     ______   ___    ___
    /\  _  \ /\_ \  /\_ \
    \ \ \L\ \\//\ \ \//\ \      __     __   _ __   ___ 
     \ \  __ \ \ \ \  \ \ \   /'__`\ /'_ `\/\`'__\/ __`\
      \ \ \/\ \ \_\ \_ \_\ \_/\  __//\ \L\ \ \ \//\ \L\ \
       \ \_\ \_\/\____\/\____\ \____\ \____ \ \_\\ \____/
        \/_/\/_/\/____/\/____/\/____/\/___L\ \/_/ \/___/
                                       /\____/
                                       \_/__/     Version 3.12


                A game programming library.

              By Shawn Hargreaves, 1994/1999.

                See the AUTHORS file for a
               complete list of contributors.



#include <std_disclaimer.h>

   "I do not accept responsibility for any effects, adverse or otherwise, 
    that this code may have on you, your computer, your sanity, your dog, 
    and anything else that you can think of. Use it at your own risk."



=======================================
============ Using Allegro ============
=======================================

See readme.txt for a general introduction, copyright details, and 
information about how to compile Allegro.

All the Allegro functions, variables, and data structures are defined in 
allegro.h. You should include this in your programs, and link with 
liballeg.a. To do this you should:

   - Put the following line at the beginning of all C or C++ files that use 
     Allegro:

         #include <allegro.h>

   - If you compile from the command line or with a makefile, add '-lalleg' 
     to the end of the gcc command, eg:

         gcc foo.c -o foo.exe -lalleg

   - If you are using Rhide, go to the Options/Libraries menu, type 'alleg' 
     into the first empty space, and make sure the box next to it is checked.

   - If you aren't using some parts of Allegro's functionality, you can 
     slightly reduce your compilation times by defining any of the following 
     preprocessor symbols before including allegro.h:

         alleg_mouse_unused
         alleg_timer_unused
         alleg_keyboard_unused
         alleg_joystick_unused
         alleg_gfx_driver_unused
         alleg_palette_unused
         alleg_graphics_unused
         alleg_vidmem_unused
         alleg_flic_unused
         alleg_sound_unused
         alleg_file_unused
         alleg_datafile_unused
         alleg_math_unused
         alleg_gui_unused

     Please note that these defines will not affect your executable size, 
     though! See towards the end of this file for information about how to 
     remove unused code at the linking stage.

int allegro_init();
   Initialises the Allegro library. It doesn't actually do very much except 
   setting up some global variables, locking some memory, and installing 
   allegro_exit() as an atexit() routine, but you must call it before doing 
   anything else. Returns zero for success (at the moment it can't fail, so 
   there is not much point checking the return value).

void allegro_exit();
   Closes down the Allegro system. This includes returning the system to 
   text mode and removing whatever mouse, keyboard, and timer routines have 
   been installed. You don't normally need to bother making an explicit call 
   to this function, because allegro_init() installs it as an atexit() 
   routine so it will be called automatically when your program exits.

extern char allegro_id[];
   Text string containing a date and version number for the library, in case 
   you want to display these somewhere.

extern char allegro_error[];
   Text string used by set_gfx_mode() and install_sound() to report error 
   messages. If they fail and you want to tell the user why, this is the 
   place to look for a description of the problem.

extern int os_type;
   Set by allegro_init() to one of the values:

      OSTYPE_UNKNOWN    - unknown, or regular MSDOS
      OSTYPE_WIN3       - Windows 3.1 or earlier
      OSTYPE_WIN95      - Windows 95
      OSTYPE_WIN98      - Windows 98
      OSTYPE_WINNT      - Windows NT
      OSTYPE_OS2        - OS/2
      OSTYPE_WARP       - OS/2 Warp 3
      OSTYPE_DOSEMU     - Linux DOSEMU
      OSTYPE_OPENDOS    - Caldera OpenDOS

extern int windows_version, windows_sub_version;
   These are set by allegro_init(), using int 0x2F, ax=0x1600 to detect the 
   presence of Microsoft Windows. Under windows 3.1, windows_version will be 
   set to 3, under win95 it will be 4, under NT it will be 0x100, and if 
   windows isn't present it will be zero. Under Linux DOSEMU, they will both 
   be set to -1.

void check_cpu();
   Detects the CPU type, setting the following global variables:

extern char cpu_vendor[];
   Set by check_cpu() to the CPU vendor name, if known.

extern int cpu_family;
   Set by check_cpu() to the CPU type: 3=386, 4=486, 5=Pentium, 6=PPro, etc.

extern int cpu_model;
   Set by check_cpu() to the CPU submodel. On a 386 (cpu_family=3), zero 
   indicates a DX chip, 2 an SX, 4 an SL, and 15 is unknown. On a 486 
   (cpu_family=4), zero or one indicates a DX chip, 2 an SX, 3 a 487 (SX) or 
   486 DX, 4 an SL, 5 an SX2, 7 a DX2 write-back enhanced, 8 a DX4 or DX4 
   overdrive, 14 a Cyrix, and 15 is unknown. On a Pentium chip 
   (cpu_family=5), 1 indicates a Pentium (510\66, 567\66), 2 is a Pentium 
   P54C, 3 is a Pentium overdrive processor, 5 is a Pentium overdrive for 
   IntelDX4, 14 is a Cyrix, and 15 is unknown.

extern int cpu_fpu;
   Set by check_cpu() to TRUE or FALSE, depending on whether a floating 
   point processor is available.

extern int cpu_mmx;
   Set by check_cpu() to TRUE or FALSE, depending on whether the MMX 
   instruction set is available.

extern int cpu_3dnow;
   Set by check_cpu() to TRUE or FALSE, depending on whether the 3DNow! 
   instruction set is available.

extern int cpu_cpuid;
   Set by check_cpu() to TRUE or FALSE, depending on whether the "cpuid" 
   instruction was available (if this is set, the other CPU variables are 
   100% reliable, otherwise there may be some mistakes).



================================================
============ Configuration routines ============
================================================

Various parts of Allegro, such as the sound routines and the 
load_joystick_data() function, require some configuration information. This 
data is stored in text files as a collection of "variable=value" lines, 
along with comments that begin with a '#' character and continue to the end 
of the line. The configuration file may optionally be divided into sections, 
which begin with a "[sectionname]" line. Each section has a unique 
namespace, to prevent variable name conflicts, but any variables that aren't 
in a section are considered to belong to all the sections simultaneously.

By default the configuration data is read from a file called allegro.cfg or 
sound.cfg, which can be located either in the same directory as the program 
executable, or the directory pointed to by the ALLEGRO environment variable. 
If you don't like this approach, you can specify any filename you like, or 
use a block of binary configuration data provided by your program (which 
could for example be loaded from a datafile).

You can store whatever custom information you like in the config file, along 
with the standard variables that are used by Allegro (see below).

void set_config_file(char *filename);
   Sets the configuration file to be used by all subsequent config 
   functions. If you don't call this function, Allegro will use the default 
   allegro.cfg file, looking first in the same directory as your program and 
   then in the directory pointed to by the ALLEGRO environment variable.

void set_config_data(char *data, int length);
   Specifies a block of data to be used by all subsequent config functions, 
   which you have already loaded from disk (eg. as part of some more 
   complicated format of your own, or in a grabber datafile). This routine 
   makes a copy of the information, so you can safely free the data after 
   calling it.

void override_config_file(char *filename);
   Specifies a file containing config overrides. These settings will be used 
   in addition to the parameters in the main config file, and where a 
   variable is present in both files this version will take priority. This 
   can be used by application programmers to override some of the config 
   settings from their code, while still leaving the main config file free 
   for the end user to customise. For example, you could specify a 
   particular sample frequency and IBK instrument file, but the user could 
   still use a sound.cfg or allegro.cfg file to specify the port settings 
   and irq numbers.

void override_config_data(char *data, int length);
   Version of override_config_file() which uses a block of data that has 
   already been read into memory.

void push_config_state();
   Pushes the current configuration state (filename, variable values, etc). 
   onto an internal stack, allowing you to select some other config source 
   and later restore the current settings by calling pop_config_state(). 
   This function is mostly intended for internal use by other library 
   functions, for example when you specify a config filename to the 
   save_joystick_data() function, it pushes the config state before 
   switching to the file you specified.

void pop_config_state();
   Pops a configuration state previously stored by push_config_state(), 
   replacing the current config source with it.

void hook_config_section(char *section,
                         int (*intgetter)(char *name, int def),
                         char *(*stringgetter)(char *name, char *def),
                         void (*stringsetter)(char *name, char *value));
   Takes control of the specified config file section, so that your hook 
   functions will be used to manipulate it instead of the normal disk file 
   access. If both the getter and setter functions are NULL, a currently 
   present hook will be unhooked. Hooked functions have the highest 
   priority. If a section is hooked, the hook will always be called, so you 
   can also hook a '#' section: even override_config_file() cannot override 
   a hooked section.

int config_is_hooked(char *section);
   Returns TRUE if the specified config section has been hooked.

char *get_config_string(char *section, char *name, char *def);
   Retrieves a string variable from the current config file. If the named 
   variable cannot be found, the value of def is returned. The section name 
   may be set to NULL to accept variables from any part of the file, or used 
   to control which set of parameters (eg. sound or joystick) you are 
   interested in reading.

int get_config_int(char *section, char *name, int def);
   Reads an integer variable from the current config file. See the comments 
   about get_config_string().

int get_config_hex(char *section, char *name, int def);
   Reads an integer variable from the current config file, in hexadecimal 
   format. See the comments about get_config_string().

float get_config_float(char *section, char *name, float def);
   Reads a floating point variable from the current config file. See the 
   comments about get_config_string().

int get_config_id(char *section, char *name, int def);
   Reads a 4-letter driver ID variable from the current config file. See the 
   comments about get_config_string().

char **get_config_argv(char *section, char *name, int *argc);
   Reads a token list (words separated by spaces) from the current config 
   file, returning a an argv style argument list, and setting argc to the 
   number of tokens (unlike argc/argv, this list is zero based). Returns 
   NULL and sets argc to zero if the variable is not present. The token list 
   is stored in a temporary buffer that will be clobbered by the next call 
   to get_config_argv(), so the data should not be expected to persist.

char *get_config_text(char *msg);
   This function is primarily intended for use by internal library code, but 
   it may perhaps be helpful to application programmers as well. It uses the 
   language.dat file to look up a translated version of the parameter in the 
   currently selected language, returning a suitable translation if one can 
   be found or a copy of the parameter if nothing else is available. This is 
   basically the same thing as calling get_config_string() with [language] 
   as the section, msg as the variable name, and msg as the default value.

void set_config_string(char *section, char *name, char *val);
   Writes a string variable to the current config file, replacing any 
   existing value it may have, or removes the variable if val is NULL. The 
   section name may be set to NULL to write the variable to the root of the 
   file, or used to control which section the variable is inserted into. The 
   altered file will be cached in memory, and not actually written to disk 
   until you call allegro_exit(). Note that you can only write to files in 
   this way, so the function will have no effect if the current config 
   source was specified with set_config_data() rather than set_config_file().

   As a special case, variable or section names that begin with a '#' 
   character are treated specially and will not be read from or written to 
   the disk. Addon packages can use this to store version info or other 
   status information into the config module, from where it can be read with 
   the get_config_string() function.

void set_config_int(char *section, char *name, int val);
   Writes an integer variable to the current config file. See the comments 
   about set_config_string().

void set_config_hex(char *section, char *name, int val);
   Writes an integer variable to the current config file, in hexadecimal 
   format. See the comments about set_config_string().

void set_config_float(char *section, char *name, float val);
   Writes a floating point variable to the current config file. See the 
   comments about set_config_string().

void set_config_id(char *section, char *name, int val);
   Writes a 4-letter driver ID variable to the current config file. See the 
   comments about set_config_string().

Allegro uses these standard variables from the configuration file:

gfx_card = x
   Specifies which graphics driver to use when the program requests 
   GFX_AUTODETECT. Multiple possible drivers can be suggested with extra 
   lines in the form "gfx_card2 = x", "gfx_card3 = x", etc. The driver ID 
   should be one of the values:

      VGA      - Standard VGA
      MODX     - Mode-X
      VBE1     - VESA 1.x
      VB2B     - VBE 2.0 (banked)
      VB2L     - VBE 2.0 (linear)
      VBE3     - VBE 3.0
      VBAF     - VBE/AF
      XTND     - Xtended mode

vbeaf_driver = x
   Specifies where to look for the VBE/AF driver (vbeaf.drv). If this 
   variable is not set, Allegro will look in the same directory as the 
   program, and then fall back on the standard locations (c:\, or the 
   directory specified with the VBEAF_PATH environment variable).

keyboard = x
   Specifies which keyboard layout to use. The parameter is the name of a 
   keyboard mapping file produced by the keyconf utility, and can either be 
   a fully qualified file path or a basename like "us" or "uk". If the 
   latter, Allegro will look first for a separate config file with that name 
   (eg. "uk.cfg") and then for an object with that name in the keyboard.dat 
   file (eg. "UK_CFG"). The config file or keyboard.dat file can be stored 
   in the same directory as the program, or in the location pointed to by 
   the ALLEGRO environment variable. Look in the keyboard.dat file to see 
   which mappings are available: at the time of this writing there are:

      BE       - Belgium
      CH       - Swiss
      CZ       - Czech
      DE       - Germany
      DK       - Denmark
      DVORAK   - Dvorak
      ES       - Spain
      FI       - Finland
      FR       - France
      IT       - Italy
      NO       - Norway
      PT       - Portugal
      RU       - Russian
      SE       - Sweden
      UK       - United Kingdom
      US       - United States

language = x
   Specifies which language file to use for error messages and other bits of 
   system text. The parameter is the name of a translation file, and can 
   either be a fully qualified file path or a basename like "en" or "sp". If 
   the latter, Allegro will look first for a separate config file with a 
   name in the form "entext.cfg", and then for an object with that name in 
   the language.dat file (eg. "ENTEXT_CFG"). The config file or language.dat 
   file can be stored in the same directory as the program, or in the 
   location pointed to by the ALLEGRO environment variable. Look in the 
   language.dat file to see which mappings are available: at the time of 
   this writing there are:

      CZ       - Czech
      DE       - German
      DK       - Danish
      EN       - English
      ES       - Spanish
      FI       - Finnish
      FR       - French
      IT       - Italian
      RU       - Russian
      SE       - Swedish

mouse = x
   Mouse type (Microsoft, Logitech, or NT). Any input method should work on 
   most systems, although the Microsoft code is probably more reliable. Some 
   Logitech drivers have a bug that causes the mouse to move at eighth 
   speed, though, and you should use the Logitech setting if you suffer from 
   this problem.

[sound]
   Section containing sound configuration information, using the variables:

digi_card = x
   Sets the driver to use for playing samples, where x is one of the values:

      0        - none
      SB       - Sound Blaster (autodetect breed)
      SB10     - Sound Blaster 1.0
      SB15     - Sound Blaster 1.5
      SB20     - Sound Blaster 2.0
      SBP      - Sound Blaster Pro
      SB16     - Sound Blaster 16
      ESS      - ESS AudioDrive
      ESC      - Ensoniq Soundscape

midi_card = x
   Sets the driver to use for MIDI music, where x is one of the values:

      0        - none
      OPL      - Adlib (autodetect OPL version)
      OPL2     - OPL2 FM synth
      OPLX     - Dual OPL2 (SB Pro-1)
      OPL3     - OPL3 FM synth
      SB       - raw SB MIDI interface
      MPU      - MPU-401 MIDI interface
      DIGI     - DIGMID software wavetable
      AWE      - AWE32

digi_input_card = x
   Sets the driver to use for digital sample input, in the same format as 
   the digi_card variable. This can usually be left blank.

midi_input_card = x
   Sets the driver to use for MIDI data input, in the same format as the 
   midi_card variable. This can usually be left blank.

digi_voices = x
   Specifies the minimum number of voices to reserve for use by the digital 
   sound driver. How many are possible depends on the driver.

midi_voices = x
   Specifies the minimum number of voices to reserve for use by the MIDI 
   sound driver. How many are possible depends on the driver.

quality = x
   Controls the sound quality vs. performance tradeoff for the sample mixing 
   code. This can be set to any of the values:
      0 - fast mixing of 8 bit data into 16 bit buffers
      1 - true 16 bit mixing (requires a 16 bit stereo soundcard)
      2 - interpolated 16 bit mixing

flip_pan = x
   Toggling this between 0 and 1 reverses the left/right panning of samples, 
   which might be needed because some SB cards (including mine :-) get the 
   stereo image the wrong way round.

sb_port = x
   Sets the port address of the SB (this is usually 220).

sb_dma = x
   Sets the DMA channel for the SB (this is usually 1).

sb_irq = x
   Sets the IRQ for the SB (this is usually 7).

sb_freq = x
   Sets the sample frequency. With the SB driver, possible rates are 11906 
   (any), 16129 (any), 22727 (SB 2.0 and above), and 45454 (only on SB 2.0 
   or SB16, not the stereo SB Pro driver). On the ESS Audiodrive, possible 
   rates are 11363, 17046, 22729, or 44194. On the Ensoniq Soundscape, 
   possible rates are 11025, 16000, 22050, or 48000. Don't worry if you set 
   some other number by mistake: Allegro will automatically round it to the 
   closest supported frequency.

fm_port = x
   Sets the port address of the OPL synth (this is usually 388).

mpu_port = x
   Sets the port address of the MPU-401 MIDI interface (this is usually 330).

mpu_irq = x
   Sets the IRQ for the MPU-401 (this is usually the same as sb_irq).

digi_volume = x
   Sets the volume for digital sample playback, from 0 to 255.

midi_volume = x
   Sets the volume for midi music playback, from 0 to 255.

ibk_file = x
   Specifies the name of a .IBK file which will be used to replace the 
   standard Adlib patch set.

ibk_drum_file = x
   Specifies the name of a .IBK file which will be used to replace the 
   standard set of Adlib percussion patches.

patches = x
   Specifies where to find the sample set for the DIGMID driver. This can 
   either be a Gravis style directory containing a collection of .pat files 
   and a default.cfg index, or an Allegro datafile produced by the pat2dat 
   utility. If this variable is not set, Allegro will look either for a 
   default.cfg or patches.dat file in the same directory as the program, the 
   directory pointed to by the ALLEGRO environment variable, and the 
   standard GUS directory pointed to by the ULTRASND environment variable.

[midimap]
   If you are using the SB MIDI output or MPU-401 drivers with an external 
   synthesiser that is not General MIDI compatible, you can use the midimap 
   section of the config file to specify a patch mapping table for 
   converting GM patch numbers into whatever bank and program change 
   messages will select the appropriate sound on your synth. This is a real 
   piece of self-indulgence. I have a Yamaha TG500, which has some great 
   sounds but no GM patch set, and I just had to make it work somehow...

   This section consists of a set of lines in the form:

p<n> = bank0 bank1 prog pitch
   With this statement, n is the GM program change number (1-128), bank0 and 
   bank1 are the two bank change messages to send to your synth (on 
   controllers #0 and #32), prog is the program change message to send to 
   your synth, and pitch is the number of semitones to shift everything that 
   is played with that sound. Setting the bank change numbers to -1 will 
   prevent them from being sent.

   For example, the line:

      p36 = 0 34 9 12

   specifies that whenever GM program 36 (which happens to be a fretless 
   bass) is selected, Allegro should send a bank change message #0 with a 
   parameter of 0, a bank change message #32 with a parameter of 34, a 
   program change with a parameter of 9, and then should shift everything up 
   by an octave.

[joystick]
   Section containing a set of variables used by the save_joystick_data() 
   and load_joystick_data() functions.



========================================
============ Mouse routines ============
========================================

The Allegro mouse handler runs on top of the DOS int33 mouse driver, so it 
will only work when the DOS driver (usually mouse.com or mouse.exe) is 
active. It is useful as a simple wrapper for the int33 functions, and also 
because it can display mouse pointers in SVGA modes, which most 
implementations of the standard driver cannot.

int install_mouse();
   Installs the Allegro mouse handler. You must do this before using any 
   other mouse functions. Returns -1 on failure (ie. if the int33 driver is 
   not loaded), otherwise the number of buttons on the mouse.

void remove_mouse();
   Removes the mouse handler. You don't normally need to bother calling 
   this, because allegro_exit() will do it for you.

extern volatile int mouse_x;
extern volatile int mouse_y;
extern volatile int mouse_b;
extern volatile int mouse_pos;
   Global variables containing the current mouse position and button state. 
   The mouse_x and mouse_y positions are integers ranging from zero to the 
   bottom right corner of the screen. The mouse_b variable is a bitfield 
   indicating the state of each button: bit 0 is the left button, bit 1 the 
   right, and bit 2 the middle button. For example:

      if (mouse_b & 1)
         printf("Left button is pressed\n");

      if (!(mouse_b & 2))
         printf("Right button is not pressed\n");

   The mouse_pos variable has the current X coordinate in the high word and 
   the Y in the low word. This may be useful in tight polling loops where a 
   mouse interrupt could occur between your reading of the two separate 
   variables, since you can copy this value into a local variable with a 
   single instruction and then split it up at your leisure.

void show_mouse(BITMAP *bmp);
   Tells Allegro to display a mouse pointer on the screen. This will only 
   work if the timer module has been installed. The mouse pointer will be 
   drawn onto the specified bitmap, which should normally be 'screen' (see 
   later for information about bitmaps). To hide the mouse pointer, call 
   show_mouse(NULL). Warning: if you draw anything onto the screen while the 
   pointer is visible, a mouse movement interrupt could occur in the middle 
   of your drawing operation. If this happens the mouse buffering and SVGA 
   bank switching code will get confused and will leave 'mouse droppings' 
   all over the screen. To prevent this, you must make sure you turn off the 
   mouse pointer whenever you draw onto the screen.

void scare_mouse();
   Helper for hiding the mouse pointer prior to a drawing operation. This 
   will temporarily get rid of the pointer, but only if that is really 
   required (ie. the mouse is visible, and is displayed on the physical 
   screen rather than some other memory surface, and it is not a hardware 
   cursor). The previous mouse state is stored for subsequent calls to 
   unscare_mouse().

void unscare_mouse();
   Undoes the effect of a previous call to scare_mouse(), restoring the 
   original pointer state.

extern int freeze_mouse_flag;
   If this flag is set, the mouse pointer won't be redrawn when the mouse 
   moves. This can avoid the need to hide the pointer every time you draw to 
   the screen, as long as you make sure your drawing doesn't overlap with 
   the current pointer position.

void position_mouse(int x, int y);
   Moves the mouse to the specified screen position. It is safe to call even 
   when a mouse pointer is being displayed.

void set_mouse_range(int x1, int y1, int x2, int y2);
   Sets the area of the screen within which the mouse can move. Pass the top 
   left corner and the bottom right corner (inclusive). If you don't call 
   this function the range defaults to (0, 0, SCREEN_W-1, SCREEN_H-1).

void set_mouse_speed(int xspeed, int yspeed);
   Sets the mouse speed. Larger values of xspeed and yspeed represent slower 
   mouse movement: the default for both is 2.

void set_mouse_sprite(BITMAP *sprite);
   You don't like my mouse pointer? No problem. Use this function to supply 
   an alternative of your own. If you change the pointer and then want to 
   get my lovely arrow back again, call set_mouse_sprite(NULL).

void set_mouse_sprite_focus(int x, int y);
   The mouse focus is the bit of the pointer that represents the actual 
   mouse position, ie. the (mouse_x, mouse_y) position. By default this is 
   the top left corner of the arrow, but if you are using a different mouse 
   pointer you might need to alter it.

void get_mouse_mickeys(int *mickeyx, int *mickeyy);
   Measures how far the mouse has moved since the last call to this 
   function. The mouse will continue to generate movement mickeys even when 
   it reaches the edge of the screen, so this form of input can be useful 
   for games that require an infinite range of mouse movement.

extern void (*mouse_callback)(int flags);
   Called by the interrupt handler whenever the mouse moves or one of the 
   buttons changes state. This function must be in locked memory, and must 
   execute _very_ quickly! It is passed the event flags that triggered the 
   call, which is a bitmask containing any of the values MOUSE_FLAG_MOVE, 
   MOUSE_FLAG_LEFT_DOWN, MOUSE_FLAG_LEFT_UP, MOUSE_FLAG_RIGHT_DOWN, 
   MOUSE_FLAG_RIGHT_UP, MOUSE_FLAG_MIDDLE_DOWN, and MOUSE_FLAG_MIDDLE_UP.



========================================
============ Timer routines ============
========================================

The standard PC clock only ticks 18.2 times a second, which is not much good 
for fast action games. Allegro can replace the system timer routine with a 
custom one, which reprograms the clock for higher tick rates while still 
calling the BIOS handler at the old speed. You can set up several virtual 
timers of your own, all going at different speeds, and Allegro will 
constantly reprogram the clock to make sure they are all called at the 
correct times. Because Allegro alters the PIT timer chip settings, it should 
not be used together with the uclock() function from the djgpp libc.

int install_timer();
   Installs the Allegro timer interrupt handler. You must do this before 
   installing any user timer routines, and also before displaying a mouse 
   pointer, playing FLI animations or MIDI music, and using any of the GUI 
   routines.

void remove_timer();
   Removes the Allegro timer handler and passes control of the clock back to 
   the BIOS. You don't normally need to bother calling this, because 
   allegro_exit() will do it for you.

int install_int(void (*proc)(), int speed);
   Installs a user timer handler, with the speed given as the number of 
   milliseconds between ticks. This is the same thing as 
   install_int_ex(proc, MSEC_TO_TIMER(speed)).

int install_int_ex(void (*proc)(), int speed);
   Adds a function to the list of user timer handlers, or if it is already 
   installed, adjusts its speed. The speed is given in hardware clock ticks, 
   of which there are 1193181 a second. You can convert from other time 
   formats to hardware clock ticks with the macros:

      SECS_TO_TIMER(secs)  - give the number of seconds between each tick
      MSEC_TO_TIMER(msec)  - give the number of milliseconds between ticks
      BPS_TO_TIMER(bps)    - give the number of ticks each second
      BPM_TO_TIMER(bpm)    - give the number of ticks per minute

   If there is no room to add a new user timer, install_int_ex() will return 
   a negative number, otherwise it returns zero. There can only be sixteen 
   timers in use at a time, and some other parts of Allegro (the GUI code, 
   the mouse pointer display routines, rest(), the FLI player, and the MIDI 
   player) need to install handlers of their own, so you should avoid using 
   too many at the same time.

   Your function will be called by the Allegro interrupt handler and not 
   directly by the processor, so it can be a normal C function and does not 
   need a special wrapper. You should be aware, however, that it will be 
   called in an interrupt context, which imposes a lot of restrictions on 
   what you can do in it. It should not use large amounts of stack, it must 
   not make any calls to DOS or use C library functions which may in turn 
   call DOS routines, and it must execute very quickly. Don't try to do lots 
   of complicated code in a timer handler: as a general rule you should just 
   set some flags and respond to these later in your main control loop.

   In a protected mode environment like djgpp, memory is virtualised and can 
   be swapped to disk. Due to the non-reentrancy of DOS, if a disk swap 
   occurs inside an interrupt handler the system will die a painful death, 
   so you need to make sure you lock all the memory (both code and data) 
   that is touched inside timer routines. Allegro will lock everything it 
   uses, but you are responsible for locking your handler functions. The 
   macros LOCK_VARIABLE(variable), END_OF_FUNCTION(function_name), and 
   LOCK_FUNCTION(function_name) can be used to simplify this task. For 
   example, if you want an interrupt handler that increments a counter 
   variable, you should write:

      volatile int counter;

      void my_timer_handler()
      {
         counter++;
      }

      END_OF_FUNCTION(my_timer_handler);

   and in your initialisation code you should lock the memory:

      LOCK_VARIABLE(counter);
      LOCK_FUNCTION(my_timer_handler);

   Obviously this can get awkward if you use complicated data structures and 
   call other functions from within your handler, so you should try to keep 
   your interrupt routines as simple as possible.

void remove_int(void (*proc)());
   Removes a function from the list of user interrupt routines. At program 
   termination, allegro_exit() does this automatically.

extern int i_love_bill;
   If set to TRUE, enables the special 'windows friendly' timer mode, which 
   locks the hardware timer interrupt to a speed of 200 ticks per second 
   rather than dynamically reprogramming it. This mode reduces the accuracy 
   of the timer (for instance, rest() will round the delay time to the 
   nearest 5 milliseconds), and prevents the vertical retrace simulator from 
   working (calls to timer_simulate_retrace() are ignored while in this 
   mode). On the plus side, though, it makes Allegro programs work under 
   windows 3.1, and stops win95 from complaining that it should be run in 
   DOS mode. This flag should be set before you install the timer module, 
   and you should not change it while the timer is active. By default, 
   allegro_init() will enable this mode if it detects the presence of 
   Windows.

void timer_simulate_retrace(int enable);
   The timer handler can be used to simulate vertical retrace interrupts. 
   An retrace interrupt can be extremely useful for implementing smooth 
   animation, but unfortunately the VGA hardware doesn't support it. The EGA 
   did, and some SVGA chipsets do, but not enough, and not in a sufficiently 
   standardised way, for it to be useful. Allegro works around this by 
   programming the timer to generate an interrupt when it thinks a retrace 
   is next likely to occur, and polling the VGA inside the interrupt handler 
   to make sure it stays in sync with the monitor refresh. This works quite 
   well in some situations, but there are a lot of caveats:

   - Don't ever use the retrace simulator in SVGA modes. It will work with 
     some chipsets, but not others, and it conflicts with most VESA 
     implementations. Retrace simulation is only reliable in VGA mode 13h 
     and mode-X.

   - Retrace simulation doesn't work under win95, because win95 returns 
     garbage when I try to read the elapsed time from the PIT. If anyone 
     knows how I can make this work, please tell me!

   - Retrace simulation involves a lot of waiting around in the timer 
     handler with interrupts disabled. This will significantly slow down 
     your entire system, and may also cause static when playing samples on 
     SB 1.0 cards (because they don't support auto-initialised DMA: SB 2.0 
     and above will be fine).

   Bearing all those problems in mind, I'd strongly advise against relying 
   on the retrace simulator. If you are coding in mode-X, and don't care 
   about your program working under win95, it is great, but it would be a 
   good idea to give the user an option to disable it.

   Retrace simulation must be enabled before you use the triple buffering 
   functions in a mode-X resolution. It can also be useful for simple 
   retrace detection, because the polling vsync() function can occasionally 
   miss retraces if a soundcard or timer interrupt occurs at exactly the 
   same time as the retrace. When retrace interrupt simulation is enabled, 
   vsync() will check the retrace_count variable rather than polling the 
   VGA, so it won't miss retraces even if they are masked by other 
   interrupts.

extern volatile int retrace_count;
   If the retrace simulator is installed, this is incremented on each 
   vertical retrace, otherwise it is incremented 70 times a second (ignoring 
   retraces). This provides a useful way of controlling the speed of your 
   program without the hassle of installing user timer functions.

   The speed of retraces varies depending on the graphics mode. In mode 13h 
   and 200/400 line mode-X resolutions there are 70 retraces a second, and 
   in 240/480 line modes there are 60. It can be as low as 50 (in 376x282 
   mode) or as high as 92 (in 400x300 mode).

extern void (*retrace_proc)();
   If the retrace simulator is installed, this function is called during 
   every vertical retrace, otherwise it is called 70 times a second 
   (ignoring retraces). Set it to NULL to disable the callback. The function 
   must obey the same rules as regular timer handlers (ie. it must be 
   locked, and mustn't call DOS or libc functions) but even more so: it must 
   execute _very_ quickly, or it will mess up the timer synchronisation. The 
   only use I can see for this function is for doing palette manipulations, 
   because triple buffering can be done with the request_scroll() function, 
   and the retrace_count variable can be used for timing your code. If you 
   want to alter the palette in the retrace_proc you should use the inline 
   _set_color() function rather than the regular set_color() or 
   set_palette(), and you shouldn't try to alter more than two or three 
   palette entries in a single retrace.

void rest(long time);
   Once Allegro has taken over the timer the standard delay() function will 
   no longer work, so you should use this routine instead. The time is given 
   in milliseconds.

void rest_callback(long time, void (*callback)())
   Like rest(), but continually calls the specified function while it is 
   waiting for the required time to elapse.



===========================================
============ Keyboard routines ============
===========================================

If you want to detect multiple keypresses at the same time, the BIOS 
keyboard routines are useless. Allegro can install a replacement keyboard 
handler, which provides both buffered input and a set of flags storing the 
state of each key. Note that it is not possible to correctly detect every 
combination of keys, due to the design of the PC keyboard. Up to two or 
three keys at a time will work fine, but if you press more than that the 
extras are likely to be ignored (exactly which combinations are possible 
seems to vary from one keyboard to another).

int install_keyboard();
   Installs the Allegro keyboard interrupt handler. You must call this 
   before using any of the keyboard input routines. Once you have set up the 
   Allegro handler, you can no longer use DOS/BIOS calls or C library 
   functions to access the keyboard.

void remove_keyboard();
   Removes the keyboard handler, returning control to the BIOS. You don't 
   normally need to bother calling this, because allegro_exit() will do it 
   for you.

void install_keyboard_hooks(int (*keypressed)(), int (*readkey)());
   You should only use this function if you *aren't* using the rest of the 
   keyboard handler. It should be called in the place of install_keyboard(), 
   and lets you provide callback routines to detect and read keypresses, 
   which will be used by the main keypressed() and readkey() functions. This 
   can be useful if you want to use Allegro's GUI code with a custom 
   keyboard handler, as it provides a way for the GUI to access keyboard 
   input from your own code. If you want to use the BIOS keyboard routines, 
   the libc function _bios_keybrd(_KEYBRD_READ) returns keypresses in the 
   correct format.

extern volatile char key[128];
   Array of flags indicating the state of each key, ordered by scancode. The 
   scancodes are defined in allegro.h as a series of KEY_* constants. For 
   example, you could write:

      if (key[KEY_SPACE])
         printf("Space is pressed\n");

   Each of these values is actually a bitfield containing either or both of 
   the flags KB_NORMAL and KB_EXTENDED, allowing you to determine which key 
   is pressed even when there are two with the same scancode. For example, 
   you could write:

      if (key[KEY_ENTER] & KB_NORMAL)
         printf("Enter (next to the right shift key) is pressed\n");

      if (key[KEY_ENTER] & KB_EXTENDED)
         printf("Enter (on the numeric keypad) is pressed\n");

extern volatile int key_shifts;
   Bitmask containing the current state of shift/ctrl/alt, the special 
   Windows keys, and the accent escape characters. This can contain any of 
   the flags:

      KB_SHIFT_FLAG
      KB_CTRL_FLAG
      KB_ALT_FLAG
      KB_LWIN_FLAG
      KB_RWIN_FLAG
      KB_MENU_FLAG
      KB_SCROLOCK_FLAG
      KB_NUMLOCK_FLAG
      KB_CAPSLOCK_FLAG
      KB_INALTSEQ_FLAG
      KB_ACCENT1_FLAG
      KB_ACCENT1_S_FLAG
      KB_ACCENT2_FLAG
      KB_ACCENT2_S_FLAG

int keypressed();
   Returns TRUE if there are keypresses waiting in the input buffer. This is 
   equivalent to the libc kbhit() function.

int readkey();
   Returns the next character from the keyboard buffer. If the buffer is 
   empty, it waits until a key is pressed. The low byte of the return value 
   contains the ASCII code of the key, and the high byte the scancode. The 
   scancode remains the same whatever the state of the shift, ctrl and alt 
   keys. The ASCII code is affected by shift and ctrl in the normal way 
   (shift changes case, ctrl+letter gives the position of that letter in the 
   alphabet, eg. ctrl+A = 1, ctrl+B = 2, etc). Pressing alt+key returns only 
   the scancode, with a zero ASCII code in the low byte. For example:

      if ((readkey() & 0xff) == 'd')         // by ASCII code
         printf("You pressed 'd'\n");

      if ((readkey() >> 8) == KEY_SPACE)     // by scancode
         printf("You pressed Space\n");

      if ((readkey() & 0xff) == 3)           // ctrl+letter
         printf("You pressed Control+C\n");

      if (readkey() == (KEY_X << 8))         // alt+letter
         printf("You pressed Alt+X\n");

void simulate_keypress(int key);
   Stuffs a key into the keyboard buffer, just as if the user had pressed 
   it. The parameter is in the same format returned by readkey().

extern int (*keyboard_callback)(int key);
   If set, this function is called by the keyboard handler in response to 
   every keypress. It is passed a copy of the value that is about to be 
   added into the input buffer, and can either return this value unchanged, 
   return zero to cause the key to be ignored, or return a modified value to 
   change what readkey() will later return. This routine executes in an 
   interrupt context, so it must be in locked memory.

extern void (*keyboard_lowlevel_callback)(int key);
   If set, this function is called by the keyboard handler in response to 
   every keyboard event, both presses and releases. It will be passed a raw 
   keyboard scancode byte, with the top bit clear if the key has been 
   pressed or set if it was released. This routine executes in an interrupt 
   context, so it must be in locked memory.

void set_leds(int leds);
   Overrides the state of the keyboard LED indicators. The parameter is a 
   bitmask containing any of the values KB_SCROLOCK_FLAG, KB_NUMLOCK_FLAG, 
   and KB_CAPSLOCK_FLAG, or -1 to restore the default behavior.

void clear_keybuf();
   Empties the keyboard buffer.

extern int three_finger_flag;
   The Allegro keyboard handler provides an 'emergency exit' sequence which 
   you can use to kill off your program. If you are running under DOS this 
   is the three finger salute, ctrl+alt+del. Most multitasking OS's will 
   trap this combination before it reaches the Allegro handler, in which 
   case you can use the alternative ctrl+alt+end. If you want to disable 
   this behaviour in release versions of your program, set this flag to 
   FALSE.

extern int key_led_flag;
   By default, the capslock, numlock, and scroll-lock keys toggle the 
   keyboard LED indicators when they are pressed. If you are using these 
   keys for input in your game (eg. capslock to fire) this may not be 
   desirable, so you can clear this flag to prevent the LED's being updated.

extern int switch_standard_kb_key;
extern int switch_standard_kb_flags;
extern int switch_custom_kb_key;
extern int switch_custom_kb_flags;
   Scancode and shift flag combinations used for the custom to standard 
   keyboard mapping switch. By default, pressing ctrl+alt+F1 will select the 
   standard US layout, while ctrl+alt+F2 selects the custom mapping. 
   These values allow you to alter the hotswap keys, or you can set them to 
   zero to disable the swapping.

void set_standard_keyboard();
   Forces Allegro to use the standard US keyboard mapping, regardless of 
   what custom keyboard layout is currently loaded.

void set_custom_keyboard();
   Forces a switch to the current custom keyboard layout, reversing the 
   effect of any previous calls to set_standard_keyboard().



===========================================
============ Joystick routines ============
===========================================

int install_joystick(int type);
   Initialises the joystick, and calibrates the centre position value. You 
   must call this before using any other joystick functions, and you should 
   make sure that the joystick is in the middle position at the time. 
   Returns zero on success. The type parameter should be one of the values:

   JOY_TYPE_AUTODETECT
      Attempts to autodetect your joystick hardware. It isn't possible to 
      reliably distinguish between all the possible input setups, so this 
      routine can only ever choose the standard joystick, Sidewider, or 
      GamePad Pro drivers, but it will use information from the 
      configuration file if one is available (this can be created using the 
      setup utility or by calling the save_joystick_data() function), so you 
      can always use JOY_TYPE_AUTODETECT in your code and then select the 
      exact hardware type from the setup program.

   JOY_TYPE_NONE
      Dummy driver for machines without any joystick.

   JOY_TYPE_STANDARD
      A normal two button stick.

   JOY_TYPE_2PADS
      Dual joystick mode (two sticks, each with two buttons).

   JOY_TYPE_4BUTTON
      Enable the extra buttons on a 4-button joystick.

   JOY_TYPE_6BUTTON
      Enable the extra buttons on a 6-button joystick.

   JOY_TYPE_8BUTTON
      Enable the extra buttons on an 8-button joystick.

   JOY_TYPE_FSPRO
      CH Flightstick Pro or compatible stick, which provides four buttons, 
      an analogue throttle control, and a 4-direction coolie hat.

   JOY_TYPE_WINGEX
      A Logitech Wingman Extreme, which should also work with any 
      Thrustmaster Mk.I compatible joystick. It provides support for four 
      buttons and a coolie hat. This also works with the Wingman Warrior, if 
      you plug in the 15 pin plug (remember to unplug the 9-pin plug!) and 
      set the tiny switch in front to the "H" position (you will not be able 
      to use the throttle or the spinner though).

   JOY_TYPE_SIDEWINDER
      The Microsoft Sidewinder digital pad (supports up to four 
      controllers, each with ten buttons and a digital direction control).

   JOY_TYPE_GAMEPAD_PRO
      The Gravis GamePad Pro (supports up to two controllers, each with ten 
      buttons and a digital direction control).

   JOY_TYPE_SNESPAD_LPT1
   JOY_TYPE_SNESPAD_LPT2
   JOY_TYPE_SNESPAD_LPT3
      SNES joypads connected to LPT1, LPT2, and LPT3 respectively.

   JOY_TYPE_WINGWARRIOR
      A Wingman Warrior joystick.

   As soon as you have installed the joystick module, you will be able to 
   read the button state and digital (on/off toggle) direction information, 
   which may be enough for some games. If you want to get full analogue 
   input, though, you need to use the calibrate_joystick() functions to 
   measure the exact range of the inputs: see below.

void remove_joystick();
   Removes the joystick handler. You don't normally need to bother calling 
   this, because allegro_exit() will do it for you.

int poll_joystick();
   Unlike the mouse and keyboard, the joystick is not interrupt driven, so 
   you need to call this function every now and again to update the global 
   position values.

extern int num_joysticks;
   Global variable containing the number of active joystick devices. The 
   current drivers support a maximum of four controllers.

extern JOYSTICK_INFO joy[n];
   Global array of joystick state information, which is updated by the 
   poll_joystick() function. Only the first num_joysticks elements will 
   contain meaningful information. The JOYSTICK_INFO structure is defined as:

   typedef struct JOYSTICK_INFO
   {
      int flags;                       - status flags for this joystick
      int num_sticks;                  - how many stick inputs?
      int num_buttons;                 - how many buttons?
      JOYSTICK_STICK_INFO stick[n];    - stick state information
      JOYSTICK_BUTTON_INFO button[n];  - button state information
   } JOYSTICK_INFO;

   The button status is stored in the structure:

   typedef struct JOYSTICK_BUTTON_INFO
   {
      int b;                           - boolean on/off flag
      char *name;                      - description of this button
   } JOYSTICK_BUTTON_INFO;

   You may wish to display the button names as part of an input 
   configuration screen to let the user choose what game function will be 
   performed by each button, but in simpler situations you can safely assume 
   that the first two elements in the button array will always be the main 
   trigger controls.

   Each joystick will provide one or more stick inputs, of varying types. 
   These can be digital controls which snap to specific positions (eg. a 
   gamepad controller, the coolie hat on a Flightstick Pro or Wingman 
   Extreme, or a normal joystick which hasn't yet been calibrated), or they 
   can be full analogue inputs with a smooth range of motion. Sticks may 
   also have different numbers of axis, for example a normal directional 
   control has two, but the Flightstick Pro throttle is only a single axis, 
   and it is possible that the system could be extended in the future to 
   support full 3d controllers. A stick input is described by the structure:

   typedef struct JOYSTICK_STICK_INFO
   {
      int flags;                       - status flags for this input
      int num_axis;                    - how many axis do we have?
      JOYSTICK_AXIS_INFO axis[n];      - axis state information
      char *name;                      - description of this input
   } JOYSTICK_STICK_INFO;

   A single joystick may provide several different stick inputs, but you can 
   safely assume that the first element in the stick array will always be 
   the main directional controller.

   Information about each of the stick axis is stored in the substructure:

   typedef struct JOYSTICK_AXIS_INFO
   {
      int pos;                         - analogue axis position
      int d1, d2;                      - digital axis position
      char *name;                      - description of this axis
   } JOYSTICK_AXIS_INFO;

   This provides both analogue input in the pos field (ranging from -128 to 
   128 or from 0 to 255, depending on the type of the control), and digital 
   values in the d1 and d2 fields. For example, when describing the X-axis 
   position, the pos field will hold the horizontal position of the 
   joystick, d1 will be set if it is moved left, and d2 will be set if it is 
   moved right. Allegro will fill in all these values regardless of whether 
   it is using a digital or analogue joystick, emulating the pos field for 
   digital inputs by snapping it to the min, middle, and maximum positions, 
   and emulating the d1 and d2 values for an analogue stick by comparing the 
   current position with the centre point.

   The joystick flags field may contain any combination of the bit flags:

   JOYFLAG_DIGITAL
      This control is currently providing digital input.

   JOYFLAG_ANALOGUE
      This control is currently providing analogue input.

   JOYFLAG_CALIB_DIGITAL
      This control will be capable of providing digital input once it has 
      been calibrated, but is not doing this at the moment.

   JOYFLAG_CALIB_ANALOGUE
      This control will be capable of providing analogue input once it has 
      been calibrated, but is not doing this at the moment.

   JOYFLAG_CALIBRATE
      Indicates that this control needs to be calibrated. Many devices 
      require multiple calibration steps, so you should call the 
      calibrate_joystick() function from a loop until this flag is cleared.

   JOYFLAG_SIGNED
      Indicates that the analogue axis position is in signed format, ranging 
      from -128 to 128. This is the case for all 2d directional controls.

   JOYFLAG_UNSIGNED
      Indicates that the analogue axis position is in unsigned format, 
      ranging from 0 to 255. This is the case for all 1d throttle controls.

   Note for people who spell funny: in case you don't like having to type 
   "analogue", there are some #define aliases in allegro.h that will allow 
   to to write "analog" instead.

char *calibrate_joystick_name(int n);
   Returns a text description for the next type of calibration that will be 
   done on the specified joystick, or NULL if no more calibration is 
   required.

int calibrate_joystick(int n);
   Most joysticks need to be calibrated before they can provide full 
   analogue input. This function performs the next operation in the 
   calibration series for the specified stick, assuming that the joystick 
   has been positioned in the manner described by a previous call to 
   calibrate_joystick_name(), returning zero on success. For example, a 
   simple routine to fully calibrate all the joysticks might look like:

      int i;

      for (i=0; i<;num_joysticks; i++) {
         while (joy[i].flags & JOYFLAG_CALIBRATE) {
            char *msg = calibrate_joystick_name(i);
            printf("%s, and press a key\n", msg);
            readkey();
            if (calibrate_joystick(i) != 0) {
               printf("oops!\n");
               exit(1);
            }
         }
      }

int save_joystick_data(char *filename);
   After all the headache of calibrating the joystick, you may not want to 
   make your poor users repeat the process every time they run your program. 
   Call this function to save the joystick calibration data into the 
   specified configuration file, from which it can later be read by 
   load_joystick_data(). Pass a NULL filename to write the data to the 
   currently selected configuration file. Returns zero on success.

int load_joystick_data(char *filename);
   Restores calibration data previously stored by save_joystick_data() or 
   the setup utility. This sets up all aspects of the joystick code: you 
   don't even need to call install_joystick() if you are using this 
   function. Pass a NULL filename to read the data from the currently 
   selected configuration file. Returns zero on success: if it fails the 
   joystick state is undefined and you must reinitialise it from scratch.



========================================
============ Graphics modes ============
========================================

void set_color_depth(int depth);
   Sets the pixel format to be used by subsequent calls to set_gfx_mode() 
   and create_bitmap(). Valid depths are 8 (the default), 15, 16, 24, and 32 
   bits.

int set_gfx_mode(int card, int w, int h, int v_w, int v_h);
   Switches into graphics mode. The card parameter should be one of the 
   values:

      GFX_TEXT          - return to text mode
      GFX_AUTODETECT    - let Allegro pick an appropriate graphics driver
      GFX_VGA           - normal VGA (320x200, 320x100, 160x120, or 80x80)
      GFX_MODEX         - select a planar, tweaked VGA mode
      GFX_VESA1         - use the VESA 1.x driver
      GFX_VESA2B        - use the VBE 2.0 banked mode driver
      GFX_VESA2L        - use the VBE 2.0 linear framebuffer driver
      GFX_VESA3         - use the VBE 3.0 driver
      GFX_VBEAF         - use the VBE/AF hardware accelerator API
      GFX_XTENDED       - use the unchained mode 640x400 driver

   The w and h parameters specify what screen resolution you want. Possible 
   modes are:

      The standard 256 color VGA mode 13h, using the GFX_VGA driver. This is 
      normally sized 320x200, which will work on any VGA but doesn't support 
      large virtual screens and hardware scrolling. Allegro also provides 
      some tweaked variants of the mode which are able to scroll, sized 
      320x100 (with a 200 pixel high virtual screen), 160x120 (with a 409 
      pixel high virtual screen), and 80x80 (with a 819 pixel high virtual 
      screen).

      Mode-X, using the GFX_MODEX driver. This will work on any VGA card, 
      and provides a range of different 256 color tweaked resolutions.

      Stable mode-X resolutions:

         Square aspect ratio: 320x240

         Skewed aspect ratio: 256x224, 256x240, 320x200, 320x400,
                              320x480, 320x600, 360x200, 360x240,
                              360x360, 360x400, 360x480

         These have worked on every card/monitor that I've tested.

      Unstable mode-X resolutions:

         Square aspect ratio: 360x270, 376x282, 400x300

         Skewed aspect ratio: 256x200, 256x256, 320x350, 360x600,
                              376x308, 376x564, 400x150, 400x600

         These only work on some monitors. They were fine on my old machine, 
         but don't get on very well with my new monitor. If you are worried 
         about the possibility of damaging your monitor by using these 
         modes, don't be. Of course I'm not providing any warranty with any 
         of this, and if your hardware does blow up that is tough, but I 
         don't think this sort of tweaking can do any damage. From the 
         documentation of Robert Schmidt's TWEAK program:

            "Some time ago, putting illegal or unsupported values or 
            combinations of such into the video card registers might prove 
            hazardous to both your monitor and your health. I have *never* 
            claimed that bad things can't happen if you use TWEAK, although 
            I'm pretty sure it never will. I've never heard of any damage 
            arising from trying out TWEAK, or from general VGA tweaking in 
            any case."

         Most of the mode-X drawing functions are slower than in mode 13h, 
         due to the complexity of the planar bitmap organisation, but solid 
         area fills and plane-aligned blits from one part of video memory to 
         another can be significantly faster, particularly on older 
         hardware. Mode-X can address the full 256k of VGA RAM, so hardware 
         scrolling and page flipping are possible, and it is possible to 
         split the screen in order to scroll the top part of the display but 
         have a static status indicator at the bottom.

      The standard SVGA modes, 640x480, 800x600, and 1024x768. These ought 
      to work with any SVGA card: if they don't, get a copy of the SciTech 
      Display Doctor and see if that fixes it. What color depths are 
      available will depend on your hardware. Most cards support both 15 and 
      16 bit resolutions, but if at all possible I would advise you to 
      support both (it's not hard...) in case one is not available. Some 
      cards provide both 24 and 32 bit truecolor, in which case it is a 
      choice between 24 (saves memory) or 32 (faster), but many older cards 
      have no 32 bit mode and some newer ones don't support 24 bit 
      resolutions. Use the vesainfo test program to see what modes your VESA 
      driver provides.

      Many cards also support 640x400, 1280x1024, and 1600x1200, but these 
      aren't available on everything, for example the S3 chipset has no 
      640x400 mode. Other weird resolution may be possible, eg. some Tseng 
      boards can do 640x350, and the Avance Logic has a 512x512 mode.

      Xtended mode, as described by Mark Feldman in the PCGPE, sized 640x400 
      (256 colors) and using the GFX_XTENDED driver. This uses VESA to 
      select an SVGA mode (so it will only work on cards supporting the VESA 
      640x400 resolution), and then unchains the VGA hardware as for mode-X. 
      This allows the entire screen to be addressed without the need for 
      bank switching, but hardware scrolling and page flipping are not 
      possible. This driver will never be autodetected (the normal VESA 
      640x400 mode will be chosen instead), so if you want to use it you 
      will have to explicitly pass GFX_XTENDED to set_gfx_mode().

      The SciTech Display Doctor provides several scrollable low resolution 
      modes in a range of different color depths (320x200, 320x240, 320x400, 
      320x480, 360x200, 360x240, 360x400, and 360x480 all work on my ET4000 
      with 8, 15, or 16 bits per pixel). These are lovely, allowing 
      scrolling and page flipping without the complexity of the mode-X 
      planar setup, but unfortunately they aren't standard so you will need 
      Display Doctor in order to use them.

      VBE/AF is a superset of the VBE 2.0 standard, which provides an API 
      for accessing hardware accelerator features. VBE/AF drivers are 
      currently only available from the FreeBE/AF project or as part of the 
      SciTech Display Doctor package, but they can give dramatic speed 
      improvements when used with suitable hardware. For a detailed 
      discussion of hardware acceleration issues, refer to the documentation 
      for the gfx_capabilities flag.

      You can use the afinfo test program to check whether you have a VBE/AF 
      driver, and to see what resolutions it supports.

      The SciTech VBE/AF drivers require nearptr access to be enabled, so 
      any stray pointers are likely to crash your machine while their 
      drivers are in use. This means it may be a good idea to use VESA while 
      debugging your program, and only switch to VBE/AF once the code is 
      working correctly. The FreeBE/AF drivers do not have this problem.

   The v_w and v_h parameters specify the minimum virtual screen size, in 
   case you need a large virtual screen for hardware scrolling or page 
   flipping. You should set them to zero if you don't care about the virtual 
   screen size. Virtual screens can cause a lot of confusion, but they are 
   really quite simple. Warning: patronising explanation coming up, so you 
   may wish to skip the rest of this paragraph :-) Think of video memory as 
   a rectangular piece of paper which is being viewed through a small hole 
   (your monitor) in a bit of cardboard. Since the paper is bigger than the 
   hole you can only see part of it at any one time, but by sliding the 
   cardboard around you can alter which portion of the image is visible. You 
   could just leave the hole in one position and ignore the parts of video 
   memory that aren't visible, but you can get all sorts of useful effects 
   by sliding the screen window around, or by drawing images in a hidden 
   part of video memory and then flipping across to display them.

   For example, you could select a 640x480 mode in which the monitor acts as 
   a window onto a 1024x1024 virtual screen, and then move the visible 
   screen around in this larger area. Initially, with the visible screen 
   positioned at the top left corner of video memory, this setup would look 
   like:

      (0,0)------------(640,0)----(1024,0)
        |                  |           |
        |  visible screen  |           |
        |                  |           |
      (0,480)----------(640,480)       |
        |                              |
        |   the rest of video memory   |
        |                              |
      (0,1024)--------------------(1024,1024)

   What's that? You are viewing this with a proportional font? Hehehe.

   When you call set_gfx_mode(), the v_w and v_h parameters represent the 
   minimum size of virtual screen that is acceptable for your program. The 
   range of possible sizes is usually very restricted, and Allegro is likely 
   to end up creating a virtual screen much larger than the one you request. 
   On an SVGA card with one megabyte of vram you can count on getting a 
   1024x1024 virtual screen (256 colors) or 1024x512 (15 or 16 bpp), and 
   with 512k vram you can get 1024x512 (256 color). Other sizes may or may 
   not be possible: don't assume that they will work. In mode-X the virtual 
   width can be any multiple of eight greater than or equal to the physical 
   screen width, and the virtual height will be set accordingly (the VGA has 
   256k of vram, so the virtual height will be 256*1024/virtual_width).

   After you select a graphics mode, the physical and virtual screen sizes 
   can be checked with the macros SCREEN_W, SCREEN_H, VIRTUAL_W, and 
   VIRTUAL_H.

   If Allegro is unable to select an appropriate mode, set_gfx_mode() 
   returns a negative number and stores a description of the problem in 
   allegro_error. Otherwise it returns zero.

extern int gfx_capabilities;
   Bitfield describing the capabilities of the current graphics driver and 
   video hardware. This may contain combination any of the flags:

   GFX_CAN_SCROLL:
      Indicates that the scroll_screen() function may be used with this 
      driver.

   GFX_CAN_TRIPLE_BUFFER:
      Indicates that the request_scroll() and poll_scroll() functions may be 
      used with this driver. As a special case, the mode-X driver only 
      supports triple buffering when the retrace simulator is installed, so 
      you must call timer_simulate_retrace() before doing any triple 
      buffering in a mode-X resolution.

   GFX_HW_CURSOR:
      Indicates that a hardware mouse cursor is in use. When this flag is 
      set, it is safe to draw onto the screen without hiding the mouse 
      pointer first. Note that not every cursor graphic can be implemented 
      in hardware: in particular VBE/AF only supports 2-color images up to 
      32x32 in size, where the second color is an exact inverse of the 
      first. This means that Allegro may need to switch between hardware and 
      software cursors at any point during the execution of your program, so 
      you should not assume that this flag will remain constant for long 
      periods of time. It only tells you whether a hardware cursor is in use 
      at the current time, and may change whenever you hide/redisplay the 
      pointer.

   GFX_HW_HLINE:
      Indicates that the normal opaque version of the hline() function is 
      implemented using a hardware accelerator. This will improve the 
      performance not only of hline() itself, but also of many other 
      functions that use it as a workhorse, for example circlefill(), 
      triangle(), and floodfill().

   GFX_HW_HLINE_XOR:
      Indicates that the XOR version of the hline() function, and any other 
      functions that use it as a workhorse, are implemented using a hardware 
      accelerator.

   GFX_HW_HLINE_SOLID_PATTERN:
      Indicates that the solid and masked pattern modes of the hline() 
      function, and any other functions that use it as a workhorse, are 
      implemented using a hardware accelerator (see note below).

   GFX_HW_HLINE_COPY_PATTERN:
      Indicates that the copy pattern mode of the hline() function, and any 
      other functions that use it as a workhorse, are implemented using a 
      hardware accelerator (see note below).

   GFX_HW_FILL:
      Indicates that the opaque version of the rectfill() function, the 
      clear() routine, and clear_to_color(), are implemented using a 
      hardware accelerator.

   GFX_HW_FILL_XOR:
      Indicates that the XOR version of the rectfill() function is 
      implemented using a hardware accelerator.

   GFX_HW_FILL_SOLID_PATTERN:
      Indicates that the solid and masked pattern modes of the rectfill() 
      function are implemented using a hardware accelerator (see note below).

   GFX_HW_FILL_COPY_PATTERN:
      Indicates that the copy pattern mode of the rectfill() function is 
      implemented using a hardware accelerator (see note below).

   GFX_HW_LINE:
      Indicates that the opaque mode line() and vline() functions are 
      implemented using a hardware accelerator.

   GFX_HW_LINE_XOR:
      Indicates that the XOR version of the line() and vline() functions are 
      implemented using a hardware accelerator.

   GFX_HW_TRIANGLE:
      Indicates that the opaque mode triangle() function is implemented 
      using a hardware accelerator.

   GFX_HW_TRIANGLE_XOR:
      Indicates that the XOR version of the triangle() function is 
      implemented using a hardware accelerator.

   GFX_HW_TEXTOUT_FIXED:
      Indicates that monochrome character expansion (for the 8x8 and 8x16 
      fixed width font formats) is implemented using a hardware accelerator.

   GFX_HW_VRAM_BLIT:
      Indicates that blitting from one part of the screen to another is 
      implemented using a hardware accelerator. If this flag is set, 
      blitting within the video memory will almost certainly be the fastest 
      possible way to display an image, so it may be worth storing some of 
      your more frequently used graphics in an offscreen portion of the 
      video memory.

   GFX_HW_VRAM_BLIT_MASKED:
      Indicates that the masked_blit() routine is capable of a hardware 
      accelerated copy from one part of video memory to another, and that 
      draw_sprite() will use a hardware copy when given a sub-bitmap of the 
      screen or a video memory bitmap as the source image. If this flag is 
      set, copying within the video memory will almost certainly be the 
      fastest possible way to display an image, so it may be worth storing 
      some of your more frequently used sprites in an offscreen portion of 
      the video memory.

      Warning: if this flag is not set, masked_blit() and draw_sprite() will 
      not work correctly when used with a video memory source image! You 
      must only try to use these functions to copy within the video memory 
      if they are supported in hardware.

   GFX_HW_MEM_BLIT:
      Indicates that blitting from a memory bitmap onto the screen is being 
      accelerated in hardware.

   GFX_HW_MEM_BLIT_MASKED:
      Indicates that the masked_blit() and draw_sprite() functions are being 
      accelerated in hardware when the source image is a memory bitmap and 
      the destination is the physical screen.

   Note: even if the capabilities information says that patterned drawing is 
   supported by the hardware, it will not be possible for every size of 
   pattern. VBE/AF only supports patterns up to 8x8 in size, so Allegro will 
   fall back on the original non-accelerated drawing routines whenever you 
   use a pattern larger than this.

   Note2: these hardware acceleration features will only take effect when 
   you are drawing directly onto the screen bitmap, a video memory bitmap, 
   or a sub-bitmap thereof. Accelerated hardware is most useful in a page 
   flipping or triple buffering setup, and is unlikely to make any 
   difference to the classic "draw onto a memory bitmap, then blit to the 
   screen" system.

int scroll_screen(int x, int y);
   Attempts to scroll the hardware screen to display a different part of the 
   virtual screen (initially it will be positioned at 0, 0, which is the top 
   left corner). Returns zero on success: it may fail if the graphics driver 
   can't handle hardware scrolling or the virtual screen isn't large enough. 
   You can use this to move the screen display around in a large virtual 
   screen space, or to page flip back and forth between two non-overlapping 
   areas of the virtual screen. Note that to draw outside the original 
   position in the screen bitmap you will have to alter the clipping 
   rectangle: see below.

   Mode-X scrolling is reliable and will work on any card. Unfortunately 
   most VESA implementations can only handle horizontal scrolling in four 
   pixel increments, so smooth horizontal panning is impossible in SVGA 
   modes. This is a shame, but I can't see any way round it. A significant 
   number of VESA implementations seem to be very buggy when it comes to 
   scrolling in truecolor video modes, so I suggest you don't depend on this 
   routine working correctly in the truecolor resolutions unless you can be 
   sure that SciTech Display Doctor is installed.

   Allegro will handle any necessary vertical retrace synchronisation when 
   scrolling the screen, so you don't need to call vsync() before it. This 
   means that scroll_screen() has the same time delay effects as vsync().

int request_scroll(int x, int y);
   This function is used for triple buffering. It requests a hardware scroll 
   to the specified position, but returns immediately rather than waiting 
   for a retrace. The scroll will then take place during the next vertical 
   retrace, but you can carry on running other code in the meantime and use 
   the poll_scroll() routine to detect when the flip has actually taken 
   place (see examples/ex20.c). Triple buffering is only possible on certain 
   hardware: it will work in any mode-X resolution if the timer retrace 
   simulator is active (but this doesn't work correctly under win95), plus 
   it is supported by the VBE 3.0 and VBE/AF drivers for a limited number of 
   high-end graphics cards. You can look at the GFX_CAN_TRIPLE_BUFFER bit in 
   the gfx_capabilities flag to see if it will work with the current driver.

int poll_scroll();
   This function is used for triple buffering. It checks the status of a 
   hardware scroll previously initiated by the request_scroll() routine, 
   returning non-zero if it is still waiting to take place, and zero if it 
   has already happened.

int show_video_bitmap(BITMAP *bitmap);
   Attempts to page flip the hardware screen to display the specified video 
   bitmap object, which must be the same size as the physical screen, and 
   should have been obtained by calling the create_video_bitmap() function. 
   Returns zero on success. This function will wait for a vertical retrace 
   if the graphics card requires it, so you don't need to call vsync() 
   yourself.

int request_video_bitmap(BITMAP *bitmap);
   This function is used for triple buffering. It requests a page flip to 
   display the specified video bitmap object, but returns immediately rather 
   than waiting for a retrace. The flip will then take place during the next 
   vertical retrace, but you can carry on running other code in the meantime 
   and use the poll_scroll() routine to detect when the flip has actually 
   taken place. Triple buffering is only possible on certain hardware: see 
   the comments about request_scroll().

int request_modex_scroll(int x, int y);
int poll_modex_scroll();
   Obsolete versions of request_scroll() and poll_scroll(), preserved for 
   backward compatibility.

void split_modex_screen(int line);
   This function is only available in mode-X. It splits the VGA display into 
   two parts at the specified line. The top half of the screen can be 
   scrolled to any part of video memory with the scroll_screen() function, 
   but the part below the specified line number will remain fixed and will 
   display from position (0, 0) of the screen bitmap. After splitting the 
   screen you will generally want to scroll so that the top part of the 
   display starts lower down in video memory, and then create two 
   sub-bitmaps to access the two sections (see examples/ex19.c for a 
   demonstration of how this could be done). To disable the split, call 
   split_modex_screen(0).



========================================
============ Bitmap objects ============
========================================

Once you have selected a graphics mode, you can draw things onto the display 
via the 'screen' bitmap. All the Allegro graphics routines draw onto BITMAP 
structures, which are areas of memory containing rectangular images, stored 
as packed byte arrays (eight bits per pixel). You can create and manipulate 
bitmaps in system RAM, or you can write to the special 'screen' bitmap which 
represents the video memory in your graphics card.

For example, to draw a pixel onto the screen you would write:

   putpixel(screen, x, y, color);

Or to implement a double-buffered system:

   BITMAP *bmp = create_bitmap(320, 200);    // make a bitmap in system RAM
   clear(bmp);                               // zero the memory bitmap
   putpixel(bmp, x, y, color);               // draw onto the memory bitmap
   blit(bmp, screen, 0, 0, 0, 0, 320, 200);  // copy it to the screen

See below for information on how to get direct access to the image memory in 
a bitmap.

Allegro supports several different types of bitmaps:

   - The screen bitmap, which represents the hardware video memory. 
     Ultimately you have to draw onto this in order for your image to be 
     visible.

   - Memory bitmaps, which are located in system RAM and can be used to 
     store graphics or as temporary drawing spaces for double buffered 
     systems. These can be obtained by calling create_bitmap(), load_pcx(), 
     or by loading a grabber datafile.

   - Sub-bitmaps. These share image memory with a parent bitmap (which 
     can be the screen, a memory bitmap, or another sub-bitmap), so drawing 
     onto them will also change their parent. They can be of any size and 
     located anywhere within the parent bitmap, and can have their own 
     clipping rectangles, so they are a useful way of dividing a bitmap into 
     several smaller units, eg. splitting a large virtual screen into two 
     sections for page flipping (see examples/ex9.c).

   - Video memory bitmaps. These are created by the create_video_bitmap() 
     function, and are usually implemented as sub-bitmaps of the screen 
     object.

extern BITMAP *screen;
   Global pointer to a bitmap, sized VIRTUAL_W x VIRTUAL_H. This is created 
   by set_gfx_mode(), and represents the hardware video memory. Only a part 
   of this bitmap will actually be visible, sized SCREEN_W x SCREEN_H. 
   Normally this is the top left corner of the larger virtual screen, so you 
   can ignore the extra invisible virtual size of the bitmap if you aren't 
   interested in hardware scrolling or page flipping. To move the visible 
   window to other parts of the screen bitmap, call scroll_screen(). 
   Initially the clipping rectangle will be limited to the physical screen 
   size, so if you want to draw onto a larger virtual screen space outside 
   this rectangle, you will need to adjust the clipping.

BITMAP *create_bitmap(int width, int height);
   Creates a memory bitmap sized width by height, and returns a pointer to 
   it. The bitmap will have clipping turned on, and the clipping rectangle 
   set to the full size of the bitmap. The image memory will not be cleared, 
   so it will probably contain garbage: you should clear the bitmap before 
   using it. This routine always uses the global pixel format, as specified 
   by calling set_color_depth().

BITMAP *create_bitmap_ex(int color_depth, int width, int height);
   Creates a bitmap in a specific color depth (8, 15, 16, 24 or 32 bits per 
   pixel).

BITMAP *create_sub_bitmap(BITMAP *parent, int x, y, width, height);
   Creates a sub-bitmap, ie. a bitmap sharing drawing memory with a 
   pre-existing bitmap, but possibly with a different size and clipping 
   settings. When creating a sub-bitmap of the mode-X screen, the x position 
   must be a multiple of four. The sub-bitmap width and height can extend 
   beyond the right and bottom edges of the parent (they will be clipped), 
   but the origin point must lie within the parent region.

BITMAP *create_video_bitmap(int width, int height);
   Allocates a video memory bitmap of the specified size, returning a 
   pointer to it on success or NULL on failure (ie. if you have run out of 
   vram). This can be used to allocate offscreen video memory for storing 
   source graphics ready for a hardware accelerated blitting operation, or 
   to create multiple video memory pages which can then be displayed by 
   calling show_video_bitmap(). Video memory bitmaps are usually allocated 
   from the same space as the screen bitmap, so they may overlap with it: it 
   is not therefore a good idea to use the global screen at the same time as 
   any surfaces returned by this function.

void destroy_bitmap(BITMAP *bitmap);
   Destroys a memory bitmap, sub-bitmap, or video memory bitmap when you are 
   finished with it.

int bitmap_color_depth(BITMAP *bmp);
   Returns the color depth of the specified bitmap (8, 15, 16, 24, or 32).

int bitmap_mask_color(BITMAP *bmp);
   Returns the mask color for the specified bitmap (the value which is 
   skipped when drawing sprites). For 256 color bitmaps this is zero, and 
   for truecolor bitmaps it is bright pink (maximum red and blue, zero 
   green).

int is_same_bitmap(BITMAP *bmp1, BITMAP *bmp2);
   Returns TRUE if the two bitmaps describe the same drawing surface, ie. 
   the pointers are equal, one is a sub-bitmap of the other, or they are 
   both sub-bitmaps of a common parent.

int is_linear_bitmap(BITMAP *bmp);
   Returns TRUE if bmp is a linear bitmap, ie. a memory bitmap, mode 13h 
   screen, or SVGA screen. Linear bitmaps can be used with the _putpixel(), 
   _getpixel(), bmp_write_line(), and bmp_read_line() functions.

int is_planar_bitmap(BITMAP *bmp);
   Returns TRUE if bmp is a planar (mode-X or Xtended mode) screen bitmap.

int is_memory_bitmap(BITMAP *bmp);
   Returns TRUE if bmp is a memory bitmap, ie. it was created by calling 
   create_bitmap() or loaded from a grabber datafile or image file. Memory 
   bitmaps can be accessed directly via the line pointers in the bitmap 
   structure, eg. bmp->line[y][x] = color.

int is_screen_bitmap(BITMAP *bmp);
   Returns TRUE if bmp is the screen bitmap, a video memory bitmap, or a 
   sub-bitmap of either.

int is_sub_bitmap(BITMAP *bmp);
   Returns TRUE if bmp is a sub-bitmap.

void set_clip(BITMAP *bitmap, int x1, int y1, int x2, int y2);
   Each bitmap has an associated clipping rectangle, which is the area of 
   the image that it is ok to draw on. Nothing will be drawn to positions 
   outside this space. Pass the two opposite corners of the clipping 
   rectangle: these are inclusive, eg. set_clip(bitmap, 16, 16, 32, 32) will 
   allow drawing to (16, 16) and (32, 32), but not to (15, 15) and (33, 33). 
   If x1, y1, x2, and y2 are all zero, clipping will be turned off, which 
   may slightly speed up some drawing operations (usually a negligible 
   difference, although every little helps) but will result in your program 
   dying a horrible death if you try to draw beyond the edges of the bitmap.



=============================================
============ Loading image files ============
=============================================

Warning: when using truecolor images, you should always set the graphics 
mode before loading any bitmap data! Otherwise the pixel format (RGB or BGR)
will not be known, so the file may be converted wrongly.

BITMAP *load_bitmap(char *filename, RGB *pal);
   Loads a bitmap from a file, returning a pointer to a bitmap and storing 
   the palette data in the specified location, which should be an array of 
   256 RGB structures. You are responsible for destroying the bitmap when 
   you are finished with it. Returns NULL on error. At present this function 
   supports BMP, LBM, PCX, and TGA files, determining the type from the file 
   extension. If the file contains a truecolor image, you must set the video 
   mode or call set_color_conversion() before loading it.

BITMAP *load_bmp(char *filename, RGB *pal);
   Loads a 256 color or 24 bit truecolor Windows or OS/2 BMP file.

BITMAP *load_lbm(char *filename, RGB *pal);
   Loads a 256 color IFF ILBM/PBM file.

BITMAP *load_pcx(char *filename, RGB *pal);
   Loads a 256 color or 24 bit truecolor PCX file.

BITMAP *load_tga(char *filename, RGB *pal);
   Loads a 256 color, 15 bit hicolor, or 24 bit truecolor TGA file.

int save_bitmap(char *filename, BITMAP *bmp, RGB *pal);
   Writes a bitmap into a file, using the specified palette, which should be 
   an array of 256 RGB structures. Returns non-zero on error. The output 
   format is determined from the filename extension: at present this 
   function supports BMP, PCX and TGA formats. One thing to watch out for: 
   if you use this to dump the screen into a file you may end up with an 
   image much larger than you were expecting, because Allegro often creates 
   virtual screens larger than the visible screen. You can get around this 
   by using a sub-bitmap to specify which part of the screen to save, eg:

      BITMAP *bmp;
      PALETTE pal;

      get_palette(pal);
      bmp = create_sub_bitmap(screen, 0, 0, SCREEN_W, SCREEN_H);
      save_bitmap("dump.pcx", bmp, pal);
      destroy_bitmap(bmp);

int save_bmp(char *filename, BITMAP *bmp, RGB *pal);
   Writes a bitmap into a 256 color or 24 bit truecolor BMP file.

int save_pcx(char *filename, BITMAP *bmp, RGB *pal);
   Writes a bitmap into a 256 color or 24 bit truecolor PCX file.

int save_tga(char *filename, BITMAP *bmp, RGB *pal);
   Writes a bitmap into a 256 color, 15 bit hicolor, or 24 bit truecolor TGA 
   file.

void register_bitmap_file_type(char *ext,
                    BITMAP *(*load)(char *filename, RGB *pal),
                    int (*save)(char *filename, BITMAP *bmp, RGB *pal));
   Informs the load_bitmap() and save_bitmap() functions of a new file type, 
   providing routines to read and write images in this format (either 
   function may be NULL).

void set_color_conversion(int mode);
   Specifies how to convert images between the various color depths when 
   reading graphics from external bitmap files or datafiles. The mode is a 
   bitmask specifying which types of conversion are allowed. If the 
   appropriate bit is set, data will be converted into the current pixel 
   format (selected by calling the set_color_depth() function), otherwise it 
   will be left in the same format as the disk file, leaving you to convert 
   it manually before the graphic can be displayed. The default mode is 
   total conversion, so that all images will be loaded in the appropriate 
   format for the current video mode. Valid bit flags are:

      COLORCONV_EXPAND_256          // expand 256 colors into truecolor
      COLORCONV_REDUCE_TO_256       // reduce truecolor to 256 colors
      COLORCONV_EXPAND_15_TO_16     // expand 15 bit hicolor to 16 bits
      COLORCONV_REDUCE_16_TO_15     // reduce 16 bit hicolor to 15 bits
      COLORCONV_EXPAND_HI_TO_TRUE   // expand 15/16 bits to 24/32 bits
      COLORCONV_REDUCE_TRUE_TO_HI   // reduce 24/32 bits to 15/16 bits
      COLORCONV_24_EQUALS_32        // convert between 24 and 32 bits
      COLORCONV_DITHER              // dither when reducing to hicolor

   For convenience, the following macros can be used to select common 
   combinations of these flags:

      COLORCONV_NONE       // disable all format conversions
      COLORCONV_TOTAL      // convert everything to current format
      COLORCONV_PARTIAL    // convert 15 <-> 16 and 24 <-> 32 bits
      COLORCONV_MOST       // all but truecolor <-> 256 conversions

   If you enable the COLORCONV_DITHER flag, dithering will be performed 
   whenever truecolor graphics are converted into a hicolor format, 
   including by the blit() function. This can produce much better looking 
   results, but is obviously much slower than a direct conversion.



==========================================
============ Palette routines ============
==========================================

All the Allegro drawing functions use integer parameters to represent 
colors. In truecolor resolutions these numbers encode the color directly as 
a collection of red, green, and blue bits, but in a regular 256 color mode 
the values are treated as indexes into the current palette, which is a table 
listing the red, green and blue intensities for each of the 256 possible 
colors.

Palette entries are stored in an RGB structure, which contains red, green 
and blue intensities in the VGA hardware format, ranging from 0-63, and is 
defined as:

typedef struct RGB
{
   unsigned char r, g, b;
} RGB;

For example:

   RGB black = { 0,  0,  0  };
   RGB white = { 63, 63, 63 };
   RGB green = { 0,  63, 0  };
   RGB grey  = { 32, 32, 32 };

The type PALETTE is defined to be an array of 256 RGB structures.

You may notice that a lot of the code in Allegro spells 'palette' as 
'pallete'. This is because the headers from my old Mark Williams compiler on 
the Atari spelt it with two l's, so that is what I'm used to. Allegro will 
happily accept either spelling, due to some #defines in allegro.h.

void vsync();
   Waits for a vertical retrace to begin. The retrace happens when the 
   electron beam in your monitor has reached the bottom of the screen and is 
   moving back to the top ready for another scan. During this short period 
   the graphics card isn't sending any data to the monitor, so you can do 
   things to it that aren't possible at other times, such as altering the 
   palette without causing flickering (snow). Allegro will automatically 
   wait for a retrace before altering the palette or doing any hardware 
   scrolling, though, so you don't normally need to bother with this 
   function.

void set_color(int index, RGB *p);
   Sets the specified palette entry to the specified RGB triplet. Unlike the 
   other palette functions this doesn't do any retrace synchronisation, so 
   you should call vsync() before it to prevent snow problems.

void _set_color(int index, RGB *p);
   This is an inline version of set_color(), intended for use in the 
   vertical retrace simulator callback function. It should only be used in 
   VGA mode 13h and mode-X, because some of the more recent SVGA chipsets 
   aren't VGA compatible (set_color() and set_palette() will use VESA calls 
   on these cards, but _set_color() doesn't know about that).

void set_palette(PALETTE p);
   Sets the entire palette of 256 colors. You should provide an array of 256 
   RGB structures. Unlike set_color(), there is no need to call vsync() 
   before this function.

void set_palette_range(PALETTE p, int from, int to, int vsync);
   Sets the palette entries between from and to (inclusive: pass 0 and 255 
   to set the entire palette). If vsync is set it waits for the vertical 
   retrace, otherwise it sets the colors immediately.

void get_color(int index, RGB *p);
   Retrieves the specified palette entry.

void get_palette(PALETTE p);
   Retrieves the entire palette of 256 colors. You should provide an array 
   of 256 RGB structures to store it in.

void get_palette_range(PALETTE p, int from, int to);
   Retrieves the palette entries between from and to (inclusive: pass 0 and 
   255 to get the entire palette).

void fade_interpolate(PALETTE source, dest, output, int pos, int from, to);
   Calculates a temporary palette part way between source and dest, 
   returning it in the output parameter. The position between the two 
   extremes is specified by the pos value: 0 returns an exact copy of 
   source, 64 returns dest, 32 returns a palette half way between the two, 
   etc. This routine only affects colors between from and to (inclusive: 
   pass 0 and 255 to interpolate the entire palette).

void fade_from_range(PALETTE source, dest, int speed, int from, to);
   Gradually fades a part of the palette from the source palette to the dest 
   palette. The speed is from 1 (the slowest) up to 64 (instantaneous). This 
   routine only affects colors between from and to (inclusive: pass 0 and 
   255 to fade the entire palette).

void fade_in_range(PALETTE p, int speed, int from, to);
   Gradually fades a part of the palette from a black screen to the 
   specified palette. The speed is from 1 (the slowest) up to 64 
   (instantaneous). This routine only affects colors between from and to 
   (inclusive: pass 0 and 255 to fade the entire palette).

void fade_out_range(int speed, int from, to);
   Gradually fades a part of the palette from the current palette to a black 
   screen. The speed is from 1 (the slowest) up to 64 (instantaneous). This 
   routine only affects colors between from and to (inclusive: pass 0 and 
   255 to fade the entire palette).

void fade_from(PALETTE source, PALETTE dest, int speed);
   Fades gradually from the source palette to the dest palette. The speed is 
   from 1 (the slowest) up to 64 (instantaneous).

void fade_in(PALETTE p, int speed);
   Fades gradually from a black screen to the specified palette. The speed 
   is from 1 (the slowest) up to 64 (instantaneous).

void fade_out(int speed);
   Fades gradually from the current palette to a black screen. The speed is 
   from 1 (the slowest) up to 64 (instantaneous).

void select_palette(PALLETE p);
   Ugly hack for use in various dodgy situations where you need to convert 
   between paletted and truecolor image formats. Sets the internal palette 
   table in the same way as the set_palette() function, so the conversion 
   will use the specified palette, but without affecting the display 
   hardware in any way. The previous palette settings are stored in an 
   internal buffer, and can be restored by calling unselect_palette().

void unselect_palette();
   Restores the palette tables that were in use before the last call to 
   select_palette().

void generate_332_palette(PALETTE pal);
   Constructs a fake truecolor palette, using three bits for red and green 
   and two for the blue. The load_bitmap() function returns this if the file 
   does not contain a palette itself (ie. you are reading a truecolor 
   bitmap).

int generate_optimized_palette(BITMAP *bmp, PALETTE pal, char rsvd[256]);
   Generates a 256 color palette suitable for making a reduced color version 
   of the specified truecolor image. The rsvd parameter points to a table 
   indicating which colors it is allowed to modify: zero for free colors, 
   non-zero for ones which are already set to fixed values and may not be 
   changed. If rsvd is NULL, the remapping will use the entire palette.

extern PALETTE black_palette;
   A palette containing solid black colors, used by the fade routines.

extern PALETTE desktop_palette;
   The palette used by the Atari ST low resolution desktop. I'm not quite 
   sure why this is still here, except that the grabber and test programs 
   use it. It is probably the only Atari legacy code left in Allegro, and it 
   would be a shame to remove it :-)



=================================================
============ Truecolor pixel formats ============
=================================================

In a truecolor video mode the red, green, and blue components for each pixel 
are packed directly into the color value, rather than using a palette lookup 
table. In a 15 bit mode there are 5 bits for each color, in 16 bit modes 
there are 5 bits each of red and blue and six bits of green, and both 24 and 
32 bit modes use 8 bits for each color (the 32 bit pixels simply have an 
extra padding byte to align the data nicely). The layout of these components 
can vary depending on your hardware, but will generally either be RGB or 
BGR. Since the layout is not known until you select the video mode you will 
be using, you must call set_gfx_mode() before using any of the following 
routines!

int makecol8(int r, int g, int b);
int makecol15(int r, int g, int b);
int makecol16(int r, int g, int b);
int makecol24(int r, int g, int b);
int makecol32(int r, int g, int b);
   These functions convert colors from a hardware independent form (red, 
   green, and blue values ranging 0-255) into various display dependent 
   pixel formats. Converting to 15, 16, 24, or 32 bit formats only takes a 
   few shifts, so it is fairly efficient. Converting to an 8 bit color 
   involves searching the palette to find the closest match, which is quite 
   slow unless you have set up an RGB mapping table (see below).

int makecol(int r, int g, int b);
   Converts colors from a hardware independent format (red, green, and blue 
   values ranging 0-255) to the pixel format required by the current video 
   mode, calling the preceding 8, 15, 16, 24, or 32 bit makecol functions as 
   appropriate.

int makecol_depth(int color_depth, int r, int g, int b);
   Converts colors from a hardware independent format (red, green, and blue 
   values ranging 0-255) to the pixel format required by the specified color 
   depth.

int makecol15_dither(int r, int g, int b, int x, int y);
int makecol16_dither(int r, int g, int b, int x, int y);
   Given both a color value and a pixel coordinate, calculate a dithered 15 
   or 16 bit RGB value. This can produce better results when reducing images 
   from truecolor to hicolor. In addition to calling these functions 
   directly, hicolor dithering can be automatically enabled when loading 
   graphics by calling the set_color_conversion() function, for example 
   set_color_conversion(COLORCONV_REDUCE_TRUE_TO_HI | COLORCONV_DITHER).

int getr8(int c);
int getg8(int c);
int getb8(int c);
int getr15(int c);
int getg15(int c);
int getb15(int c);
int getr16(int c);
int getg16(int c);
int getb16(int c);
int getr24(int c);
int getg24(int c);
int getb24(int c);
int getr32(int c);
int getg32(int c);
int getb32(int c);
   Given a color in a display dependent format, these functions extract one 
   of the red, green, or blue components (ranging 0-255).

int getr(int c);
int getg(int c);
int getb(int c);
   Given a color in the format being used by the current video mode, these 
   functions extract one of the red, green, or blue components (ranging 
   0-255), calling the preceding 8, 15, 16, 24, or 32 bit get functions as 
   appropriate.

int getr_depth(int color_depth, int c);
int getg_depth(int color_depth, int c);
int getb_depth(int color_depth, int c);
   Given a color in the format being used by the specified color depth, 
   these functions extract one of the red, green, or blue components 
   (ranging 0-255).

extern int palette_color[256];
   Table mapping palette index colors (0-255) into whatever pixel format is 
   being used by the current display mode. In a 256 color mode this just 
   maps onto the array index. In truecolor modes it looks up the specified 
   entry in the current palette, and converts that RGB value into the 
   appropriate packed pixel format.

#define MASK_COLOR_8 0
#define MASK_COLOR_15 (5.5.5 pink)
#define MASK_COLOR_16 (5.6.5 pink)
#define MASK_COLOR_24 (8.8.8 pink)
#define MASK_COLOR_32 (8.8.8 pink)
   Constants representing the colors used to mask transparent sprite pixels 
   for each color depth. In 256 color resolutions this is zero, and in 
   truecolor modes it is bright pink (maximum red and blue, zero green).



============================================
============ Drawing primitives ============
============================================

Except for _putpixel(), all these routines are affected by the current 
drawing mode and the clipping rectangle of the destination bitmap.

void putpixel(BITMAP *bmp, int x, int y, int color);
   Writes a pixel to the specified position in the bitmap, using the current 
   drawing mode and the bitmap's clipping rectangle.

void _putpixel(BITMAP *bmp, int x, int y, unsigned char color);
   Like the regular putpixel(), but much faster because it is implemented as 
   an inline assembler function. This won't work in mode-X or truecolor 
   video modes, doesn't perform any clipping (it will crash if you try to 
   draw outside the bitmap!), and ignores the drawing mode. It clobbers the 
   %fs register, so you shouldn't mix it with code that uses the 
   _farsetsel() and _farns*() functions.

int getpixel(BITMAP *bmp, int x, int y);
   Reads a pixel from point x, y in the bitmap. Returns -1 if the point lies 
   outside the bitmap.

unsigned char _getpixel(BITMAP *bmp, int x, int y);
   Faster inline version of getpixel(). This won't work in mode-X or 
   truecolor video modes, and it doesn't do any clipping, so you must make 
   sure the point lies inside the bitmap.

void vline(BITMAP *bmp, int x, int y1, int y2, int color);
   Draws a vertical line onto the bitmap, from point (x, y1) to (x, y2).

void hline(BITMAP *bmp, int x1, int y, int x2, int color);
   Draws a horizontal line onto the bitmap, from point (x1, y) to (x2, y).

void do_line(BITMAP *bmp, int x1, y1, x2, y2, int d, void (*proc)());
   Calculates all the points along a line from point (x1, y1) to (x2, y2), 
   calling the supplied function for each one. This will be passed a copy of 
   the bmp parameter, the x and y position, and a copy of the d parameter, 
   so it is suitable for use with putpixel().

void line(BITMAP *bmp, int x1, int y1, int x2, int y2, int color);
   Draws a line onto the bitmap, from point (x1, y1) to (x2, y2).

void triangle(BITMAP *bmp, int x1, y1, x2, y2, x3, y3, int color);
   Draws a filled triangle between the three points.

void polygon(BITMAP *bmp, int vertices, int *points, int color);
   Draws a filled polygon with an arbitrary number of corners. Pass the 
   number of vertices and an array containing a series of x, y points (a 
   total of vertices*2 values).

void rect(BITMAP *bmp, int x1, int y1, int x2, int y2, int color);
   Draws an outline rectangle with the two points as its opposite corners.

void rectfill(BITMAP *bmp, int x1, int y1, int x2, int y2, int color);
   Draws a solid, filled rectangle with the two points as its opposite 
   corners.

void do_circle(BITMAP *bmp, int x, int y, int radius, int d, void (*proc)());
   Calculates all the points in a circle around point (x, y) with radius r, 
   calling the supplied function for each one. This will be passed a copy of 
   the bmp parameter, the x and y position, and a copy of the d parameter, 
   so it is suitable for use with putpixel().

void circle(BITMAP *bmp, int x, int y, int radius, int color);
   Draws a circle with the specified centre and radius.

void circlefill(BITMAP *bmp, int x, int y, int radius, int color);
   Draws a filled circle with the specified centre and radius.

void do_ellipse(BITMAP *bmp, int x, y, int rx, ry, int d, void (*proc)());
   Calculates all the points in an ellipse around point (x, y) with radius 
   rx and ry, calling the supplied function for each one. This will be 
   passed a copy of the bmp parameter, the x and y position, and a copy of 
   the d parameter, so it is suitable for use with putpixel().

void ellipse(BITMAP *bmp, int x, int y, int rx, int ry, int color);
   Draws an ellipse with the specified centre and radius.

void ellipsefill(BITMAP *bmp, int x, int y, int rx, int ry, int color);
   Draws a filled ellipse with the specified centre and radius.

void do_arc(BITMAP *bmp, int x, y, fixed a1, a2, int r, d, void (*proc)());
   Calculates all the points in a circular arc around point (x, y) with 
   radius r, calling the supplied function for each one. This will be passed 
   a copy of the bmp parameter, the x and y position, and a copy of the d 
   parameter, so it is suitable for use with putpixel(). The arc will be 
   plotted in an anticlockwise direction starting from the angle a1 and 
   ending when it reaches a2. These values are specified in 16.16 fixed 
   point format, with 256 equal to a full circle, 64 a right angle, etc. 
   Zero is to the right of the centre point, and larger values rotate 
   anticlockwise from there.

void arc(BITMAP *bmp, int x, y, fixed ang1, ang2, int r, int color);
   Draws a circular arc with centre x, y and radius r, in an anticlockwise 
   direction starting from the angle a1 and ending when it reaches a2. These 
   values are specified in 16.16 fixed point format, with 256 equal to a 
   full circle, 64 a right angle, etc. Zero is to the right of the centre 
   point, and larger values rotate anticlockwise from there.

void calc_spline(int points[8], int npts, int *x, int *y);
   Calculates a series of npts values along a bezier spline, storing them in 
   the output x and y arrays. The bezier curve is specified by the four x/y 
   control points in the points array: points[0] and points[1] contain the 
   coordinates of the first control point, points[2] and points[3] are the 
   second point, etc. Control points 0 and 3 are the ends of the spline, and 
   points 1 and 2 are guides. The curve probably won't pass through points 1 
   and 2, but they affect the shape of the curve between points 0 and 3 (the 
   lines p0-p1 and p2-p3 are tangents to the spline). The easiest way to 
   think of it is that the curve starts at p0, heading in the direction of 
   p1, but curves round so that it arrives at p3 from the direction of p2. 
   In addition to their role as graphics primitives, spline curves can be 
   useful for constructing smooth paths around a series of control points, 
   as in ex23.c.

void spline(BITMAP *bmp, int points[8], int color);
   Draws a bezier spline using the four control points specified in the 
   points array.

void floodfill(BITMAP *bmp, int x, int y, int color);
   Floodfills an enclosed area, starting at point (x, y), with the specified 
   color.



==============================================
============ Blitting and sprites ============
==============================================

All these routines are affected by the clipping rectangle of the destination 
bitmap.

void clear(bitmap);
   Clears the bitmap to color 0.

void clear_to_color(BITMAP *bitmap, int color);
   Clears the bitmap to the specified color.

void blit(BITMAP *source, BITMAP *dest, int source_x, int source_y,
          int dest_x, int dest_y, int width, int height);
   Copies a rectangular area of the source bitmap to the destination bitmap. 
   The source_x and source_y parameters are the top left corner of the area 
   to copy from the source bitmap, and dest_x and dest_y are the 
   corresponding position in the destination bitmap. This routine respects 
   the destination clipping rectangle, and it will also clip if you try to 
   blit from areas outside the source bitmap.

   You can blit between any parts of any two bitmaps, even if the two memory 
   areas overlap (ie. source and dest are the same, or one is sub-bitmap of 
   the other). You should be aware, however, that a lot of SVGA cards don't 
   provide separate read and write banks, which means that blitting from one 
   part of the screen to another requires the use of a temporary bitmap in 
   memory, and is therefore extremely slow. As a general rule you should 
   avoid blitting from the screen onto itself in SVGA modes.

   In mode-X, on the other hand, blitting from one part of the screen to 
   another can be significantly faster than blitting from memory onto the 
   screen, as long as the source and destination are correctly aligned with 
   each other. Copying between overlapping screen rectangles is slow, but if 
   the areas don't overlap, and if they have the same plane alignment (ie. 
   (source_x%4) == (dest_x%4)), the VGA latch registers can be used for a 
   very fast data transfer. To take advantage of this, in mode-X it is often 
   worth storing tile graphics in a hidden area of video memory (using a 
   large virtual screen), and blitting them from there onto the visible part 
   of the screen.

   If the GFX_HW_VRAM_BLIT bit in the gfx_capabilities flag is set, the 
   current driver supports hardware accelerated blits from one part of the 
   screen onto another. This is extremely fast, so when this flag is set it 
   may be worth storing some of your more frequently used graphics in an 
   offscreen portion of the video memory.

   Unlike most of the graphics routines, blit() allows the source and 
   destination bitmaps to be of different color depths, so it can be used to 
   convert images from one pixel format to another.

void masked_blit(BITMAP *source, BITMAP *dest, int source_x, int source_y,
                  int dest_x, int dest_y, int width, int height);
   Like blit(), but skips transparent pixels, which are marked by a zero in 
   256 color modes or bright pink for truecolor data (maximum red and blue, 
   zero green). The source and destination regions must not overlap.

   If the GFX_HW_VRAM_BLIT_MASKED bit in the gfx_capabilities flag is set, 
   the current driver supports hardware accelerated masked blits from one 
   part of the screen onto another. This is extremely fast, so when this 
   flag is set it may be worth storing some of your more frequently used 
   sprites in an offscreen portion of the video memory.

   Warning: if the hardware acceleration flag is not set, masked_blit() will 
   not work correctly when used with a video memory source image, and the 
   input graphic must always be a memory bitmap!

void stretch_blit(BITMAP *source, BITMAP *dest,
                  int source_x, source_y, source_width, source_height,
                  int dest_x, dest_y, dest_width, dest_height);
   Like blit(), except it can scale images so the source and destination 
   rectangles don't need to be the same size. This routine doesn't do as 
   much safety checking as the regular blit: in particular you must take 
   care not to copy from areas outside the source bitmap, and you cannot 
   blit between overlapping regions, ie. you must use different bitmaps for 
   the source and the destination. Also, the source must be a memory bitmap 
   or sub-bitmap, not the hardware screen.

void draw_sprite(BITMAP *bmp, BITMAP *sprite, int x, int y);
   Draws a copy of the sprite bitmap onto the destination bitmap at the 
   specified position. This is almost the same as blit(sprite, bmp, 0, 0, x, 
   y, sprite->w, sprite->h), but it uses a masked drawing mode where 
   transparent pixels are skipped, so the background image will show through 
   the masked parts of the sprite. Transparent pixels are marked by a zero 
   in 256 color modes or bright pink for truecolor data (maximum red and 
   blue, zero green).

   If the GFX_HW_VRAM_BLIT_MASKED bit in the gfx_capabilities flag is set, 
   the current driver supports hardware accelerated sprite drawing when the 
   source image is a video memory bitmap or a sub-bitmap of the screen. This 
   is extremely fast, so when this flag is set it may be worth storing some 
   of your more frequently used sprites in an offscreen portion of the video 
   memory.

   Warning: if the hardware acceleration flag is not set, draw_sprite() will 
   not work correctly when used with a video memory source image, and the 
   input graphic must always be a memory bitmap!

   Although generally not supporting graphics of mixed color depths, as a 
   special case this function can be used to draw 256 color source images 
   onto truecolor destination bitmaps, so you can use palette effects on 
   specific sprites within a truecolor program.

void draw_sprite_v_flip(BITMAP *bmp, BITMAP *sprite, int x, int y);
void draw_sprite_h_flip(BITMAP *bmp, BITMAP *sprite, int x, int y);
void draw_sprite_vh_flip(BITMAP *bmp, BITMAP *sprite, int x, int y);
   These are like draw_sprite(), but they flip the image about the vertical, 
   horizontal, or diagonal, axis. This produces exact mirror images, which 
   is not the same as rotating the sprite (and it is a lot faster than the 
   rotation routine). The sprite must be a memory bitmap.

void draw_trans_sprite(BITMAP *bmp, BITMAP *sprite, int x, int y);
   Uses the global color_map table or truecolor blender functions to overlay 
   the sprite on top of the existing image. This must only be used after you 
   have set up the color mapping table (for 256 color modes) or blender map 
   (for truecolor modes). Because it involves reading as well as writing the 
   bitmap memory, translucent drawing is very slow if you draw directly to 
   video RAM, so wherever possible you should use a memory bitmap instead.

void draw_lit_sprite(BITMAP *bmp, BITMAP *sprite, int x, int y, int color);
   Tints the sprite image to the specified color or light level, using the 
   global color_map table, and draws the resulting image to the destination 
   bitmap. This must only be used after you have set up the color mapping 
   table (for 256 color modes) or blender map (for truecolor modes).

void draw_gouraud_sprite(BITMAP *bmp, BITMAP *sprite, int x, int y,
                         int c1, int c2, int c3, int c4);
   Tints the sprite to the specified color or light level, interpolating the 
   four corner colors across the surface of the image. This must only be 
   used after you have set up the color mapping table (for 256 color modes) 
   or blender map (for truecolor modes).

void draw_character(BITMAP *bmp, BITMAP *sprite, int x, int y, int color);
   Draws a copy of the sprite bitmap onto the destination bitmap at the 
   specified position, drawing transparent pixels in the current text mode 
   (skipping them if the text mode is -1, otherwise drawing them in the text 
   background color), and setting all other pixels to the specified color. 
   Transparent pixels are marked by a zero in 256 color modes or bright pink 
   for truecolor data (maximum red and blue, zero green). The sprite must be 
   an 8 bit image, even if the destination is a truecolor bitmap.

void rotate_sprite(BITMAP *bmp, BITMAP *sprite, int x, int y, fixed angle);
   Draws the sprite image onto the bitmap at the specified position, 
   rotating it by the specified angle. The angle is a fixed point 16.16 
   number in the same format used by the fixed point trig routines, with 256 
   equal to a full circle, 64 a right angle, etc. The sprite must be a 
   memory bitmap.

void rotate_scaled_sprite(BITMAP *bmp, BITMAP *sprite, int x, int y,
                          fixed angle, fixed scale);
   Like rotate_sprite(), but stretches or shrinks the image at the same time 
   as rotating it.

void stretch_sprite(BITMAP *bmp, BITMAP *sprite, int x, int y, int w, int h);
   Draws the sprite image onto the bitmap at the specified position, 
   stretching it to the specified width and height. The difference between 
   stretch_sprite() and stretch_blit() is that stretch_sprite() masks out 
   transparent pixels, which are marked by a zero in 256 color modes or 
   bright pink for truecolor data (maximum red and blue, zero green).



=====================================
============ RLE sprites ============
=====================================

Because bitmaps can be used in so many different ways, the bitmap structure 
is quite complicated, and it contains a lot of data. In many situations, 
though, you will find yourself storing images that are only ever copied to 
the screen, rather than being drawn onto or used as filling patterns, etc. 
If this is the case you may be better off storing your images in RLE_SPRITE 
or COMPILED_SPRITE structures rather than regular bitmaps.

RLE sprites store the image in a simple run-length encoded format, where 
repeated zero pixels are replaced by a single length count, and strings of 
non-zero pixels are preceded by a counter giving the length of the solid 
run. RLE sprites are usually much smaller then normal bitmaps, both because 
of the run length compression, and because they avoid most of the overhead 
of the bitmap structure. They are often also faster than normal bitmaps, 
because rather than having to compare every single pixel with zero to 
determine whether it should be drawn, it is possible to skip over a whole 
run of zeros with a single add, or to copy a long run of non-zero pixels 
with fast string instructions.

Every silver lining has a cloud, though, and in the case of RLE sprites it 
is a lack of flexibility. You can't draw onto them, and you can't flip them, 
rotate them, or stretch them. In fact the only thing you can do with them is 
to blast them onto a bitmap with the draw_rle_sprite() function, which is 
equivalent to using draw_sprite() with a regular bitmap. You can convert 
bitmaps into RLE sprites at runtime, or you can create RLE sprite structures 
in grabber datafiles by making a bitmap object and checking the 'RLE' button.

RLE_SPRITE *get_rle_sprite(BITMAP *bitmap);
   Creates an RLE sprite based on the specified bitmap (which must be a 
   memory bitmap).

void destroy_rle_sprite(RLE_SPRITE *sprite);
   Destroys an RLE sprite structure previously returned by get_rle_sprite().

void draw_rle_sprite(BITMAP *bmp, RLE_SPRITE *sprite, int x, int y);
   Draws an RLE sprite onto a bitmap at the specified position.

void draw_trans_rle_sprite(BITMAP *bmp, RLE_SPRITE *sprite, int x, int y);
   Translucent version of draw_rle_sprite(). See the description of 
   draw_trans_sprite(). This must only be used after you have set up the 
   color mapping table (for 256 color modes) or blender map (for truecolor 
   modes).

void draw_lit_rle_sprite(BITMAP *bmp, RLE_SPRITE *sprite, int x, y, color);
   Tinted version of draw_rle_sprite(). See the description of
   draw_lit_sprite(). This must only be used after you have set up the color 
   mapping table (for 256 color modes) or blender map (for truecolor modes).



==========================================
============ Compiled sprites ============
==========================================

Compiled sprites are stored as actual machine code instructions that draw a 
specific image onto a bitmap, using mov instructions with immediate data 
values. This is the fastest way to draw a masked image: on my machine 
drawing compiled sprites is about five times as fast as using draw_sprite() 
with a regular bitmap. Compiled sprites are big, so if memory is tight you 
should use RLE sprites instead, and what you can do with them is even more 
restricted than with RLE sprites, because they don't support clipping. If 
you try to draw one off the edge of a bitmap, you will corrupt memory and 
probably crash the system. You can convert bitmaps into compiled sprites at 
runtime, or you can create compiled sprite structures in grabber datafiles 
by making a bitmap object and checking the 'Linear Compiled' or 'Mode-X 
Compiled' buttons. 

COMPILED_SPRITE *get_compiled_sprite(BITMAP *bitmap, int planar);
   Creates a compiled sprite based on the specified bitmap (which must be a 
   memory bitmap). Compiled sprites are device-dependent, so you have to 
   specify whether to compile it into a linear or planar format. Pass FALSE 
   as the second parameter if you are going to be drawing it onto memory 
   bitmaps or mode 13h and SVGA screen bitmaps, and pass TRUE if you are 
   going to draw it onto mode-X or Xtended mode screen bitmaps.

void destroy_compiled_sprite(COMPILED_SPRITE *sprite);
   Destroys a compiled sprite structure previously returned by 
   get_compiled_sprite().

void draw_compiled_sprite(BITMAP *bmp, COMPILED_SPRITE *sprite, int x, int y);
   Draws a compiled sprite onto a bitmap at the specified position. The 
   sprite must have been compiled for the correct type of bitmap (linear or 
   planar). This function does not support clipping.

   Hint: if not being able to clip compiled sprites is a problem, a neat 
   trick is to set up a work surface (memory bitmap, mode-X virtual screen, 
   or whatever) a bit bigger than you really need, and use the middle of it 
   as your screen. That way you can draw slightly off the edge without any 
   trouble... 



=====================================
============ Text output ============
=====================================

Allegro provides text output routines that work with both proportional 
(individual characters can be any size) and fixed size 8x8 or 8x16 fonts. A 
font contains the ASCII characters 32 to 255: all other characters will be 
drawn as spaces. The grabber program can create fonts from sets of 
characters drawn on a PCX file (see grabber.txt for more information), and 
can also import GRX or BIOS format font files.

extern FONT *font;
   A simple 8x8 fixed size font (the mode 13h BIOS default). If you want to 
   alter the font used by the GUI routines, change this to point to one of 
   your own fonts.

void text_mode(int mode);
   Sets the mode in which text will be drawn. If mode is zero or positive, 
   text output will be opaque and the background of the characters will be 
   set to color #mode. If mode is negative, text will be drawn transparently 
   (ie. the background of the characters will not be altered). The default 
   is a mode of zero.

void textout(BITMAP *bmp, FONT *f, unsigned char *s, int x, y, int color);
   Writes the string s onto the bitmap at position x, y, using the current 
   text mode and the specified font and foreground color. If the color is -1 
   and a proportional font is in use, it will be drawn using the colors from 
   the original font bitmap (the one you imported into the grabber program), 
   which allows multicolored text output.

void textout_centre(BITMAP *bmp, FONT *f, unsigned char *s, int x, y, color);
   Like textout(), but interprets the x coordinate as the centre rather than 
   the left edge of the string.

void textout_justify(BITMAP *bmp, FONT *f, unsigned char *s, 
                     int x1, int x2, int y, int diff, int color);
   Draws justified text within the region x1-x2. If the amount of spare 
   space is greater than the diff value, it will give up and draw regular 
   left justified text instead.

void textprintf(BITMAP *bmp, FONT *f, int x, y, color, char *fmt, ...);
   Formatted text output, using a printf() style format string.

void textprintf_centre(BITMAP *bmp, FONT *f, int x, y, color, char *fmt, ...);
   Like textprintf(), but interprets the x coordinate as the centre rather 
   than the left edge of the string.

int text_length(FONT *f, unsigned char *str);
   Returns the length (in pixels) of a string in the specified font.

int text_height(FONT *f)
   Returns the height (in pixels) of the specified font.

void destroy_font(FONT *f);
   Frees the memory being used by a font structure.



===========================================
============ Polygon rendering ============
===========================================

void polygon3d(BITMAP *bmp, int type, BITMAP *texture, int vc, V3D *vtx[]);
void polygon3d_f(BITMAP *bmp, int type, BITMAP *texture, int vc, V3D_f *vtx[]);
   Draw 3d polygons onto the specified bitmap, using the specified rendering 
   mode. Unlike the regular polygon() function, these routines don't support 
   concave or self-intersecting shapes, and they can't draw onto mode-X 
   screen bitmaps (if you want to write 3d code in mode-X, draw onto a 
   memory bitmap and then blit to the screen). The width and height of the 
   texture bitmap must be powers of two, but can be different, eg. a 64x16 
   texture is fine, but a 17x3 one is not. The vertex count parameter (vc) 
   should be followed by an array containing the appropriate number of 
   pointers to vertex structures: polygon3d() uses the fixed point V3D 
   structure, while polygon3d_f() uses the floating point V3D_f structure. 
   These are defined as:

   typedef struct V3D
   {
      fixed x, y, z;       - position
      fixed u, v;          - texture map coordinates
      int c;               - color
   } V3D;

   typedef struct V3D_f
   {
      float x, y, z;       - position
      float u, v;          - texture map coordinates
      int c;               - color
   } V3D_f;

   How the vertex data is used depends on the rendering mode:

   The x and y values specify the position of the vertex in 2d screen 
   coordinates.

   The z value is only required when doing perspective correct texture 
   mapping, and specifies the depth of the point in 3d world coordinates.

   The u and v coordinates are only required when doing texture mapping, and 
   specify the position of the point in the texture bitmap, for example 0, 0 
   maps the vertex onto the top left corner of the texture, and if the 
   texture is sized 32x32, setting u=32 and v=16 maps the vertex to the 
   point half way down the right edge of the texture. The u/v coordinates 
   wrap at the edge of the texture, so with a 32x32 texture, u=v=32 is the 
   same as u=v=0. This can be used to tile textures several times across a 
   polygon.

   The c value specifies the vertex color, and is interpreted differently by 
   various rendering modes.

   The type parameter specifies the polygon rendering mode, and can be any 
   of the values:

   POLYTYPE_FLAT:
      A simple flat shaded polygon, taking the color from the c value of the 
      first vertex. This polygon type is affected by the drawing_mode() 
      function, so it can be used to render XOR or translucent polygons.

   POLYTYPE_GCOL:
      A single-color gouraud shaded polygon. The colors for each vertex are 
      taken from the c value, and interpolated across the polygon. This is 
      very fast, but will only work in 256 color modes if your palette 
      contains a smooth gradient between the colors. In truecolor modes it 
      interprets the color as a packed, display-format value as produced by 
      the makecol() function.

   POLYTYPE_GRGB:
      A gouraud shaded polygon which interpolates RGB triplets rather than a 
      single color. In 256 color modes this uses the global rgb_map table to 
      convert the result to an 8 bit paletted color, so it must only be used 
      after you have set up the RGB mapping table! The colors for each 
      vertex are taken from the c value, which is interpreted as a 24 bit 
      RGB triplet (0xFF0000 is red, 0x00FF00 is green, and 0x0000FF is 
      blue).

   POLYTYPE_ATEX:
      An affine texture mapped polygon. This stretches the texture across 
      the polygon with a simple 2d linear interpolation, which is fast but 
      not mathematically correct. It can look ok if the polygon is fairly 
      small or flat-on to the camera, but because it doesn't deal with 
      perspective foreshortening, it can produce strange warping artifacts. 
      To see what I mean, run test.exe and see what happens to the 
      polygon3d() test when you zoom in very close to the cube.

   POLYTYPE_PTEX:
      A perspective-correct texture mapped polygon. This uses the z value 
      from the vertex structure as well as the u/v coordinates, so textures 
      are displayed correctly regardless of the angle they are viewed from. 
      Because it involves division calculations in the inner texture mapping 
      loop, this mode is a lot slower than POLYTYPE_ATEX, and it uses 
      floating point so it will be very slow on anything less than a Pentium 
      (even with an FPU, a 486 can't overlap floating point division with 
      other integer operations like the Pentium can).

   POLYTYPE_ATEX_MASK:
   POLYTYPE_PTEX_MASK:
      Like POLYTYPE_ATEX and POLYTYPE_PTEX, but zero texture map pixels are 
      skipped, allowing parts of the texture map to be transparent.

   POLYTYPE_ATEX_LIT:
   POLYTYPE_PTEX_LIT:
      Like POLYTYPE_ATEX and POLYTYPE_PTEX, but the global color_map table 
      (for 256 color modes) or blender function (for non-MMX truecolor 
      modes) is used to blend the texture with a light level taken from the 
      c value in the vertex structure. This must only be used after you have 
      set up the color mapping table or blender functions!

   POLYTYPE_ATEX_MASK_LIT:
   POLYTYPE_PTEX_MASK_LIT:
      Like POLYTYPE_ATEX_LIT and POLYTYPE_PTEX_LIT, but zero texture map 
      pixels are skipped, allowing parts of the texture map to be 
      transparent.

   If the cpu_mmx global variable is set, the GRGB and truecolor *LIT 
   routines will be optimised using MMX instructions. If the cpu_3dnow 
   global variable is set, the truecolor PTEX*LIT routines will take 
   advantage of the 3DNow! CPU extensions. For this reason, it is a good 
   idea to call check_cpu() before using the polygon functions.

   Using MMX for *LIT routines has a side effect: normally (without MMX), 
   these routines use the blender functions used also for other lighting 
   functions, set with set_trans_blender() or set_blender_mode(). The MMX 
   versions only use the RGB value passed to set_trans_blender() and do the 
   linear interpolation themselves. Therefore a new set of blender functions 
   passed to set_blender_mode() is ignored.

void triangle3d(BITMAP *bmp, int type, BITMAP *tex, V3D *v1, *v2, *v3);
void triangle3d_f(BITMAP *bmp, int type, BITMAP *tex, V3D_f *v1, *v2, *v3);
   Draw 3d triangles, using either fixed or floating point vertex 
   structures. These are equivalent to calling
      polygon3d(bmp, type, tex, 3, v1, v2, v3);
   or
      polygon3d_f(bmp, type, tex, 3, v1, v2, v3);

void quad3d(BITMAP *bmp, int type, BITMAP *tex, V3D *v1, *v2, *v3, *v4);
void quad3d_f(BITMAP *bmp, int type, BITMAP *tex, V3D_f *v1, *v2, *v3, *v4);
   Draw 3d quads, using either fixed or floating point vertex structures. 
   These are equivalent to calling
      polygon3d(bmp, type, tex, 4, v1, v2, v3, v4);
   or
      polygon3d_f(bmp, type, tex, 4, v1, v2, v3, v4);

int clip3d_f(int type, float min_z, float max_z, int vc,
             V3D_f *vtx[], V3D_f *vout[], V3D_f *vtmp[], int out[]);
   Clips the polygon given in vtx. The number of vertices is vc, the result 
   goes in vout, and vtmp and out are needed for internal purposes. The 
   pointers in vtx, vout and vtmp must point to valid V3D_f structures. The 
   size of vout, vtmp and out should be at least vc*2 (additional vertices 
   may appear in the process of clipping). The frustum (viewing volume) is 
   defined by -z<x<z, -z<y<z, 0<min_z<z<max_z. If 
   max_z<=min_z, the z<max_z clipping is not done. As you can see, 
   clipping is done in the camera space, with perspective in mind, so this 
   routine should be called after you apply the camera matrix, but before 
   the perspective projection. The routine will correctly interpolate u, v, 
   and c in the vertex structure. However, no provision is made for 
   high/truecolor GCOL.



============================================================
============ Transparency and patterned drawing ============
============================================================

void drawing_mode(int mode, BITMAP *pattern, int x_anchor, int y_anchor);
   Sets the graphics drawing mode. This only affects the geometric routines 
   like putpixel, lines, rectangles, circles, polygons, floodfill, etc, not 
   the text output, blitting, or sprite drawing functions. The mode should 
   be one of the values:

      DRAW_MODE_SOLID               - the default, solid color drawing
      DRAW_MODE_XOR                 - exclusive-or drawing
      DRAW_MODE_COPY_PATTERN        - multicolored pattern fill
      DRAW_MODE_SOLID_PATTERN       - single color pattern fill
      DRAW_MODE_MASKED_PATTERN      - masked pattern fill
      DRAW_MODE_TRANS               - translucent color blending

   In DRAW_MODE_XOR, pixels are written to the bitmap with an exclusive-or 
   operation rather than a simple copy, so drawing the same shape twice will 
   erase it. Because it involves reading as well as writing the bitmap 
   memory, xor drawing is a lot slower than the normal replace mode.

   With the patterned modes, you provide a pattern bitmap which is tiled 
   across the surface of the shape. Allegro stores a pointer to this bitmap 
   rather than copying it, so you must not destroy the bitmap while it is 
   still selected as the pattern. The width and height of the pattern must 
   be powers of two, but they can be different, eg. a 64x16 pattern is fine, 
   but a 17x3 one is not. The pattern is tiled in a grid starting at point 
   (x_anchor, y_anchor). Normally you should just pass zero for these 
   values, which lets you draw several adjacent shapes and have the patterns 
   meet up exactly along the shared edges. Zero alignment may look peculiar 
   if you are moving a patterned shape around the screen, however, because 
   the shape will move but the pattern alignment will not, so in some 
   situations you may wish to alter the anchor position.

   When you select DRAW_MODE_COPY_PATTERN, pixels are simply copied from the 
   pattern bitmap onto the destination bitmap. This allows the use of 
   multicolored patterns, and means that the color you pass to the drawing 
   routine is ignored. This is the fastest of the patterned modes.

   In DRAW_MODE_SOLID_PATTERN, each pixel in the pattern bitmap is compared 
   with the mask color, which is zero in 256 color modes or bright pink for 
   truecolor data (maximum red and blue, zero green). If the pattern pixel 
   is solid, a pixel of the color you passed to the drawing routine is 
   written to the destination bitmap, otherwise a zero is written. The 
   pattern is thus treated as a monochrome bitmask, which lets you use the 
   same pattern to draw different shapes in different colors, but prevents 
   the use of multicolored patterns.

   DRAW_MODE_MASKED_PATTERN is almost the same as DRAW_MODE_SOLID_PATTERN, 
   but the masked pixels are skipped rather than being written as zeros, so 
   the background shows through the gaps.

   In DRAW_MODE_TRANS, the global color_map table or truecolor blender 
   functions are used to overlay pixels on top of the existing image. This 
   must only be used after you have set up the color mapping table (for 256 
   color modes) or blender map (for truecolor modes). Because it involves 
   reading as well as writing the bitmap memory, translucent drawing is very 
   slow if you draw directly to video RAM, so wherever possible you should 
   use a memory bitmap instead.

void xor_mode(int xor);
   This is a shortcut for toggling xor drawing mode on and off. Calling 
   xor_mode(TRUE) is equivalent to drawing_mode(DRAW_MODE_XOR, NULL, 0, 0);

void solid_mode();
   This is a shortcut for selecting solid drawing mode. It is equivalent to 
   calling drawing_mode(DRAW_MODE_SOLID, NULL, 0, 0);

In paletted video modes, translucency and lighting are implemented with a 
64k lookup table, which contain the result of combining any two colors c1 
and c2. You must set up this table before you use any of the lighting or 
translucency routines. Depending on how you construct the table, a range of 
different effects are possible. For example, translucency can be implemented 
by using a color halfway between c1 and c2 as the result of the combination. 
Lighting is achieved by treating one of the colors as a light level (0-255) 
rather than a color, and setting up the table appropriately. A range of 
specialised effects are possible, for instance replacing any color with any 
other color and making individual source or destination colors completely 
solid or invisible.

Color mapping tables can be precalculated with the colormap utility, or 
generated at runtime. The COLOR_MAP structure is defined as:

typedef struct {
   unsigned char data[PAL_SIZE][PAL_SIZE];
} COLOR_MAP;

extern COLOR_MAP *color_map;
   Global pointer to the color mapping table. This must be set before using 
   any translucent or lit drawing functions in a 256 color video mode!

void create_light_table(COLOR_MAP *table, PALETTE pal,
                        int r, g, b, void (*callback)(int pos));
   Fills the specified color mapping table with lookup data for doing 
   lighting effects with the specified palette. When combining the colors c1 
   and c2 with this table, c1 is treated as a light level from 0-255. At 
   light level 255 the table will output color c2 unchanged, at light level 
   0 it will output the r, g, b value you specify to this function, and at 
   intermediate light levels it will output a color somewhere between the 
   two extremes. If the callback function is not NULL, it will be called 256 
   times during the calculation, allowing you to display a progress 
   indicator.

void create_trans_table(COLOR_MAP *table, PALETTE pal,
                        int r, g, b, void (*callback)(int pos));
   Fills the specified color mapping table with lookup data for doing 
   translucency effects with the specified palette. When combining the 
   colors c1 and c2 with this table, the result will be a color somewhere 
   between the two. The r, g, and b parameters specify the solidity of each 
   color component, ranging from 0 (totally transparent) to 255 (totally 
   solid). For 50% solidity, pass 128. This function treats source color #0 
   as a special case, leaving the destination unchanged whenever a zero 
   source pixel is encountered, so that masked sprites will draw correctly. 
   If the callback function is not NULL, it will be called 256 times during 
   the calculation, allowing you to display a progress indicator.

void create_color_table(COLOR_MAP *table, PALETTE pal,
                        RGB (*blend)(PALETTE pal, int x, int y),
                        void (*callback)(int pos));
   Fills the specified color mapping table with lookup data for doing 
   customised effects with the specified palette, calling the blend function 
   to determine the results of each color combination. Your blend routine 
   will be passed a pointer to the palette and the two colors which are to 
   be combined, and should return the desired result in RGB (0-63) format. 
   Allegro will then search the palette for the closest match to the RGB 
   color that you requested, so it doesn't matter if the palette has no 
   exact match for this color. If the callback function is not NULL, it will 
   be called 256 times during the calculation, allowing you to display a 
   progress indicator.

In truecolor video modes, translucency and lighting are implemented with a 
set of blender functions in the form:

unsigned long (*BLENDER_FUNC)(unsigned long x, unsigned long y);

This routine takes two color parameters, decomposes them into their red, 
green, and blue components, combines them in whatever way is appropriate, 
and then merges the result back into a single return color value. To support 
varying alpha (ie. different amounts of translucency or light levels), there 
are 256 versions of the blender function: one for every possible alpha 
value. These are stored in a table:

typedef struct {
   BLENDER_FUNC blend[256];
} BLENDER_MAP;

Since these routines may be used from various different color depths, there 
are three such tables, one for use with 15 bit 5.5.5 pixels, one for 16 bit 
5.6.5 pixels, and one for 24 bit 8.8.8 pixels (this can be shared between 
the 24 and 32 bit code since the bit packing is the same).

void set_trans_blender(int r, int g, int b, int a);
   Selects the default set of truecolor blender routines, which perform a 
   simple linear interpolation between the source and destination colors. 
   When a translucent drawing function is called, the alpha parameter set by 
   this routine is used to select one of the blenders from the table, and 
   that function is then called to blend each pixel with the existing 
   destination color (ie. the alpha parameter controls the solidity of the 
   drawing, from 0 to 255). When a lit sprite drawing function is called, 
   the alpha value passed to this routine is ignored, and instead the color 
   passed to the sprite function is used to select an alpha level. The 
   blender routine will then be used to interpolate between the sprite color 
   and the RGB value that was passed to this function (ranging 0-255).

void set_blender_mode(BLENDER_MAP *b15, *b16, *b24, int r, g, b, a);
   Specifies a custom set of truecolor blender routines, providing a table 
   of function pointers for every possible color depth (these parameters may 
   be NULL if you aren't going to use that pixel format).



==========================================================
============ Converting between color formats ============
==========================================================

In general, Allegro is designed to be used in only one color depth at a 
time, so you will call set_color_depth() once and then store all your 
bitmaps in the same format. If you want to mix several different pixel 
formats, you can use create_bitmap_ex() in place of create_bitmap(), and 
call bitmap_color_depth() to query the format of a specific image. Most of 
the graphics routines require all their input parameters to be in the same 
format (eg. you cannot stretch a 15 bit source bitmap onto a 24 bit 
destination), but there are three exceptions: blit() can copy between 
bitmaps of any format, converting the data as required, draw_sprite() can 
draw 256 color source images onto destinations of any format, and 
draw_character() _always_ uses a 256 color source bitmap, whatever the 
format of the destination.

Expanding a 256 color source onto a truecolor destination is fairly fast 
(obviously you must set the correct palette before doing this conversion!). 
Converting between different truecolor formats is slightly slower, and 
reducing truecolor images to a 256 color destination is very slow (it can be 
sped up significantly if you set up the global rgb_map table before doing 
the conversion).

int bestfit_color(PALLETE pal, int r, int g, int b);
   Searches the specified palette for the closest match to the requested 
   color, which are specified in the VGA hardware 0-63 format. Normally you 
   should call makecol8() instead, but this lower level function may be 
   useful if you need to use a palette other than the currently selected 
   one, or specifically don't want to use the rgb_map lookup table.

extern RGB_MAP *rgb_map;
   To speed up reducing RGB values to 8 bit paletted colors, Allegro uses a 
   32k lookup table (5 bits for each color component). You must set up this 
   table before using the gouraud shading routines, and if present the table 
   will also vastly accelerate the makecol8() function. RGB tables can be 
   precalculated with the rgbmap utility, or generated at runtime. The 
   RGB_MAP structure is defined as:

   typedef struct {
      unsigned char data[32][32][32];
   } RGB_MAP;

void create_rgb_table(RGB_MAP *table, PALETTE pal, void (*callback)(int pos));
   Fills the specified RGB mapping table with lookup data for the specified 
   palette. If the callback function is not NULL, it will be called 256 
   times during the calculation, allowing you to display a progress 
   indicator.

void hsv_to_rgb(float h, float s, float v, int *r, int *g, int *b);
void rgb_to_hsv(int r, int g, int b, float *h, float *s, float *v);
   Convert color values between the HSV and RGB colorspaces. The RGB values 
   range from 0 to 255, hue is from 0 to 360, and saturation and value are 
   from 0 to 1.



=======================================================
============ Direct access to video memory ============
=======================================================

The bitmap structure looks like:

typedef struct BITMAP
{
   int w, h;               - size of the bitmap in pixels
   int clip;               - non-zero if clipping is turned on
   int cl, cr, ct, cb;     - clip rectangle left, right, top, and bottom
   int seg;                - segment for use with the line pointers
   unsigned char *line[];  - pointers to the start of each line
} BITMAP;

There is some other stuff in the structure as well, but it is liable to 
change and you shouldn't use anything except the above. The clipping 
rectangle is inclusive on the left and top (0 allows drawing to position 0) 
but exclusive on the right and bottom (10 allows drawing to position 9, but 
not to 10). Note this is not the same format as you pass to set_clip(), 
which takes inclusive coordinates for all four corners.

There are several ways to get direct access to the image memory of a bitmap, 
varying in complexity depending on what sort of bitmap you are using.

The simplest approach will only work with memory bitmaps (obtained from 
create_bitmap(), grabber datafiles, and image files) and sub-bitmaps of 
memory bitmaps. This uses a table of char pointers, called 'line', which is 
a part of the bitmap structure and contains pointers to the start of each 
line of the image. For example, a simple memory bitmap putpixel function is:

   void memory_putpixel(BITMAP *bmp, int x, int y, int color)
   {
      bmp->line[y][x] = color;
   }

For truecolor modes you need to cast the line pointer to the appropriate 
type, for example:

   void memory_putpixel_15_or_16_bpp(BITMAP *bmp, int x, int y, int color)
   {
      ((short *)bmp->line[y])[x] = color;
   }

   void memory_putpixel_32(BITMAP *bmp, int x, int y, int color)
   {
      ((long *)bmp->line[y])[x] = color;
   }

If you want to write to the screen as well as to memory bitmaps, you need to 
use far pointers. Rewriting the above function to use the routines in 
sys/farptr.h will allow it to be used with screen bitmaps, as long as they 
do not require bank switching (this means mode 13h screens and linear 
framebuffer SVGA screens). Using far pointers, the putpixel is:

   #include <sys/farptr.h>

   void farptr_putpixel(BITMAP *bmp, int x, int y, int color)
   {
      _farpokeb(bmp->seg, (unsigned long)bmp->line[y]+x, color);
   }

Obviously for truecolor modes you should replace the _farpokeb() with 
_farpokew() or _farpokel(), and multiply the x offset by the number of bytes 
per pixel.

This still won't work in banked SVGA modes, however. For more 
flexible access to bitmap memory, you need to call the bank switching 
functions:

unsigned long bmp_write_line(BITMAP *bmp, int line);
   Selects the line of a bitmap that you are going to draw onto.

unsigned long bmp_read_line(BITMAP *bmp, int line);
   Selects the line of a bitmap that you are going to read from.

These are implemented as inline assembler routines, so they are not as 
inefficient as they might seem. If the bitmap doesn't require bank switching 
(ie. it is a memory bitmap, mode 13h screen, etc), these functions just 
return bmp->line[line].

Although SVGA bitmaps are banked, Allegro provides linear access to the 
memory within each scanline, so you only need to pass a y coordinate to 
these functions. Various x positions can be obtained by simply adding the x 
coordinate to the returned address. The return value is an unsigned long 
rather than a char pointer because the bitmap memory may not be in your data 
segment, and you need to access it with far pointers. For example, a 
putpixel using the bank switching functions is:

   #include <sys/farptr.h>

   void banked_putpixel(BITMAP *b, int x, int y, int color)
   {
      unsigned long address = bmp_write_line(bmp, y);
      _farpokeb(bmp->seg, address+x, color);
   }

You will notice that Allegro provides separate functions for setting the 
read and write banks. It is important that you distinguish between these, 
because on some graphics cards the banks can be set individually, and on 
others the video memory is read and written at different addresses. Life is 
never quite as simple as we might wish it to be, though (this is true even 
when we _aren't_ talking about graphics coding :-) and so of course some 
cards only provide a single bank. On these the read and write bank functions 
will behave identically, so you shouldn't assume that you can read from one 
part of video memory and write to another at the same time. You can call 
bmp_read_line(), and read whatever you like from that line, and then call 
bmp_write_line() with the same or a different line number, and write 
whatever you like to this second line, but you mustn't call bmp_read_line() 
and bmp_write_line() together and expect to be able to read one line and 
write the other simultaneously. It would be nice if this was possible, but 
if you do it, your code won't work on single banked SVGA cards.

And then there's mode-X. If you've never done any mode-X graphics coding, 
you probably won't understand this, but for those of you who want to know 
how Allegro sets up the mode-X screen bitmaps, here goes...

The line pointers are still present, and they contain planar addresses, ie. 
the actual location at which you access the first pixel in the line. These 
addresses are guaranteed to be quad aligned, so you can just set the write 
plane, divide your x coordinate by four, and add it to the line pointer. For 
example, a mode-X putpixel is:

   #include <pc.h>
   #include <sys/farptr.h>

   void modex_putpixel(BITMAP *b, int x, int y, int color)
   {
      outportw(0x3C4, (0x100<<(x&3))|2);
      _farpokeb(bmp->seg, (unsigned long)bmp->line[y]+(x>>2), color);
   }

Oh yeah: the nearptr hack. Personally I don't like this very much, but a lot 
of people swear by it because it can give you direct access to the screen 
memory via a normal C pointer. Warning: this method will only work in VGA 
mode 13h and linear framebuffer modes!

In your setup code:

   #include <sys/nearptr.h>

   unsigned char *screenmemory;
   unsigned long screen_base_addr;

   __djgpp_nearptr_enable();

   __dpmi_get_segment_base_address(screen->seg, &screen_base_addr);

   screenmemory = (unsigned char *)(screen_base_addr + 
                                    screen->line[0] - 
                                    __djgpp_base_address);

Then:

   void nearptr_putpixel(int x, int y, int color)
   {
      screenmemory[x + y*SCREEN_W] = color;
   }



=======================================
============ FLIC routines ============
=======================================

There are two high level functions for playing FLI/FLC animations: 
play_fli(), which reads the data directly from disk, and play_memory_fli(), 
which uses data that has already been loaded into RAM. Apart from the 
different sources of the data, these two functions behave identically. They 
draw the animation onto the specified bitmap, which should normally be the 
screen. Frames will be aligned with the top left corner of the bitmap: if 
you want to position them somewhere else you will need to create a 
sub-bitmap for the FLI player to draw onto. If loop is set the player will 
cycle when it reaches the end of the file, otherwise it will play through 
the animation once and then return. If the callback function is not NULL it 
will be called once for each frame, allowing you to perform background tasks 
of your own. This callback should normally return zero: if it returns 
non-zero the player will terminate (this is the only way to stop an 
animation that is playing in looped mode). The FLI player returns FLI_OK if 
it reached the end of the file, FLI_ERROR if something went wrong, and the 
value returned by the callback function if that was what stopped it. If you 
need to distinguish between different return values, your callback should 
return positive integers, since FLI_OK is zero and FLI_ERROR is negative. 
Note that the FLI player will only work when the timer module is installed, 
and that it will alter the palette according to whatever palette data is 
present in the animation file.

Occasionally you may need more detailed control over how an FLI is played, 
for example if you want to superimpose a text scroller on top of the 
animation, or to play it back at a different speed. You could do both of 
these with the lower level functions described below.

int play_fli(char *filename, BITMAP *bmp, int loop, int (*callback)());
   Plays an Autodesk Animator FLI or FLC animation file, reading the data 
   from disk as it is required.

int play_memory_fli(void *fli_data, BITMAP *bmp, int loop, int (*callback)());
   Plays an Autodesk Animator FLI or FLC animation, reading the data from a 
   copy of the file which is held in memory. You can obtain the fli_data 
   pointer by mallocing a block of memory and reading an FLI file into it, 
   or by importing an FLI into a grabber datafile. Playing animations from 
   memory is obviously faster than cueing them directly from disk, and is 
   particularly useful with short, looped FLI's. Animations can easily get 
   very large, though, so in most cases you will probably be better just 
   using play_fli().

int open_fli(char *filename);
int open_memory_fli(void *fli_data);
   Open FLI files ready for playing, reading the data from disk or memory 
   respectively. Return FLI_OK on success. Information about the current FLI 
   is held in global variables, so you can only have one animation open at a 
   time.

void close_fli();
   Closes an FLI file when you have finished reading from it.

int next_fli_frame(int loop);
   Reads the next frame of the current animation file. If loop is set the 
   player will cycle when it reaches the end of the file, otherwise it will 
   return FLI_EOF. Returns FLI_OK on success, FLI_ERROR or FLI_NOT_OPEN on 
   error, and FLI_EOF on reaching the end of the file. The frame is read 
   into the global variables fli_bitmap and fli_palette.

extern BITMAP *fli_bitmap;
   Contains the current frame of the FLI/FLC animation.

extern PALETTE fli_palette;
   Contains the current FLI palette.

extern int fli_bmp_dirty_from;
extern int fli_bmp_dirty_to;
   These variables are set by next_fli_frame() to indicate which part of the 
   fli_bitmap has changed since the last call to reset_fli_variables(). If 
   fli_bmp_dirty_from is greater than fli_bmp_dirty_to, the bitmap has not 
   changed, otherwise lines fli_bmp_dirty_from to fli_bmp_dirty_to 
   (inclusive) have altered. You can use these when copying the fli_bitmap 
   onto the screen, to avoid moving data unnecessarily.

extern int fli_pal_dirty_from;
extern int fli_pal_dirty_to;
   These variables are set by next_fli_frame() to indicate which part of the 
   fli_palette has changed since the last call to reset_fli_variables(). If 
   fli_pal_dirty_from is greater than fli_pal_dirty_to, the palette has not 
   changed, otherwise colors fli_pal_dirty_from to fli_pal_dirty_to 
   (inclusive) have altered. You can use these when updating the hardware 
   palette, to avoid unnecessary calls to set_palette().

void reset_fli_variables();
   Once you have done whatever you are going to do with the fli_bitmap and 
   fli_palette, call this function to reset the fli_bmp_dirty_* and 
   fli_pal_dirty_* variables.

extern int fli_frame;
   Global variable containing the current frame number in the FLI file. This 
   is useful for synchronising other events with the animation, for instance 
   you could check it in a play_fli() callback function and use it to 
   trigger a sample at a particular point.

extern volatile int fli_timer;
   Global variable for timing FLI playback. When you open an FLI file, a 
   timer interrupt is installed which increments this variable every time a 
   new frame should be displayed. Calling next_fli_frame() decrements it, so 
   you can test it and know that it is time to display a new frame if it is 
   greater than zero.



=============================================
============ Sound init routines ============
=============================================

int detect_digi_driver(int driver_id);
   Detects whether the specified digital sound device is available, using 
   the same ID values as install_sound(). Returns the maximum number of 
   voices that the driver can provide, or zero if the hardware is not 
   present. This function must be called _before_ install_sound().

int detect_midi_driver(int driver_id);
   Detects whether the specified MIDI sound device is available, using the 
   same ID values as install_sound(). Returns the maximum number of voices 
   that the driver can provide, or zero if the hardware is not present. 
   There are two special-case return values that you should watch out for: 
   if this function returns -1 it is a note-stealing driver (eg. DIGMID) 
   that shares voices with the current digital sound driver, and if it 
   returns 0xFFFF it is an external device like an MPU-401 where there is no 
   way to determine how many voices are available. This function must be 
   called _before_ install_sound(). 

void reserve_voices(int digi_voices, int midi_voices);
   Call this function to specify the number of voices that are to be used by 
   the digital and MIDI sound drivers respectively. This must be done 
   _before_ calling install_sound(). If you reserve too many voices, 
   subsequent calls to install_sound() will fail. How many voices are 
   available depends on the driver, and in some cases you will actually get 
   more than you reserve (eg. the FM synth drivers will always provide 9 
   voices on an OPL2 and 18 on an OPL3, and the SB digital driver will round 
   the number of voices up to the nearest power of two). Pass negative 
   values to restore the default settings. You should be aware that the 
   sound quality is usually inversely related to how many voices you use, so 
   don't reserve any more than you really need.

int install_sound(int digi_card, int midi_card, char *cfg_path); 
   Initialises the sound module, returning zero on success. The digi_card 
   parameter should be one of the values:

      DIGI_AUTODETECT      - let Allegro pick a digital sound driver
      DIGI_NONE            - no digital sound
      DIGI_SB              - Sound Blaster (autodetect type)
      DIGI_SB10            - SB 1.0 (8 bit mono single shot dma)
      DIGI_SB15            - SB 1.5 (8 bit mono single shot dma)
      DIGI_SB20            - SB 2.0 (8 bit mono auto-initialised dma)
      DIGI_SBPRO           - SB Pro (8 bit stereo)
      DIGI_SB16            - SB16 (16 bit stereo)
      DIGI_AUDIODRIVE      - ESS AudioDrive
      DIGI_SOUNDSCAPE      - Ensoniq Soundscape

   The midi_card should be one of the values:

      MIDI_AUTODETECT      - let Allegro pick a MIDI sound driver
      MIDI_NONE            - no MIDI sound
      MIDI_ADLIB           - Adlib or SB FM synth (autodetect type)
      MIDI_OPL2            - OPL2 synth (mono, used in Adlib and SB)
      MIDI_2XOPL2          - dual OPL2 synths (stereo, used in SB Pro-I)
      MIDI_OPL3            - OPL3 synth (stereo, SB Pro-II and above)
      MIDI_SB_OUT          - SB MIDI interface
      MIDI_MPU             - MPU-401 MIDI interface
      MIDI_DIGMID          - sample-based software wavetable player
      MIDI_AWE32           - AWE32 (EMU8000 chip)

   You should normally give the DIGI_AUTODETECT and MIDI_AUTODETECT 
   parameters to this function, in which case Allegro will read hardware 
   settings from the current configuration file. This allows the user to 
   select different values with the setup utility: see the config section 
   for details.

   The cfg_path parameter is only present for compatibility with previous 
   versions of Allegro, and has no effect on anything.

   Returns zero if the sound is successfully installed, and -1 on failure. 
   If it fails it will store a description of the problem in allegro_error.

void remove_sound();
   Cleans up after you are finished with the sound routines. You don't 
   normally need to call this, because allegro_exit() will do it for you.

void set_volume(int digi_volume, int midi_volume);
   Alters the global sound output volume. Specify volumes for both digital 
   samples and MIDI playback, as integers from 0 to 255, or pass a negative 
   value to leave one of the settings unchanged. If possible this routine 
   will use a hardware mixer to control the volume, otherwise it will tell 
   the sample and MIDI players to simulate a mixer in software.



=================================================
============ Digital sample routines ============
=================================================

SAMPLE *load_sample(char *filename);
   Loads a sample from a file, returning a pointer to it, or NULL on error. 
   At present this function supports both mono and stereo WAV and mono VOC 
   files, in 8 or 16 bit formats.

SAMPLE *load_wav(char *filename);
   Loads a sample from a RIFF WAV file.

SAMPLE *load_voc(char *filename);
   Loads a sample from a Creative Labs VOC file.

SAMPLE *create_sample(int bits, int stereo, int freq, int len);
   Constructs a new sample structure of the specified type. The data field 
   points to a block of waveform data: see the structure definition in 
   allegro.h for details.

void destroy_sample(SAMPLE *spl);
   Destroys a sample structure when you are done with it. It is safe to call 
   this even when the sample might be playing, because it checks and will 
   kill it off if it is active.

int play_sample(SAMPLE *spl, int vol, int pan, int freq, int loop);
   Triggers a sample at the specified volume, pan position, and frequency. 
   The volume and pan range from 0 (min/left) to 255 (max/right). Frequency 
   is relative rather than absolute: 1000 represents the frequency that the 
   sample was recorded at, 2000 is twice this, etc. If the loop flag is set, 
   the sample will repeat until you call stop_sample(), and can be 
   manipulated while it is playing by calling adjust_sample().

void adjust_sample(SAMPLE *spl, int vol, int pan, int freq, int loop);
   Alters the parameters of a sample while it is playing (useful for 
   manipulating looped sounds). You can alter the volume, pan, and 
   frequency, and can also clear the loop flag, which will stop the sample 
   when it next reaches the end of its loop. If there are several copies of 
   the same sample playing, this will adjust the first one it comes across. 
   If the sample is not playing it has no effect.

void stop_sample(SAMPLE *spl);
   Kills off a sample, which is required if you have set a sample going in 
   looped mode. If there are several copies of the sample playing, it will 
   stop them all.

If you need more detailed control over how samples are played, you can use 
the lower level voice functions rather than just calling play_sample(). This 
is rather more work, because you have to explicitly allocate and free the 
voices rather than them being automatically released when they finish 
playing, but allows far more precise specification of exactly how you want 
everything to sound. You may also want to modify a couple of fields from the 
sample structure:

   int priority;
      Ranging 0-255 (default 128), this controls how voices are allocated if 
      you attempt to play more than the driver can handle. This may be used 
      to ensure that the less important sounds are cut off while the 
      important ones are preserved.

   unsigned long loop_start;
   unsigned long loop_end;
      Loop position in sample units, by default set to the start and end of 
      the sample.

int allocate_voice(SAMPLE *spl);
   Allocates a soundcard voice and prepares it for playing the specified 
   sample, setting up sensible default parameters (maximum volume, centre 
   pan, no change of pitch, no looping). When you are finished with the 
   voice you must free it by calling deallocate_voice() or release_voice(). 
   Returns the voice number, or -1 if no voices are available.

void deallocate_voice(int voice);
   Frees a soundcard voice, stopping it from playing and releasing whatever 
   resources it is using.

void reallocate_voice(int voice, SAMPLE *spl);
   Switches an already-allocated voice to use a different sample. Calling 
   reallocate_voice(voice, sample) is equivalent to:

      deallocate_voice(voice);
      voice = allocate_voice(sample);

void release_voice(int voice);
   Releases a soundcard voice, indicating that you are no longer interested 
   in manipulating it. The sound will continue to play, and any resources 
   that it is using will automatically be freed when it finishes. This is 
   essentially the same as deallocate_voice(), but it waits for the sound to 
   stop playing before taking effect.

void voice_start(int voice);
   Activates a voice, using whatever parameters have been set for it.

void voice_stop(int voice);
   Stops a voice, storing the current position and state so that it may 
   later be resumed by calling voice_start().

void voice_set_priority(int voice, int priority);
   Sets the priority of a voice (range 0-255). This is used to decide which 
   voices should be chopped off, if you attempt to play more than the 
   soundcard driver can handle.

SAMPLE *voice_check(int voice);
   Checks whether a voice is currently allocated. It returns a copy of the 
   sample that the voice is using, or NULL if the voice is inactive (ie. it 
   has been deallocated, or the release_voice() function has been called and 
   the sample has then finished playing).

int voice_get_position(int voice);
   Returns the current position of a voice, in sample units, or -1 if it has 
   finished playing.

void voice_set_position(int voice, int position);
   Sets the position of a voice, in sample units.

void voice_set_playmode(int voice, int playmode);
   Adjusts the loop status of the specified voice. This can be done while 
   the voice is playing, so you can start a sample in looped mode (having 
   set the loop start and end positions to the appropriate values), and then 
   clear the loop flag when you want to end the sound, which will cause it 
   to continue past the loop end, play the subsequent part of the sample, 
   and finish in the normal way. The mode parameter is a bitfield containing 
   the following values:
   
   PLAYMODE_PLAY
      Plays the sample a single time. This is the default if you don't set 
      the loop flag.
   
   PLAYMODE_LOOP
      Loops repeatedly through the sample, jumping back to the loop start 
      position upon reaching the loop end.
   
   PLAYMODE_FORWARD
      Plays the sample from beginning to end. This is the default if you 
      don't set the backward flag.
   
   PLAYMODE_BACKWARD
      Reverses the direction of the sample. If you combine this with the 
      loop flag, the sample jumps to the loop end position upon reaching the 
      loop start (ie. you do not need to reverse the loop start and end 
      values when you play the sample in reverse).
   
   PLAYMODE_BIDIR
      When used in combination with the loop flag, causes the sample to 
      change direction each time it reaches one of the loop points, so it 
      alternates between playing forwards and in reverse.
   
int voice_get_volume(int voice);
   Returns the current volume of the voice, range 0-255.

void voice_set_volume(int voice, int volume);
   Sets the volume of the voice, range 0-255.

void voice_ramp_volume(int voice, int time, int endvol);
   Starts a volume ramp (crescendo or diminuendo) from the current volume to 
   the specified ending volume, lasting for time milliseconds.

void voice_stop_volumeramp(int voice);
   Interrupts a volume ramp operation.

int voice_get_frequency(int voice);
   Returns the current pitch of the voice, in Hz.

void voice_set_frequency(int voice, int frequency);
   Sets the pitch of the voice, in Hz.

void voice_sweep_frequency(int voice, int time, int endfreq);
   Starts a frequency sweep (glissando) from the current pitch to the 
   specified ending pitch, lasting for time milliseconds.

void voice_stop_frequency_sweep(int voice);
   Interrupts a frequency sweep operation.

int voice_get_pan(int voice);
   Returns the current pan position, from 0 (left) to 255 (right).

void voice_set_pan(int voice, int pan);
   Sets the pan position, ranging from 0 (left) to 255 (right).

void voice_sweep_pan(int voice, int time, int endpan);
   Starts a pan sweep (left  right movement) from the current position to 
   the specified ending position, lasting for time milliseconds.

void voice_stop_pan_sweep(int voice);
   Interrupts a pan sweep operation.

void voice_set_echo(int voice, int strength, int delay);
   Sets the echo parameters for a voice (not currently implemented).

void voice_set_tremolo(int voice, int rate, int depth);
   Sets the tremolo parameters for a voice (not currently implemented).

void voice_set_vibrato(int voice, int rate, int depth);
   Sets the vibrato parameters for a voice (not currently implemented).



=============================================
============ MIDI music routines ============
=============================================

MIDI *load_midi(char *filename);
   Loads a MIDI file (handles both format 0 and format 1), returning a 
   pointer to a MIDI structure, or NULL on error.

void destroy_midi(MIDI *midi);
   Destroys a MIDI structure when you are done with it. It is safe to call 
   this even when the MIDI file might be playing, because it checks and will 
   kill it off if it is active.

int play_midi(MIDI *midi, int loop);
   Starts playing the specified MIDI file, first stopping whatever music was 
   previously playing. If the loop flag is set, the data will be repeated 
   until replaced with something else, otherwise it will stop at the end of 
   the file. Passing a NULL pointer will stop whatever music is currently 
   playing. Returns non-zero if an error occurs (this may happen if a 
   patch-caching wavetable driver is unable to load the required samples, or 
   at least it might in the future when somebody writes some patch-caching 
   wavetable drivers :-)

int play_looped_midi(MIDI *midi, int loop_start, int loop_end);
   Starts playing a MIDI file with a user-defined loop position. When the 
   player reaches the loop end position or the end of the file (loop_end may 
   be -1 to only loop at EOF), it will wind back to the loop start point. 
   Both positions are specified in the same beat number format as the 
   midi_pos variable.

void stop_midi();
   Stops whatever music is currently playing. This is the same thing as 
   calling play_midi(NULL, FALSE).

void midi_pause();
   Pauses the MIDI player.

void midi_resume();
   Resumes playback of a paused MIDI file.

int midi_seek(int target);
   Seeks to the given midi_pos in the current MIDI file. If the target is 
   earlier in the file than the current midi_pos it seeks from the 
   beginning; otherwise it seeks from the current position. Returns zero if 
   successful, non-zero if it hit the end of the file (1 means it stopped 
   playing, 2 means it looped back to the start). If this function stops 
   because it reached EOF, midi_pos will be set to the negative length of 
   the MIDI file.

void midi_out(unsigned char *data, int length);
   Streams a block of MIDI commands into the player in realtime, allowing 
   you to trigger notes, jingles, etc, over the top of whatever MIDI file is 
   currently playing.

int load_midi_patches();
   Forces the MIDI driver to load the entire set of patches ready for use. 
   You will not normally need to call this, because Allegro automatically 
   loads whatever data is required for the current MIDI file, but you must 
   call it before sending any program change messages via the midi_out() 
   command. Returns non-zero if an error occurred.

extern volatile long midi_pos;
   Stores the current position (beat number) in the MIDI file, or contains 
   a negative number if no music is currently playing. Useful for 
   synchronising animations with the music, and for checking whether a MIDI 
   file has finished playing.

extern long midi_loop_start;
extern long midi_loop_end;
   The loop start and end points, set by the play_looped_midi() function. 
   These may safely be altered while the music is playing, but you should be 
   sure they are always set to sensible values (start < end). If you are 
   changing them both at the same time, make sure to alter them in the right 
   order in case a MIDI interrupt happens to occur in between your two 
   writes! Setting these values to -1 represents the start and end of the 
   file respectively.

extern void (*midi_msg_callback)(int msg, int byte1, int byte2);
extern void (*midi_meta_callback)(int type, unsigned char *data, int length);
extern void (*midi_sysex_callback)(unsigned char *data, int length);
   Hook functions allowing you to intercept MIDI player events. If set to 
   anything other than NULL, these routines will be called for each MIDI 
   message, meta-event, and system exclusive data block respectively. They 
   will execute in an interrupt handler context, so all the code and data 
   they use should be locked, and they must not call any operating system 
   functions. In general you just use these routines to set some flags and 
   respond to them later in your mainline code.

int load_ibk(char *filename, int drums);
   Reads in a .IBK patch definition file for use by the Adlib driver. If 
   drums is set, it will load it as a percussion patch set, otherwise it 
   will use it as a replacement set of General MIDI instruments. You may 
   call this before or after initialising the sound code, or can simply set 
   the ibk_file and ibk_drum_file variables in the configuration file to 
   have the data loaded automatically. Note that this function has no effect 
   on any drivers other than the Adlib one!



===============================================
============ Audio stream routines ============
===============================================

The audio stream functions are for playing digital sounds that are too big 
to fit in a regular SAMPLE structure, either because they are huge files 
that you want to load in pieces as the data is required, or because you are 
doing something clever like generating the waveform on the fly.

AUDIOSTREAM *play_audio_stream(int len, bits, stereo, freq, vol, pan);
   This function creates a new audio stream and starts it playing. The 
   length is the size of each transfer buffer (in samples), which should be 
   at least 2K: larger buffers are more efficient and require fewer updates, 
   but result in more latency between you providing the data and it actually 
   being played. The bits parameter must be 8 or 16, freq is the sample rate 
   of the data, and the vol and pan values use the same 0-255 ranges as the 
   regular sample playing functions. If you want to adjust the pitch, 
   volume, or panning of a stream once it is playing, you can use the 
   regular voice_*() functions with stream->voice as a parameter. The sample 
   data is always in unsigned format, with stereo waveforms consisting of 
   alternate left/right samples.

void stop_audio_stream(AUDIOSTREAM *stream);
   Destroys an audio stream when it is no longer required.

void *get_audio_stream_buffer(AUDIOSTREAM *stream);
   You must call this function at regular intervals while an audio stream is 
   playing, to provide the next buffer of sample data (the smaller the 
   stream buffer size, the more often it must be called). If it returns 
   NULL, the stream is still playing the previous lot of data, so you don't 
   need to do anything. If it returns a value, that is the location of the 
   next buffer to be played, and you should load the appropriate number of 
   samples (however many you specified when creating the stream) to that 
   address, for example using an fread() from a disk file. After filling the 
   buffer with data, call free_audio_stream_buffer() to indicate that the 
   new data is now valid. Note that this function should not be called from 
   a timer handler...

void free_audio_stream_buffer(AUDIOSTREAM *stream);
   Call this function after get_audio_stream_buffer() returns a non-NULL 
   address, to indicate that you have loaded a new block of samples to that 
   location and the data is now ready to be played.



============================================
============ Recording routines ============
============================================

int install_sound_input(int digi_card, int midi_card);
   Initialises the sound recorder module, returning zero on success. You 
   must install the normal sound playback system before calling this 
   routine. The two card parameters should use the same constants as 
   install_sound(), including DIGI_NONE and MIDI_NONE to disable parts of 
   the module, or DIGI_AUTODETECT and MIDI_AUTODETECT to guess the hardware.

void remove_sound_input();
   Cleans up after you are finished with the sound input routines. You don't 
   normally need to call this, because remove_sound() and/or allegro_exit() 
   will do it for you.

int get_sound_input_cap_bits();
   Checks which sample formats are supported by the current audio input 
   driver, returning one of the bitfield values:

      0 = audio input not supported
      8 = eight bit audio input is supported
      16 = sixteen bit audio input is supported
      24 = both eight and sixteen bit audio input are supported

int get_sound_input_cap_stereo();
   Checks whether the current audio input driver is capable of stereo 
   recording.

int get_sound_input_cap_rate(int bits, int stereo);
   Returns the maximum possible sample frequency for recording in the 
   specified format, or zero if these settings are not supported.

int get_sound_input_cap_parm(int rate, int bits, int stereo);
   Checks whether the specified recording frequency, number of bits, and 
   mono/stereo mode are supported by the current audio driver, returning one 
   of the values:

      0 = it is impossible to record in this format
      1 = recording is possible, but audio output will be suspended
      2 = recording is possible at the same time as playing other sounds
      -n = sampling rate not supported, but rate 'n' would work instead

int set_sound_input_source(int source);
   Selects the audio input source, returning zero on success or -1 if the 
   hardware does not provide an input select register. The parameter should 
   be one of the values:

      SOUND_INPUT_MIC
      SOUND_INPUT_LINE
      SOUND_INPUT_CD

int start_sound_input(int rate, int bits, int stereo);
   Starts recording in the specified format, suspending audio playback as 
   necessary (this will always happen with the current drivers). Returns the 
   buffer size in bytes if successful, or zero on error.

void stop_sound_input();
   Stops audio recording, switching the card back into the normal playback 
   mode.

int read_sound_input(void *buffer);
   Retrieves the most recently recorded audio buffer into the specified 
   location, returning non-zero if a buffer has been copied or zero if no 
   new data is yet available. The buffer size can be obtained by checking 
   the return value from start_sound_input(). You must be sure to call this 
   function at regular intervals during the recording (typically around 100 
   times a second), or some data will be lost. If you are unable to do this 
   often enough from the mainline code, use the digi_recorder() callback to 
   store the waveform into a larger buffer of your own. Note: many cards 
   produce a click or popping sound when switching between record and 
   playback modes, so it is often a good idea to discard the first buffer 
   after you start a recording. The waveform is always stored in unsigned 
   format, with stereo data consisting of alternate left/right samples.

extern void (*digi_recorder)();
   If set, this function is called by the input driver whenever a new sample 
   buffer becomes available, at which point you can use read_sound_input() 
   to copy the data into a more permenent location. This routine runs in an 
   interrupt context, so it must execute very quickly, the code and all 
   memory that it touches must be locked, and you cannot call any operating 
   system routines or access disk files.

extern void (*midi_recorder)(unsigned char data);
   If set, this function is called by the MIDI input driver whenever a new 
   byte of MIDI data becomes available. It runs in an interrupt context, so 
   it must execute very quickly and all the code/data must be locked.



=======================================================
============ File and compression routines ============
=======================================================

The following routines implement a fast buffered file I/O system, which 
supports the reading and writing of compressed files using a ring buffer 
algorithm based on the LZSS compressor by Haruhiko Okumura. This does not 
achieve quite such good compression as programs like zip and lha, but 
unpacking is very fast and it does not require much memory. Packed files 
always begin with the 32 bit value F_PACK_MAGIC, and autodetect files with 
the value F_NOPACK_MAGIC.

char *fix_filename_case(char *path);
   Converts a filename to a standardised case. On DOS platforms, they will 
   be entirely uppercase. Returns a copy of the path parameter.

char *fix_filename_slashes(char *path);
   Converts all the directory separators in a filename to a standard 
   character. On DOS platforms, this is a backslash. Returns a copy of the 
   path parameter.

char *fix_filename_path(char *dest, char *path, int size);
   Converts a partial filename into a full path, generating up to the 
   specified maximum number of characters. Returns a copy of the dest 
   parameter.

char *replace_filename(char *dest, char *path, char *filename, int size);
   Replaces the specified path+filename with a new filename tail, generating 
   up to the specified maximum number of characters. Returns a copy of the 
   dest parameter.

char *replace_extension(char *dest, char *filename, char *ext, int size);
   Replaces the specified filename+extension with a new extension tail, 
   generating up to the specified maximum number of characters. Returns a 
   copy of the dest parameter.

char *append_filename(char *dest, char *path, char *filename, int size);
   Concatenates the specified filename onto the end of the specified path, 
   generating up to the specified maximum number of characters. Returns a 
   copy of the dest parameter.

char *get_filename(char *path);
   When passed a completely specified file path, this returns a pointer to 
   the filename portion. Both '\' and '/' are recognized as directory 
   separators.

char *get_extension(char *filename);
   When passed a complete filename (with or without path information) this 
   returns a pointer to the file extension.

void put_backslash(char *filename);
   If the last character of the filename is not a '\', '/', or '#', this 
   routine will concatenate a '\' on to it.

int file_exists(char *filename, int attrib, int *aret);
   Checks whether a file matching the given name and attributes exists, 
   returning non-zero if it does. The file attribute may contain any of the 
   FA_* constants from dir.h. If aret is not NULL, it will be set to the 
   attributes of the matching file. If an error occurs the system error code 
   will be stored in errno.

int exists(char *filename);
   Shortcut version of file_exists(), which checks for normal files, which 
   may have the archive or read-only bits set, but are not hidden, 
   directories, system files, etc.

long file_size(char *filename);
   Returns the size of a file, in bytes. If the file does not exist or an 
   error occurs, it will return zero and store the system error code in 
   errno.

long file_time(char *filename);
   Returns the modification time of a file.

int delete_file(char *filename);
   Removes a file from the disk.

int for_each_file(char *name, int attrib, void (*callback)(), int param);
   Finds all the files on the disk which match the given wildcard 
   specification and file attributes, and executes callback() once for each. 
   callback() will be passed three arguments, the first a string which 
   contains the completed filename, the second being the attributes of the 
   file, and the third an int which is simply a copy of param (you can use 
   this for whatever you like). If an error occurs an error code will be 
   stored in errno, and callback() can cause for_each_file() to abort by 
   setting errno itself. Returns the number of successful calls made to 
   callback(). The file attribute may contain any of the FA_* flags from 
   dir.h.

void packfile_password(char *password);
   Sets the encryption password to be used for all future read/writes of 
   compressed files. Files written with an encryption password cannot be 
   read unless the same password is selected, so be careful: if you forget 
   the key, I can't make your data come back again! Pass NULL or an empty 
   string to return to the normal, non-encrypted mode. If you are using this 
   function to prevent people getting access to your datafiles, be careful 
   not to store an obvious copy of the password in your executable: if there 
   are any strings like "I'm the password for the datafile", it would be 
   fairly easy to get access to your data :-)

PACKFILE *pack_fopen(char *filename, char *mode);
   Opens a file according to mode, which may contain any of the flags:

      'r' - open file for reading.

      'w' - open file for writing, overwriting any existing data.

      'p' - open file in packed mode. Data will be compressed as it is 
            written to the file, and automatically uncompressed during read 
            operations. Files created in this mode will produce garbage if 
            they are read without this flag being set. 

      '!' - open file for writing in normal, unpacked mode, but add the 
            value F_NOPACK_MAGIC to the start of the file, so that it can 
            later be opened in packed mode and Allegro will automatically 
            detect that the data does not need to be decompressed.

   Instead of these flags, one of the constants F_READ, F_WRITE, 
   F_READ_PACKED, F_WRITE_PACKED or F_WRITE_NOPACK may be used as the mode 
   parameter. On success, pack_fopen() returns a pointer to a file 
   structure, and on error it returns NULL and stores an error code in 
   errno. An attempt to read a normal file in packed mode will cause errno 
   to be set to EDOM.

   The packfile functions also understand several "magic" filenames that are 
   used for special purposes. These are:

      "#" - read data that has been appended to your executable file with 
      the exedat utility, as if it was a regular independent disk file.

      'filename.dat#object_name' - open a specific object from a datafile, 
      and read from it as if it was a regular file. You can treat nested 
      datafiles exactly like a normal directory structure, for example 
      you could open 'filename.dat#graphics/level1/mapdata'.

      '#object_name' - combination of the above, reading an object from a 
      datafile that has been appended onto your executable.

   With these special filenames, the contents of a datafile object or 
   appended file can be read in an identical way to a normal disk file, so 
   any of the file access functions in Allegro (eg. load_pcx() and 
   set_config_file()) can be used to read from them. Note that you can't 
   write to these special files, though: the fake file is read only. Also, 
   you must save your datafile uncompressed or with per-object compression 
   if you are planning on loading individual objects from it (otherwise 
   there will be an excessive amount of seeking when it is read). Finally, 
   be aware that the special Allegro object types aren't the same format as 
   the files you import the data from. When you import data like bitmaps or 
   samples into the grabber, they are converted into a special 
   Allegro-specific format, but the '#' marker file syntax reads the objects 
   as raw binary chunks. This means that if, for example, you want to use 
   load_pcx to read an image from a datafile, you should import it as a 
   binary block rather than as a BITMAP object.

int  pack_fclose(PACKFILE *f);
int  pack_fseek(PACKFILE *f, int offset);
int  pack_feof(PACKFILE *f);
int  pack_ferror(PACKFILE *f);
int  pack_getc(PACKFILE *f);
int  pack_putc(int c, PACKFILE *f);
int  pack_igetw(PACKFILE *f);
long pack_igetl(PACKFILE *f);
int  pack_iputw(int w, PACKFILE *f);
long pack_iputl(long l, PACKFILE *f);
int  pack_mgetw(PACKFILE *f);
long pack_mgetl(PACKFILE *f);
int  pack_mputw(int w, PACKFILE *f);
long pack_mputl(long l, PACKFILE *f);
long pack_fread(void *p, long n, PACKFILE *f);
long pack_fwrite(void *p, long n, PACKFILE *f);
char *pack_fgets(char *p, int max, PACKFILE *f);
int  pack_fputs(char *p, PACKFILE *f);

   These work like the equivalent stdio functions, except that pack_fread() 
   and pack_fwrite() take a single size parameter instead of that silly size 
   and num_elements system, seeking only supports forward movement relative 
   to the current position, and the pack_fgets() function does not include a 
   trailing carriage return in the returned string. The pack_i* and pack_m* 
   routines read and write 16 and 32 bit values using the Intel and Motorola 
   byte ordering systems (endianness) respectively. Note that seeking is 
   very slow when reading compressed files, and so should be avoided unless 
   you are sure that the file is not compressed.

PACKFILE *pack_fopen_chunk(PACKFILE *f, int pack);
   Opens a sub-chunk of a file. Chunks are primarily intended for use by the 
   datafile code, but they may also be useful for your own file routines. A 
   chunk provides a logical view of part of a file, which can be compressed 
   as an individual entity and will automatically insert and check length 
   counts to prevent reading past the end of the chunk. To write a chunk to 
   the file f, use the code:

      /* assumes f is a PACKFILE * which has been opened in write mode */
      f = pack_fopen_chunk(f, pack);
      write some data to f
      f = pack_fclose_chunk(f);

   The data written to the chunk will be prefixed with two length counts (32 
   bit, big-endian). For uncompressed chunks these will both be set to the 
   size of the data in the chunk. For compressed chunks (created by setting 
   the pack flag), the first length will be the raw size of the chunk, and 
   the second will be the negative size of the uncompressed data.

   To read the chunk, use the code:

      /* assumes f is a PACKFILE * which has been opened in read mode */
      f = pack_fopen_chunk(f, FALSE);
      read data from f
      f = pack_fclose_chunk(f);

   This sequence will read the length counts created when the chunk was 
   written, and automatically decompress the contents of the chunk if it 
   was compressed. The length will also be used to prevent reading past the 
   end of the chunk (Allegro will return EOF if you attempt this), and to 
   automatically skip past any unread chunk data when you call 
   pack_fclose_chunk().

   Chunks can be nested inside each other by making repeated calls to 
   pack_fopen_chunk(). When writing a file, the compression status is 
   inherited from the parent file, so you only need to set the pack flag if 
   the parent is not compressed but you want to pack the chunk data. If the 
   parent file is already open in packed mode, setting the pack flag will 
   result in data being compressed twice: once as it is written to the 
   chunk, and again as the chunk passes it on to the parent file.

PACKFILE *pack_fclose_chunk(PACKFILE *f);
   Closes a sub-chunk of a file, previously obtained by calling 
   pack_fopen_chunk().



===========================================
============ Datafile routines ============
===========================================

Datafiles are created by the grabber utility, and have a .dat extension. 
They can contain bitmaps, palettes, fonts, samples, MIDI music, FLI/FLC 
animations, and any other binary data that you import.

Warning: when using truecolor images, you should always set the graphics 
mode before loading any bitmap data! Otherwise the pixel format (RGB or BGR)
will not be known, so the file may be converted wrongly.

See the documentation for pack_fopen() for information about how to read 
directly from a specific datafile object.

DATAFILE *load_datafile(char *filename);
   Loads a datafile into memory, and returns a pointer to it, or NULL on 
   error. If the datafile has been encrypted, you must first use the 
   packfile_password() function to set the appropriate key. See grabber.txt 
   for more information. If the datafile contains truecolor graphics, you 
   must set the video mode or call set_color_conversion() before loading it.

void unload_datafile(DATAFILE *dat);
   Frees all the objects in a datafile.

DATAFILE *load_datafile_object(char *filename, char *objectname);
   Loads a specific object from a datafile. This won't work if you strip the 
   object names from the file, and it will be very slow if you save the file 
   with global compression. See grabber.txt for more information.

void unload_datafile_object(DATAFILE *dat);
   Frees an object previously loaded by load_datafile_object().

char *get_datafile_property(DATAFILE *dat, int type);
   Returns the specified property string for the object, or an empty string 
   if the property isn't present. See grabber.txt for more information.

void register_datafile_object(int id, void *(*load)(PACKFILE *f, long size),
                                      void (*destroy)(void *data));
   Used to add custom object types, specifying functions to load and destroy 
   objects of this type. See grabber.txt for more information.

void fixup_datafile(DATAFILE *data);
   If you are using compiled datafiles (produced by the dat2s utility) that 
   contain truecolor images, you must call this function once after your set 
   the video mode that you will be using, to convert the color values into 
   the appropriate format. It handles flipping between RGB and BGR formats, 
   and converting between different color depths whenever that can be done 
   without changing the size of the image (ie. changing between 15<->16 
   bit hicolor for both bitmaps and RLE sprites, and 24<->32 bit 
   truecolor for RLE sprites).

When you load a datafile, you will obtain a pointer to an array of DATAFILE 
structures:

typedef struct DATAFILE
{
   void *dat;     - pointer to the actual data
   int type;      - type of the data
   long size;     - size of the data in bytes
   void *prop;    - list of object properties
} DATAFILE;

The type field will be one of the values:
   DAT_FILE       - dat points to a nested datafile
   DAT_DATA       - dat points to a block of binary data
   DAT_FONT       - dat points to a font object
   DAT_SAMPLE     - dat points to a sample structure
   DAT_MIDI       - dat points to a MIDI file
   DAT_PATCH      - dat points to a GUS patch file
   DAT_FLI        - dat points to an FLI/FLC animation
   DAT_BITMAP     - dat points to a BITMAP structure
   DAT_RLE_SPRITE - dat points to a RLE_SPRITE structure
   DAT_C_SPRITE   - dat points to a linear compiled sprite
   DAT_XC_SPRITE  - dat points to a mode-X compiled sprite
   DAT_PALETTE    - dat points to an array of 256 RGB structures
   DAT_END        - special flag to mark the end of the data list

The grabber program can also produce a header file defining the index of 
each object within the file as a series of #defined constants, using the 
names you gave the objects in the grabber. So, for example, if you have made 
a datafile called foo.dat which contains a bitmap called THE_IMAGE, you 
could display it with the code fragment:

   #include "foo.h"

   DATAFILE *data = load_datafile("foo.dat");
   draw_sprite(screen, data[THE_IMAGE].dat, x, y);

If you are programming in C++ you will get an error because the dat field is 
a void pointer and draw_sprite() expects a BITMAP pointer. You can get 
around this with a cast, eg:

   draw_sprite(screen, (BITMAP *)data[THE_IMAGE].dat, x, y);




===================================================
============ Fixed point math routines ============
===================================================

Allegro provides some routines for working with fixed point numbers, and 
defines the type 'fixed' to be a signed 32 bit integer. The high word is 
used for the integer part and the low word for the fraction, giving a range 
of -32768 to 32767 and an accuracy of about four or five decimal places. 
Fixed point numbers can be assigned, compared, added, subtracted, negated 
and shifted (for multiplying or dividing by powers of two) using the normal 
integer operators, but you should take care to use the appropriate 
conversion routines when mixing fixed point with integer or floating point 
values. Writing 'fixed_point_1 + fixed_point_2' is ok, but 'fixed_point + 
integer' is not.

fixed itofix(int x);
   Converts an integer to fixed point. This is the same thing as x<<16.

int fixtoi(fixed x);
   Converts fixed point to integer, rounding as required. If you want to 
   avoid the rounding, use x>>16 instead.

fixed ftofix(float x);
   Converts a floating point value to fixed point.

float fixtof(fixed x);
   Converts fixed point to floating point.

fixed fmul(fixed x, fixed y);
   A fixed point value can be multiplied or divided by an integer with the 
   normal '*' and '/' operators. To multiply two fixed point values, though, 
   you must use this function.

   If an overflow or division by zero occurs, errno will be set and the 
   maximum possible value will be returned, but errno is not cleared if the 
   operation is successful. This means that if you are going to test for 
   overflow you should set errno=0 before calling fmul().

fixed fdiv(fixed x, fixed y);
   Fixed point division: see comments about fmul().

fixed fadd(fixed x, fixed y);
   Although fixed point numbers can be added with the normal '+' integer 
   operator, that doesn't provide any protection against overflow. If 
   overflow is a problem, you should use this function instead. It is slower 
   than using integer operators, but if an overflow occurs it will clamp the 
   result, rather than just letting it wrap, and set errno.

fixed fsub(fixed x, fixed y);
   Fixed point subtraction: see comments about fadd().

The fixed point square root, sin, cos, tan, inverse sin, and inverse cos 
functions are implemented using lookup tables, which are very fast but not 
particularly accurate. At the moment the inverse tan uses an iterative 
search on the tan table, so it is a lot slower than the others.

Angles are represented in a binary format with 256 equal to a full circle, 
64 being a right angle and so on. This has the advantage that a simple 
bitwise 'and' can be used to keep the angle within the range zero to a full 
circle, eliminating all those tiresome 'if (angle >= 360)' checks.

fixed fsin(fixed x);
   Lookup table sine.

fixed fcos(fixed x);
   Lookup table cosine.

fixed ftan(fixed x);
   Lookup table tangent.

fixed fasin(fixed x);
   Lookup table inverse sine.

fixed facos(fixed x);
   Lookup table inverse cosine.

fixed fatan(fixed x);
   Fixed point inverse tangent.

fixed fatan2(fixed y, fixed x);
   Fixed point version of the libc atan2() routine.

fixed fsqrt(fixed x);
   Fixed point square root.


If you are programming in C++ you can ignore all the above and use the fix 
class instead, which overloads a lot of operators to provide automatic 
conversion to and from integer and floating point values, and calls the 
above routines as they are required. You should not mix the fix class with 
the fixed typedef though, because the compiler will mistake the fixed values 
for regular integers and insert unnecessary conversions. For example, if x 
is an object of class fix, calling fsqrt(x) will return the wrong result. 
You should use the overloaded sqrt(x) or x.sqrt() instead.



==========================================
============ 3D math routines ============
==========================================

Allegro contains some 3d helper functions for manipulating vectors, 
constructing and using transformation matrices, and doing perspective 
projections from 3d space onto the screen. It is not, and never will be, a 
fully fledged 3d library (my goal is to supply generic support routines, not 
shrink-wrapped graphics code :-) but these functions may be useful for 
developing your own 3d code.

All the 3d math functions are available in two versions: one which uses 
fixed point arithmetic, and another which uses floating point. The syntax 
for these is identical, but the floating point functions and structures are 
postfixed with '_f', eg. the fixed point function cross_product() has a 
floating point equivalent cross_product_f(). If you are programming in C++, 
Allegro also overloads these functions for use with the 'fix' class.

3d transformations are accomplished by the use of a modelling matrix. This 
is a 4x4 array of numbers that can be multiplied with a 3d point to produce 
a different 3d point. By putting the right values into the matrix, it can be 
made to do various operations like translation, rotation, and scaling. The 
clever bit is that you can multiply two matrices together to produce a third 
matrix, and this will have the same effect on points as applying the 
original two matrices one after the other. For example, if you have one 
matrix that rotates a point and another that shifts it sideways, you can 
combine them to produce a matrix that will do the rotation and the shift in 
a single step. You can build up extremely complex transformations in this 
way, while only ever having to multiply each point by a single matrix.

Allegro actually cheats in the way it implements the matrix structure. 
Rotation and scaling of a 3d point can be done with a simple 3x3 matrix, but 
in order to translate it and project it onto the screen, the matrix must be 
extended to 4x4, and the point extended into 4d space by the addition of an 
extra coordinate, w=1. This is a bad thing in terms of efficiency, but 
fortunately an optimisation is possible. Given the 4x4 matrix:

   ( a, b, c, d )
   ( e, f, g, h )
   ( i, j, k, l )
   ( m, n, o, p )

a pattern can be observed in which parts of it do what. The top left 3x3 
grid implements rotation and scaling. The three values in the top right 
column (d, h, and l) implement translation, and as long as the matrix is 
only used for affine transformations, m, n and o will always be zero and p 
will always be 1. If you don't know what affine means, read Foley & Van 
Damme: basically it covers scaling, translation, and rotation, but not 
projection. Since Allegro uses a separate function for projection, the 
matrix functions only need to support affine transformations, which means 
that there is no need to store the bottom row of the matrix. Allegro 
implicitly assumes that it contains (0,0,0,1), and optimises the matrix 
manipulation functions accordingly.

Matrices are stored in the structures:

typedef struct MATRIX            - fixed point matrix structure
{
   fixed v[3][3];                - 3x3 scaling and rotation component
   fixed t[3];                   - x/y/z translation component
} MATRIX;

typedef struct MATRIX_f          - floating point matrix structure
{
   float v[3][3];                - 3x3 scaling and rotation component
   float t[3];                   - x/y/z translation component
} MATRIX_f

extern MATRIX identity_matrix;
extern MATRIX_f identity_matrix_f;
   Global variables containing the 'do nothing' identity matrix. Multiplying 
   by the identity matrix has no effect.

void get_translation_matrix(MATRIX *m, fixed x, fixed y, fixed z);
void get_translation_matrix_f(MATRIX_f *m, float x, float y, float z);
   Constructs a translation matrix, storing it in m. When applied to the 
   point (px, py, pz), this matrix will produce the point (px+x, py+y, 
   pz+z). In other words, it moves things sideways.

void get_scaling_matrix(MATRIX *m, fixed x, fixed y, fixed z);
void get_scaling_matrix_f(MATRIX_f *m, float x, float y, float z);
   Constructs a scaling matrix, storing it in m. When applied to the point 
   (px, py, pz), this matrix will produce the point (px*x, py*y, pz*z). In 
   other words, it stretches or shrinks things.

void get_x_rotate_matrix(MATRIX *m, fixed r);
void get_x_rotate_matrix_f(MATRIX_f *m, float r);
   Construct X axis rotation matrices, storing them in m. When applied to a 
   point, these matrices will rotate it about the X axis by the specified 
   angle (given in binary, 256 degrees to a circle format).

void get_y_rotate_matrix(MATRIX *m, fixed r);
void get_y_rotate_matrix_f(MATRIX_f *m, float r);
   Construct Y axis rotation matrices, storing them in m. When applied to a 
   point, these matrices will rotate it about the Y axis by the specified 
   angle (given in binary, 256 degrees to a circle format).

void get_z_rotate_matrix(MATRIX *m, fixed r);
void get_z_rotate_matrix_f(MATRIX_f *m, float r);
   Construct Z axis rotation matrices, storing them in m. When applied to a 
   point, these matrices will rotate it about the Z axis by the specified 
   angle (given in binary, 256 degrees to a circle format).

void get_rotation_matrix(MATRIX *m, fixed x, fixed y, fixed z);
void get_rotation_matrix_f(MATRIX_f *m, float x, float y, float z);
   Constructs a transformation matrix which will rotate points around all 
   three axis by the specified amounts (given in binary, 256 degrees to a 
   circle format).

void get_align_matrix(MATRIX *m, fixed xfront, yfront, zfront, 
                                 fixed xup, fixed yup, fixed zup);
   Rotates a matrix so that it is aligned along the specified coordinate 
   vectors (they need not be normalized or perpendicular, but the up and 
   front must not be equal). A front vector of 1,0,0 and up vector of 0,1,0 
   will return the identity matrix.

void get_align_matrix_f(MATRIX *m, float xfront, yfront, zfront, 
                                   float xup, yup, zup);
   Floating point version of get_align_matrix().

void get_vector_rotation_matrix(MATRIX *m, fixed x, y, z, fixed a);
void get_vector_rotation_matrix_f(MATRIX_f *m, float x, y, z, float a);
   Constructs a transformation matrix which will rotate points around the 
   specified x,y,z vector by the specified angle (given in binary, 256 
   degrees to a circle format).

void get_transformation_matrix(MATRIX *m, fixed scale,
                               fixed xrot, yrot, zrot, x, y, z);
   Constructs a transformation matrix which will rotate points around all 
   three axis by the specified amounts (given in binary, 256 degrees to a 
   circle format), scale the result by the specified amount (pass 1 for no 
   change of scale), and then translate to the requested x, y, z position.

void get_transformation_matrix_f(MATRIX_f *m, float scale,
                                 float xrot, yrot, zrot, x, y, z);
   Floating point version of get_transformation_matrix().

void get_camera_matrix(MATRIX *m, fixed x, y, z, xfront, yfront, zfront,
                       fixed xup, yup, zup, fov, aspect);
   Constructs a camera matrix for translating world-space objects into a 
   normalised view space, ready for the perspective projection. The x, y, 
   and z parameters specify the camera position, xfront, yfront, and zfront 
   are the 'in front' vector specifying which way the camera is facing (this 
   can be any length: normalisation is not required), and xup, yup, and zup 
   are the 'up' direction vector. The fov parameter specifies the field of 
   view (ie. width of the camera focus) in binary, 256 degrees to the circle 
   format. For typical projections, a field of view in the region 32-48 will 
   work well. Finally, the aspect ratio is used to scale the Y dimensions of 
   the image relative to the X axis, so you can use it to adjust the 
   proportions of the output image (set it to 1 for no scaling).

void get_camera_matrix_f(MATRIX_f *m, float x, y, z, xfront, yfront, zfront,
                         float xup, yup, zup, fov, aspect);
   Floating point version of get_camera_matrix().

void qtranslate_matrix(MATRIX *m, fixed x, fixed y, fixed z);
void qtranslate_matrix_f(MATRIX_f *m, float x, float y, float z);
   Optimised routine for translating an already generated matrix: this 
   simply adds in the translation offset, so there is no need to build two 
   temporary matrices and then multiply them together.

void qscale_matrix(MATRIX *m, fixed scale);
void qscale_matrix_f(MATRIX_f *m, float scale);
   Optimised routine for scaling an already generated matrix: this simply 
   adds in the scale factor, so there is no need to build two temporary 
   matrices and then multiply them together.

void matrix_mul(MATRIX *m1, MATRIX *m2, MATRIX *out);
void matrix_mul_f(MATRIX_f *m1, MATRIX_f *m2, MATRIX_f *out);
   Multiplies two matrices, storing the result in out (this must be 
   different from the two input matrices). The resulting matrix will have 
   the same effect as the combination of m1 and m2, ie. when applied to a 
   point p, (p * out) = ((p * m1) * m2). Any number of transformations can 
   be concatenated in this way. Note that matrix multiplication is not 
   commutative, ie. matrix_mul(m1, m2) != matrix_mul(m2, m1).

fixed vector_length(fixed x, fixed y, fixed z);
float vector_length_f(float x, float y, float z);
   Calculates the length of the vector (x, y, z), using that good 'ole 
   Pythagoras theorem.

void normalize_vector(fixed *x, fixed *y, fixed *z);
void normalize_vector_f(float *x, float *y, float *z);
   Converts the vector (*x, *y, *z) to a unit vector. This points in the 
   same direction as the original vector, but has a length of one.

fixed dot_product(fixed x1, y1, z1, x2, y2, z2);
float dot_product_f(float x1, y1, z1, x2, y2, z2);
   Calculates the dot product (x1, y1, z1) . (x2, y2, z2), returning the 
   result.

void cross_product(fixed x1, y1, z1, x2, y2, z2, *xout, *yout, *zout);
void cross_product_f(float x1, y1, z1, x2, y2, z2, *xout, *yout, *zout);
   Calculates the cross product (x1, y1, z1) x (x2, y2, z2), storing the 
   result in (*xout, *yout, *zout). The cross product is perpendicular to 
   both of the input vectors, so it can be used to generate polygon normals.

fixed polygon_z_normal(V3D *v1, V3D *v2, V3D *v3);
float polygon_z_normal_f(V3D_f *v1, V3D_f *v2, V3D_f *v3);
   Finds the Z component of the normal vector to the specified three 
   vertices (which must be part of a convex polygon). This is used mainly in 
   back-face culling. The back-faces of closed polyhedra are never visible 
   to the viewer, therefore they never need to be drawn. This can cull on 
   average half the polygons from a scene. If the normal is negative the 
   polygon can safely be culled. If it is zero, the polygon is perpendicular 
   to the screen.

void apply_matrix(MATRIX *m, fixed x, y, z, *xout, *yout, *zout);
void apply_matrix_f(MATRIX_f *m, float x, y, z, *xout, *yout, *zout);
   Multiplies the point (x, y, z) by the transformation matrix m, storing 
   the result in (*xout, *yout, *zout).

void set_projection_viewport(int x, int y, int w, int h);
   Sets the viewport used to scale the output of the persp_project() 
   function. Pass the dimensions of the screen area you want to draw onto, 
   which will typically be 0, 0, SCREEN_W, and SCREEN_H.

void persp_project(fixed x, y, z, *xout, *yout);
void persp_project_f(float x, y, z, *xout, *yout);
   Projects the 3d point (x, y, z) into 2d screen space, storing the result 
   in (*xout, *yout) and using the scaling parameters previously set by 
   calling set_projection_viewport(). This function projects from the 
   normalized viewing pyramid, which has a camera at the origin and facing 
   along the positive z axis. The x axis runs left/right, y runs up/down, 
   and z increases with depth into the screen. The camera has a 90 degree 
   field of view, ie. points on the planes x=z and -x=z will map onto the 
   left and right edges of the screen, and the planes y=z and -y=z map to 
   the top and bottom of the screen. If you want a different field of view 
   or camera location, you should transform all your objects with an 
   appropriate viewing matrix, eg. to get the effect of panning the camera 
   10 degrees to the left, rotate all your objects 10 degrees to the right.



======================================
============ GUI routines ============
======================================

Allegro contains an object-oriented dialog manager, which was originally 
based on the Atari GEM system (form_do(), objc_draw(), etc: old ST 
programmers will know what I'm talking about :-) You can use the GUI as-is 
to knock out simple interfaces for things like the test program and grabber 
utility, or you can use it as a basis for more complicated systems of your 
own. Allegro lets you define your own object types by writing new dialog 
procedures, so you can take complete control over the visual aspects of the 
interface while still using Allegro to handle input from the mouse, 
keyboard, joystick, etc.

A GUI dialog is stored as an array of DIALOG objects, each one containing 
the fields:

typedef struct DIALOG
{
   int (*proc)(int, DIALOG *, int); - dialog procedure (message handler)
   int x, y, w, h;                  - position and size of the object
   int fg, bg;                      - foreground and background colors
   int key;                         - ASCII keyboard shortcut
   int flags;                       - flags about the status of the object
   int d1, d2;                      - whatever you want to use them for
   void *dp, *dp2, *dp3;            - pointers to more object-specific data
} DIALOG;

The array should end with an object which has the proc pointer set to NULL.

The flags field may contain any combination of the bit flags:
   D_EXIT          - this object should close the dialog when it is clicked
   D_SELECTED      - this object is selected
   D_GOTFOCUS      - this object has got the input focus
   D_GOTMOUSE      - the mouse is currently on top of this object
   D_HIDDEN        - this object is hidden and inactive
   D_DISABLED      - this object is greyed-out and inactive
   D_INTERNAL      - don't use this! It is for internal use by the library...
   D_USER          - any powers of two above this are free for your own use

Each object is controlled by a dialog procedure, which is stored in the proc 
pointer. This will be called by the dialog manager whenever any action 
concerning the object is required, or you can call it directly with the 
SEND_MESSAGE macro. The dialog procedure should follow the form:

   int foo(int msg, DIALOG *d, int c);

It will be passed a flag (msg) indicating what action it should perform, a 
pointer to the object concerned (d), and if msg is MSG_CHAR or MSG_XCHAR, 
the key that was pressed (c). Note that d is a pointer to a specific object, 
and not to the entire dialog.

The dialog procedure should return one of the values:
   D_O_K          - normal return status
   D_CLOSE        - tells the dialog manager to close the dialog
   D_REDRAW       - tells the dialog manager to redraw the entire dialog
   D_WANTFOCUS    - requests that the input focus be given to this object
   D_USED_CHAR    - MSG_CHAR and MSG_XCHAR return this if they used the key

Dialog procedures may be called with any of the messages:

MSG_START:
   Tells the object to initialise itself. The dialog manager sends this to 
   all the objects in a dialog just before it displays the dialog.

MSG_END:
   Sent to all objects when closing a dialog, allowing them to perform 
   whatever cleanup operations they require.

MSG_DRAW:
   Tells the object to draw itself onto the screen. The mouse pointer will 
   be turned off when this message is sent, so the drawing code does not 
   need to worry about it.

MSG_CLICK:
   Informs the object that a mouse button has been clicked while the mouse 
   was on top of the object. Typically an object will perform its own mouse 
   tracking as long as the button is held down, and only return from this 
   message handler when it is released.

MSG_DCLICK:
   Sent when the user double-clicks on an object. A MSG_CLICK will be sent 
   when the button is first pressed, then MSG_DCLICK if it is released and 
   pressed again within a short space of time.

MSG_KEY:
   Sent when the keyboard shortcut for the object is pressed, or if enter, 
   space, or a joystick button is pressed while it has the input focus.

MSG_CHAR:
   When a key is pressed, this message is sent to the object that has the 
   input focus. If the object deals with the keypress it should return 
   D_USED_CHAR, otherwise it should return D_O_K to allow the default 
   keyboard interface to operate.

MSG_XCHAR:
   When a key is pressed, Allegro will send a MSG_CHAR to the object with 
   the input focus. If this object doesn't process the key (ie. it returns 
   D_O_K rather than D_USED_CHAR), the dialog manager will look for an 
   object with a matching keyboard shortcut in the key field, and send it a 
   MSG_KEY. If this fails, it broadcasts a MSG_XCHAR to all objects in the 
   dialog, allowing them to respond to special keypresses even when they 
   don't have the input focus. Normally you should ignore this message 
   (return D_O_K rather than D_USED_CHAR), in which case Allegro will 
   perform default actions such as moving the focus in response to the arrow 
   keys and closing the dialog if ESC is pressed.

MSG_WANTFOCUS:
   Queries whether an object is willing to accept the input focus. It should 
   return D_WANTFOCUS if it does, or D_O_K if it isn't interested in getting 
   user input.

MSG_GOTFOCUS:
MSG_LOSTFOCUS:
   Sent whenever an object gains or loses the input focus. These messages 
   will always be followed by a MSG_DRAW, to let objects display themselves 
   differently when they have the input focus. If you return D_WANTFOCUS in 
   response to a MSG_LOSTFOCUS event, this will prevent your object from 
   losing the focus when the mouse moves off it onto the screen background 
   or some inert object, so it will only lose the input focus when some 
   other object is ready to take over the focus (this trick is used by the 
   d_edit_proc() object).

MSG_GOTMOUSE:
MSG_LOSTMOUSE:
   Sent when the mouse moves on top of or away from an object. Unlike the 
   focus messages, these are not followed by a MSG_DRAW, so if the object is 
   displayed differently when the mouse is on top of it, it is responsible 
   for redrawing itself in response to these messages.

MSG_IDLE:
   Sent whenever the dialog manager has nothing better to do.

MSG_RADIO:
   Sent by radio button objects to deselect other buttons in the same group 
   when they are clicked. The group number is passed in the c message 
   parameter.

MSG_USER:
   The first free message value. Any numbers from here on (MSG_USER, 
   MSG_USER+1, MSG_USER+2, ... MSG_USER+n) are free to use for whatever you 
   like.

Allegro provides several standard dialog procedures. You can use these as 
they are to provide simple user interface objects, or you can call them from 
within your own dialog procedures, resulting in a kind of OOP inheritance. 
For instance, you could make an object which calls d_button_proc to draw 
itself, but handles the click message in a different way, or an object which 
calls d_button_proc for everything except drawing itself, so it would behave 
like a normal button but could look completely different.

int d_clear_proc(int msg, DIALOG *d, int c);
   This just clears the screen when it is drawn. Useful as the first object 
   in a dialog.

int d_box_proc(int msg, DIALOG *d, int c);
int d_shadow_box_proc(int msg, DIALOG *d, int c);
   These draw boxes onto the screen, with or without a shadow.

int d_bitmap_proc(int msg, DIALOG *d, int c);
   This draws a bitmap onto the screen, which should be pointed to by the 
   dp field.

int d_text_proc(int msg, DIALOG *d, int c);
int d_ctext_proc(int msg, DIALOG *d, int c);
   These draw text onto the screen. The dp field should point to the string 
   to display. d_ctext_proc() centres the string around the x coordinate. 
   Any '&' characters in the string will be replaced with lines underneath 
   the following character, for displaying keyboard shortcuts (as in MS 
   Windows). To display a single ampersand, put "&&". To use draw the text 
   in something other than the default font, set the dp2 field to point to 
   your custom font data.

int d_button_proc(int msg, DIALOG *d, int c);
   A button object (the dp field points to the text string). This object can 
   be selected by clicking on it with the mouse or by pressing its keyboard 
   shortcut. If the D_EXIT flag is set, selecting it will close the dialog, 
   otherwise it will toggle on and off. Like d_text_proc(), ampersands can 
   be used to display the keyboard shortcut of the button.

int d_check_proc(int msg, DIALOG *d, int c);
   This is an example of how you can derive objects from other objects. Most 
   of the functionality comes from d_button_proc(), but it displays itself 
   as a check box.

int d_radio_proc(int msg, DIALOG *d, int c);
   A radio button object. A dialog can contain any number of radio button 
   groups: selecting a radio button causes other buttons within the same 
   group to be deselected. The dp field points to the text string, d1 
   specifies the group number, and d2 is the button style (0=circle, 
   1=square).

int d_icon_proc(int msg, DIALOG *d, int c);
   A bitmap button. The fg color is used for the dotted line showing focus, 
   and the bg color for the shadow used to fill in the top and left sides of 
   the button when "pressed". d1 is the "push depth", ie. the number of 
   pixels the icon will be shifted to the right and down when selected 
   (default 2) if there is no "selected" image. d2 is the distance by which 
   the dotted line showing focus is indented (default 2). dp points to a 
   bitmap for the icon, while dp2 and dp3 are the selected and disabled 
   images respectively (optional, may be NULL).

int d_keyboard_proc(int msg, DIALOG *d, int c);
   This is an invisible object for implementing keyboard shortcuts. You can 
   put an ASCII code in the key field of the dialog object (a character such 
   as 'a' to respond to a simple keypress, or a number 1-26 to respond to a 
   control key a-z), or you can put a keyboard scancode in the d1 and/or d2 
   fields. When one of these keys is pressed, the object will call the 
   function pointed to by dp. This should return an int, which will be 
   passed back to the dialog manager, so it can return D_O_K, D_REDRAW, 
   D_CLOSE, etc.

int d_edit_proc(int msg, DIALOG *d, int c);
   An editable text object (the dp field points to the string). When it has 
   the input focus (obtained by clicking on it with the mouse), text can be 
   typed into this object. The d1 field specifies the maximum number of 
   characters that it will accept, and d2 is the text cursor position within 
   the string.

int d_list_proc(int msg, DIALOG *d, int c);
   A list box object. This will allow the user to scroll through a list of 
   items and to select one by clicking or with the arrow keys. If the D_EXIT 
   flag is set, double clicking on a list item will close the dialog. The 
   index of the selected item is held in the d1 field, and d2 is used to 
   store how far it has scrolled through the list. The dp field points to a 
   function which will be called to obtain information about the contents of 
   the list. This should follow the form:

      char *foobar(int index, int *list_size);

   If index is zero or positive, the function should return a pointer to the 
   string which is to be displayed at position index in the list. If index 
   is negative, it should return NULL and list_size should be set to the 
   number of items in the list. 

   To create a multiple selection listbox, set the dp2 field to an array of 
   byte flags indicating the selection state of each list item (non-zero for 
   selected entries). This table must be at least as big as the number of 
   objects in the list!

int d_textbox_proc(int msg, DIALOG *d, int c);
   A text box object. The dp field points to the text which is to be 
   displayed in the box. If the text is long, there will be a vertical 
   scrollbar on the right hand side of the object which can be used to 
   scroll through the text. The default is to print the text with word 
   wrapping, but if the D_SELECTED flag is set, the text will be printed 
   with character wrapping. The d1 field is used internally to store the 
   number of lines of text, and d2 is used to store how far it has scrolled 
   through the text.

int d_slider_proc(int msg, DIALOG *d, int c);
   A slider control object. This object holds a value in d2, in the range 
   from 0 to d1. It will display as a vertical slider if h is greater than 
   or equal to w, otherwise it will display as a horizontal slider. The dp 
   field can contain an optional bitmap to use for the slider handle, and 
   dp2 can contain an optional callback function, which is called each time 
   d2 changes. The callback function should have the following prototype:

      int function(void *dp3, int d2);

   The d_slider_proc object will return the value of the callback function.

int d_menu_proc(int msg, DIALOG *d, int c);
   This object is a menu bar which will drop down child menus when it is 
   clicked or if an alt+key corresponding to one of the shortcuts in the 
   menu is pressed. It ignores a lot of the fields in the dialog structure, 
   in particular the color is taken from the gui_*_color variables, and the 
   width and height are calculated automatically. The dp field points to an 
   array of menu structures: see do_menu() for more information. The top 
   level menu will be displayed as a horizontal bar, but when child menus 
   drop down from it they will be in the normal vertical format used by 
   do_menu(). When a menu item is selected, the return value from the menu 
   callback function is passed back to the dialog manager, so your callbacks 
   should return D_O_K, D_REDRAW, or D_CLOSE.

The behaviour of the dialog manager can be controlled by the variables:

extern int gui_mouse_focus;
   If set, the input focus follows the mouse pointer around the dialog, 
   otherwise a click is required to move it.

extern int gui_fg_color, gui_bg_color;
   The foreground and background colors for the standard dialogs (alerts, 
   menus, and the file selector). They default to 255 and 0.

extern int gui_mg_color;
   The color used for displaying greyed-out dialog objects (with the 
   D_DISABLED flag set). Defaults to 8.

extern int gui_font_baseline;
   If set to a non-zero value, adjusts the keyboard shortcut underscores to 
   account for the height of the descenders in your font.

extern int (*gui_mouse_x)();
extern int (*gui_mouse_y)();
extern int (*gui_mouse_b)();
   Hook functions, used by the GUI routines whenever they need to access the 
   mouse state. By default these just return copies of the mouse_x, mouse_y, 
   and mouse_b variables, but they could be used to offset or scale the 
   mouse position, or read input from a different source entirely.

You can change the global 'font' pointer to make the GUI objects use 
something other than the standard 8x8 font. The standard dialog procedures, 
menus, and alert boxes, will work with fonts of any size, but the 
file_select() and gfx_mode_select() dialogs will look wrong with anything 
other than 8x8 fonts.

int gui_textout(BITMAP *bmp, unsigned char *s, int x, y, color, centre);
   Helper function for use by the GUI routines. Draws a text string onto the 
   screen, interpreting the '&' character as an underbar for displaying 
   keyboard shortcuts. Returns the width of the output string in pixels.

int gui_strlen(unsigned char *s);
   Helper function for use by the GUI routines. Returns the length of a 
   string in pixels, ignoring '&' characters.

void centre_dialog(DIALOG *dialog);
   Moves an array of dialog objects so that it is centered in the screen.

void set_dialog_color(DIALOG *dialog, int fg, int bg);
   Sets the foreground and background colors of an array of dialog objects.

int find_dialog_focus(DIALOG *dialog);
   Searches the dialog for the object which has the input focus, returning 
   an index or -1 if the focus is not set. This is useful if you are calling 
   do_dialog() several times in a row and want to leave the focus in the 
   same place it was when the dialog was last displayed, as you can call 
   do_dialog(dlg, find_dialog_focus(dlg));

int dialog_message(DIALOG *dialog, int msg, int c, int *obj);
   Sends a message to all the objects in an array. If any of the dialog 
   procedures return values other than D_O_K, it returns the value and sets 
   obj to the index of the object which produced it.

int broadcast_dialog_message(int msg, int c);
   Broadcasts a message to all the objects in the active dialog. If any of 
   the dialog procedures return values other than D_O_K, it returns that 
   value.

int do_dialog(DIALOG *dialog, int focus_obj);
   The basic dialog manager function. This displays a dialog (an array of 
   dialog objects, terminated by one with a NULL dialog procedure), and sets 
   the input focus to the focus_obj (-1 if you don't want anything to have 
   the focus). It interprets user input and dispatches messages as they are 
   required, until one of the dialog procedures tells it to close the 
   dialog, at which point it returns the index of the object that caused it 
   to exit.

int popup_dialog(DIALOG *dialog, int focus_obj);
   Like do_dialog(), but it stores the data on the screen before drawing the 
   dialog and restores it when the dialog is closed. The screen area to be 
   stored is calculated from the dimensions of the first object in the 
   dialog, so all the other objects should lie within this one.

DIALOG_PLAYER *init_dialog(DIALOG *dialog, int focus_obj);
   This function provides lower level access to the same functionality as 
   do_dialog(), but allows you to combine a dialog box with your own program 
   control structures. It initialises a dialog, returning a pointer to a 
   player object that can be used with update_dialog() and 
   shutdown_dialog(). With these functions, you could implement your own 
   version of do_dialog() with the lines:

      void *player = init_dialog(dialog, focus_obj);

      while (update_dialog(player))
         ;

      return shutdown_dialog(player);

int update_dialog(DIALOG_PLAYER *player);
   Updates the status of a dialog object returned by init_dialog(). Returns 
   TRUE if the dialog is still active, or FALSE if it has terminated. Upon a 
   return value of FALSE, it is up to you whether to call shutdown_dialog() 
   or to continue execution. The object that requested the exit can be 
   determined from the player->obj field.

int shutdown_dialog(DIALOG_PLAYER *player);
   Destroys a dialog player object returned by init_dialog(), returning the 
   object that caused it to exit (this is the same as the return value from 
   do_dialog()).

extern DIALOG *active_dialog;
   Global pointer to the most recent activated dialog. This may be useful if 
   an object needs to iterate through a list of all its siblings.

Popup or pulldown menus are created as an array of the structures:

typedef struct MENU
{
   char *text;                   - the text to display for the menu item
   int (*proc)();                - called when the menu item is clicked
   struct MENU *child;           - nested child menu
   int flags;                    - disabled or checked state
   void *dp;                     - pointer to any data you need
} MENU;

Each menu item contains a text string. This can use the '&' character to 
indicate keyboard shortcuts, or can be an zero-length string to display the 
item as a non-selectable splitter bar. If the string contains a "\t" tab 
character, any text after this will be right-justified, eg. for displaying 
keyboard shortcut information. The proc pointer is a function which will be 
called when the menu item is selected, and child points to another menu, 
allowing you to create nested menus. Both proc and child may be NULL. The 
proc function returns an integer which is ignored if the menu was brought up 
by calling do_menu(), but which is passed back to the dialog manager if it 
was created by a d_menu_proc() object. The array of menu items is terminated 
by an entry with a NULL text pointer.

Menu items can be disabled (greyed-out) by setting the D_DISABLED bit in the 
flags field, and a check mark can be displayed next to them by setting the 
D_SELECTED bit. With the default alignment and font this will usually 
overlap the menu text, so if you are going to use checked menu items it 
would be a good idea to prefix all your options with a space or two, to 
ensure there is room for the check.

int do_menu(MENU *menu, int x, int y)
   Displays and animates a popup menu at the specified screen coordinates 
   (these will be adjusted if the menu does not entirely fit on the screen). 
   Returns the index of the menu item that was selected, or -1 if the menu 
   was cancelled. Note that the return value cannot indicate selection from 
   child menus, so you will have to use the callback functions if you want 
   multi-level menus.

extern MENU *active_menu;
   When a menu callback procedure is triggered, this will be set to the menu 
   item that was selected, so your routine can determine where it was called 
   from.

int alert(char *s1, *s2, *s3, char *b1, *b2, int c1, c2);
   Displays a popup alert box, containing three lines of text (s1-s3), and 
   with either one or two buttons. The text for these buttons is passed in 
   b1 and b2 (b2 may be NULL), and the keyboard shortcuts in c1 and c2. 
   Returns 1 or 2 depending on which button was clicked. If the alert is 
   dismissed by pressing ESC when ESC is not one of the keyboard shortcuts, 
   it treats it as a click on the second button (this is consistent with the 
   common "Ok", "Cancel" alert).

int alert3(char *s1, *s2, *s3, char *b1, *b2, *b3, int c1, c2, c3);
   Like alert(), but with three buttons. Returns 1, 2, or 3.

int file_select(char *message, char *path, char *ext);
   Displays the Allegro file selector, with the message as caption. The path 
   parameter contains the initial filename to display (this can be used to 
   set the starting directory, or to provide a default filename for a 
   save-as operation). The user selection is returned by altering path, so 
   it should have room for at least 80 characters. The list of files is 
   filtered according to the file extensions in ext. Passing NULL includes 
   all files, "PCX;BMP" includes only files with .PCX or .BMP extensions. 
   Returns zero if it was closed with the Cancel button, non-zero if it was 
   OK'd.

int gfx_mode_select(int *card, int *w, int *h);
   Displays the Allegro graphics mode selection dialog, which allows the 
   user to select a screen mode and graphics card. Stores the selection in 
   the three variables, and returns zero if it was closed with the Cancel 
   button or non-zero if it was OK'd.

int gfx_mode_select_ex(int *card, int *w, int *h, int *color_depth);
   Extended version of the graphics mode selection dialog, which allows the 
   user to select the color depth as well as the resolution and hardware 
   driver. This version of the function reads the initial values from the 
   parameters when it activates, so you can specify the default values.



=======================================================
============ Reducing your executable size ============
=======================================================

Some people complain that Allegro produces very large executables. This is 
certainly true: a simple "hello world" program will be about 200k. But don't 
worry, it is a relatively fixed overhead and won't increase as your program 
gets larger. As George Foot so succinctly put it, anyone who is concerned 
about the ratio between library and program code should just get to work and 
write more program code to catch up :-)

Having said that, there are several things you can do to make your programs 
smaller:

For starters, read the djgpp FAQ section 8.15, and take note of the -s 
switch and DJP executable compressor. And don't forget to compile your 
program with optimisation enabled!

If your program is only going to run in a limited number of graphics modes, 
you can specify which graphics drivers you would like to include with the 
code:

   BEGIN_GFX_DRIVER_LIST
      driver1
      driver2
      etc...
   END_GFX_DRIVER_LIST

where the driver names are any of the defines:

   GFX_DRIVER_VBEAF
   GFX_DRIVER_VGA
   GFX_DRIVER_MODEX
   GFX_DRIVER_VESA3
   GFX_DRIVER_VESA2L
   GFX_DRIVER_VESA2B
   GFX_DRIVER_XTENDED
   GFX_DRIVER_VESA1

This construct must be included in only one of your C source files (_not_ a 
header file!). The ordering of the names is important, because the 
autodetection routine works down from the top of the list until it finds the 
first driver that is able to support the requested mode. I suggest you stick 
to the default ordering given above, and simply delete whatever entries you 
aren't going to use.

As a rule, removing the MODEX and XTENDED drivers will save a fair amount of 
space, as will removing the VESA and VBE/AF drivers if you are only going to 
use standard VGA modes.

If your program doesn't need to use all the possible color depths, you can 
specify which pixel formats you want to support with the code:

   BEGIN_COLOR_DEPTH_LIST
      depth1
      depth2
      etc...
   END_COLOR_DEPTH_LIST

where the color depth names are any of the defines:

   COLOR_DEPTH_8
   COLOR_DEPTH_15
   COLOR_DEPTH_16
   COLOR_DEPTH_24
   COLOR_DEPTH_32

Removing any of the color depths will save quite a bit of space, with the 
exception of the 15 and 16 bit modes: these share a great deal of code, so 
if you are including one of them, there is no reason not to use both. Be 
warned that if you try to use a color depth which isn't in this list, your 
program will crash horribly!

In the same way as the above, you can specify which sound drivers you want 
to support with the code:

   BEGIN_DIGI_DRIVER_LIST
      driver1
      driver2
      etc...
   END_DIGI_DRIVER_LIST

using the digital sound driver defines:

   DIGI_DRIVER_SOUNDSCAPE
   DIGI_DRIVER_AUDIODRIVE
   DIGI_DRIVER_SB

and for the MIDI music:

   BEGIN_MIDI_DRIVER_LIST
      driver1
      driver2
      etc...
   END_MIDI_DRIVER_LIST

using the MIDI driver defines:

   MIDI_DRIVER_AWE32
   MIDI_DRIVER_DIGMID
   MIDI_DRIVER_ADLIB
   MIDI_DRIVER_MPU
   MIDI_DRIVER_SB_OUT

If you are going to use either of these constructs, you must use both. If 
you only want to include digital sound drivers, simply write 
DECLARE_MIDI_DRIVER_LIST() to prevent the inclusion of any music drivers at 
all.

Likewise for the joystick drivers, you can declare an inclusion list:

   BEGIN_JOYSTICK_DRIVER_LIST
      driver1
      driver2
      etc...
   END_JOYSTICK_DRIVER_LIST

using the joystick driver defines:

   JOYSTICK_DRIVER_STANDARD
   JOYSTICK_DRIVER_SIDEWINDER
   JOYSTICK_DRIVER_GAMEPAD_PRO
   JOYSTICK_DRIVER_SNESPAD
   JOYSTICK_DRIVER_WINGWARRIOR

The standard driver includes support for the dual joysticks, increased 
numbers of buttons, Flightstick Pro, and Wingman Extreme, because these are 
all quite minor variations on the basic code.

If you are _really_ serious about this size, thing, have a look at the top 
of allegro.h and you will see the lines:

   #define ALLEGRO_COLOR16
   #define ALLEGRO_COLOR24
   #define ALLEGRO_COLOR32

If you comment out any of these definitions and then rebuild the library, 
you will get a version without any support for the absent color depths, 
which will be even smaller than using the DECLARE_COLOR_DEPTH_LIST() macro. 
Removing the ALLEGRO_COLOR16 define will get rid of the support for both 15 
and 16 bit hicolor modes, since these share a lot of the same code.

If you are distributing a copy of the setup program along with your game, 
you may be able to get a dramatic size reduction by merging the setup code 
into your main program, so that only one copy of the Allegro routines will 
need to be linked. To do this, rename the main() function from setup.c to 
something like setup_main(), link setup.c and setupdat.s into your program, 
and add a switch (eg. -setup command line option) that will call 
setup_main() instead of your main program code. After compressing the 
executable, this will probably save you about 200k compared to having two 
separate programs for the setup and the game itself.



===================================
============ Debugging ============
===================================

By default Allegro builds with maximum optimisation, no debug information, 
and the -fomit-frame-pointer option. This produces the most efficient code, 
but makes it very hard to debug your program if it is crashing inside an 
Allegro function. There are several variables to control the code generation 
options, which may be useful while debugging or if you want to modify 
Allegro. These can either be set as environment variables from the command 
prompt (eg. "set DEBUGMODE=1") or passed to make on the command line (eg. 
"make install DEBUGMODE=1"). To build a new version, run "make clean" to get 
rid of the previous build, set the appropriate variable, and then run "make 
install" to build a version with the new options. Don't forget to backup 
your previous liballeg.a first, or to recompile with the normal options when 
you are done!

   DEBUGMODE
   Creates a version with no optimisation and full debug information, so you 
   can get proper symify tracebacks and debug inside the Allegro routines. 
   This library will be very slow!

   PROFILEMODE
   Creates an optimised version that includes profiling information (using 
   the -pg option).

   SYMBOLMODE
   Builds a normal optimised version, but doesn't strip out the symbol table 
   at the link stage. This is really only useful if you are modifying 
   Allegro itself, and want to run symify on the test programs.

   PGCC
   Uses optimisation options suitable for the Pentium optimising compiler, 
   rather than the stock djgpp version of gcc.

   WARNMODE
   Enables stricter compiler warnings. I strongly recommend that all Allegro 
   developers enable this option: I won't accept any code that doesn't 
   compile cleanly with it.

If you only want to recompile a specific test program or utility, you can 
specify it as an argument to make, eg. "make demo" or "make grabber". The 
makefile also provides some special pseudo-targets:

   'all'
   The normal build process. Compiles the library, all the test and example 
   programs, and the documentation files, and then copies allegro.h, 
   liballeg.a, and allegro.inf into your djgpp directories.

   'lib'
   Builds a new version of liballeg.a, without installing it into your 
   djgpp/lib/ directory.

   'docs'
   Converts the documentation files from the ._tx sources.

   'install'
   Updates the versions of allegro.h, liballeg.a, and allegro.inf in your 
   djgpp directories, recompiling them as necessary.

   'compress'
   Uses the DJP or UPX executable compressors (whichever you have installed) 
   to compress the example executables and utility programs, which can 
   recover a significant amount of disk space.

   'clean'
   Removes all the temporary generated files from your Allegro directories, 
   forcing a complete rebuild the next time you run make. This requires the 
   'rm' utility from the fileutils package.

   'veryclean'
   Removes every generated file from your Allegro directories, including all 
   the example executables and utility programs. This requires the 'rm' 
   utility from the fileutils package.

   'uninstall'
   Removes allegro.h, liballeg.a, and allegro.inf from your djgpp 
   directories. This requires the 'rm' utility from the fileutils package.



====================================
============ Conclusion ============
====================================

All good things must come to an end. Writing documentation is not a good 
thing, though, and that means it goes on for ever. There is always something 
I've forgotten to explain, or some essential detail I've left out, but for 
now you will have to make do with this. Feel free to ask if you can't figure 
something out.

Enjoy. I hope you find some of this stuff useful.


By Shawn Hargreaves.

http://www.talula.demon.co.uk/allegro/




