File: HACKING

package info (click to toggle)
sgt-puzzles 7983-1
links: PTS
area: main
in suites: lenny
size: 6,792 kB
ctags: 4,927
sloc: ansic: 65,310; perl: 1,527; objc: 1,206; makefile: 152; sh: 23
file content (4375 lines) | stat: -rw-r--r-- 182,648 bytes
Developer documentation for Simon Tatham's puzzle collection
============================================================

This is a guide to the internal structure of Simon Tatham's Portable
Puzzle Collection (henceforth referred to simply as `Puzzles'), for
use by anyone attempting to implement a new puzzle or port to a new
platform.

This guide is believed correct as of r6190. Hopefully it will be updated
along with the code in future, but if not, I've at least left this
version number in here so you can figure out what's changed by tracking
commit comments from there onwards.

1. Introduction
---------------

The Puzzles code base is divided into four parts: a set of
interchangeable front ends, a set of interchangeable back ends, a
universal `middle end' which acts as a buffer between the two, and a
bunch of miscellaneous utility functions. In the following sections I
give some general discussion of each of these parts.

1.1. Front end
--------------

The front end is the non-portable part of the code: it's the bit that
you replace completely when you port to a different platform. So it's
responsible for all system calls, all GUI interaction, and anything else
platform-specific.

The current front ends in the main code base are for Windows, GTK and
MacOS X; I also know of a third-party front end for PalmOS.

The front end contains main() or the local platform's equivalent. Top-
level control over the application's execution flow belongs to the front
end (it isn't, for example, a set of functions called by a universal
main() somewhere else).

The front end has complete freedom to design the GUI for any given
port of Puzzles. There is no centralised mechanism for maintaining the
menu layout, for example. This has a cost in consistency (when I _do_
want the same menu layout on more than one platform, I have to edit
two pieces of code in parallel every time I make a change), but the
advantage is that local GUI conventions can be conformed to and local
constraints adapted to. For example, MacOS X has strict human interface
guidelines which specify a different menu layout from the one I've used
on Windows and GTK; there's nothing stopping the OS X front end from
providing a menu layout consistent with those guidelines.

Although the front end is mostly caller rather than the callee in its
interactions with other parts of the code, it is required to implement
a small API for other modules to call, mostly of drawing functions for
games to use when drawing their graphics. The drawing API is documented
in chapter 3; the other miscellaneous front end API functions are
documented in section 4.26.

1.2. Back end
-------------

A `back end', in this collection, is synonymous with a `puzzle'. Each
back end implements a different game.

At the top level, a back end is simply a data structure, containing a
few constants (flag words, preferred pixel size) and a large number of
function pointers. Back ends are almost invariably callee rather than
caller, which means there's a limitation on what a back end can do on
its own initiative.

The persistent state in a back end is divided into a number of data
structures, which are used for different purposes and therefore likely
to be switched around, changed without notice, and otherwise updated by
the rest of the code. It is important when designing a back end to put
the right pieces of data into the right structures, or standard midend-
provided features (such as Undo) may fail to work.

The functions and variables provided in the back end data structure are
documented in chapter 2.

1.3. Middle end
---------------

Puzzles has a single and universal `middle end'. This code is common to
all platforms and all games; it sits in between the front end and the
back end and provides standard functionality everywhere.

People adding new back ends or new front ends should generally not need
to edit the middle end. On rare occasions there might be a change that
can be made to the middle end to permit a new game to do something not
currently anticipated by the middle end's present design; however, this
is terribly easy to get wrong and should probably not be undertaken
without consulting the primary maintainer (me). Patch submissions
containing unannounced mid-end changes will be treated on their merits
like any other patch; this is just a friendly warning that mid-end
changes will need quite a lot of merits to make them acceptable.

Functionality provided by the mid-end includes:

 -  Maintaining a list of game state structures and moving back and
    forth along that list to provide Undo and Redo.

 -  Handling timers (for move animations, flashes on completion, and in
    some cases actually timing the game).

 -  Handling the container format of game IDs: receiving them, picking
    them apart into parameters, description and/or random seed, and
    so on. The game back end need only handle the individual parts
    of a game ID (encoded parameters and encoded game description);
    everything else is handled centrally by the mid-end.

 -  Handling standard keystrokes and menu commands, such as `New Game',
    `Restart Game' and `Quit'.

 -  Pre-processing mouse events so that the game back ends can rely on
    them arriving in a sensible order (no missing button-release events,
    no sudden changes of which button is currently pressed, etc).

 -  Handling the dialog boxes which ask the user for a game ID.

 -  Handling serialisation of entire games (for loading and saving a
    half-finished game to a disk file, or for handling application
    shutdown and restart on platforms such as PalmOS where state is
    expected to be saved).

Thus, there's a lot of work done once by the mid-end so that individual
back ends don't have to worry about it. All the back end has to do is
cooperate in ensuring the mid-end can do its work properly.

The API of functions provided by the mid-end to be called by the front
end is documented in chapter 4.

1.4. Miscellaneous utilities
----------------------------

In addition to these three major structural components, the Puzzles code
also contains a variety of utility modules usable by all of the above
components. There is a set of functions to provide platform-independent
random number generation; functions to make memory allocation easier;
functions which implement a balanced tree structure to be used as
necessary in complex algorithms; and a few other miscellaneous
functions. All of these are documented in chapter 5.

1.5. Structure of this guide
----------------------------

There are a number of function call interfaces within Puzzles, and
this guide will discuss each one in a chapter of its own. After that,
(chapter 6) discusses how to design new games, with some general design
thoughts and tips.

2. Interface to the back end
----------------------------

This chapter gives a detailed discussion of the interface that each back
end must implement.

At the top level, each back end source file exports a single global
symbol, which is a `const struct game' containing a large number of
function pointers and a small amount of constant data. This structure is
called by different names depending on what kind of platform the puzzle
set is being compiled on:

 -  On platforms such as Windows and GTK, which build a separate binary
    for each puzzle, the game structure in every back end has the same
    name, `thegame'; the front end refers directly to this name, so that
    compiling the same front end module against a different back end
    module builds a different puzzle.

 -  On platforms such as MacOS X and PalmOS, which build all the puzzles
    into a single monolithic binary, the game structure in each back end
    must have a different name, and there's a helper module `list.c'
    (constructed automatically by the same Perl script that builds the
    Makefiles) which contains a complete list of those game structures.

On the latter type of platform, source files may assume that the
preprocessor symbol `COMBINED' has been defined. Thus, the usual code to
declare the game structure looks something like this:

  #ifdef COMBINED
  #define thegame net    /* or whatever this game is called */
  #endif
  
  const struct game thegame = {
      /* lots of structure initialisation in here */
  };

Game back ends must also internally define a number of data structures,
for storing their various persistent state. This chapter will first
discuss the nature and use of those structures, and then go on to give
details of every element of the game structure.

2.1. Data structures
--------------------

Each game is required to define four separate data structures. This
section discusses each one and suggests what sorts of things need to be
put in it.

2.1.1. `game_params'
--------------------

The `game_params' structure contains anything which affects the
automatic generation of new puzzles. So if puzzle generation is
parametrised in any way, those parameters need to be stored in
`game_params'.

Most puzzles currently in this collection are played on a grid of
squares, meaning that the most obvious parameter is the grid size. Many
puzzles have additional parameters; for example, Mines allows you to
control the number of mines in the grid independently of its size, Net
can be wrapping or non-wrapping, Solo has difficulty levels and symmetry
settings, and so on.

A simple rule for deciding whether a data item needs to go in
`game_params' is: would the user expect to be able to control this data
item from either the preset-game-types menu or the `Custom' game type
configuration? If so, it's part of `game_params'.

`game_params' structures are permitted to contain pointers to subsidiary
data if they need to. The back end is required to provide functions to
create and destroy `game_params', and those functions can allocate and
free additional memory if necessary. (It has not yet been necessary to
do this in any puzzle so far, but the capability is there just in case.)

`game_params' is also the only structure which the game's compute_size()
function may refer to; this means that any aspect of the game which
affects the size of the window it needs to be drawn in must be stored in
`game_params'. In particular, this imposes the fundamental limitation
that random game generation may not have a random effect on the window
size: game generation algorithms are constrained to work by starting
from the grid size rather than generating it as an emergent phenomenon.
(Although this is a restriction in theory, it has not yet seemed to be a
problem.)

2.1.2. `game_state'
-------------------

While the user is actually playing a puzzle, the `game_state' structure
stores all the data corresponding to the current state of play.

The mid-end keeps `game_state's in a list, and adds to the list every
time the player makes a move; the Undo and Redo functions step back and
forth through that list.

Therefore, a good means of deciding whether a data item needs to go in
`game_state' is: would a player expect that data item to be restored on
undo? If so, put it in `game_state', and this will automatically happen
without you having to lift a finger. If not - for example, the deaths
counter in Mines is precisely something that does _not_ want to be reset
to its previous state on an undo - then you might have found a data item
that needs to go in `game_ui' instead.

During play, `game_state's are often passed around without an
accompanying `game_params' structure. Therefore, any information in
`game_params' which is important during play (such as the grid size)
must be duplicated within the `game_state'. One simple method of doing
this is to have the `game_state' structure _contain_ a `game_params'
structure as one of its members, although this isn't obligatory if you
prefer to do it another way.

2.1.3. `game_drawstate'
-----------------------

`game_drawstate' carries persistent state relating to the current
graphical contents of the puzzle window. The same `game_drawstate'
is passed to every call to the game redraw function, so that it can
remember what it has already drawn and what needs redrawing.

A typical use for a `game_drawstate' is to have an array mirroring the
array of grid squares in the `game_state'; then every time the redraw
function was passed a `game_state', it would loop over all the squares,
and physically redraw any whose description in the `game_state' (i.e.
what the square needs to look like when the redraw is completed) did
not match its description in the `game_drawstate' (i.e. what the square
currently looks like).

`game_drawstate' is occasionally completely torn down and reconstructed
by the mid-end, if the user somehow forces a full redraw. Therefore, no
data should be stored in `game_drawstate' which is _not_ related to the
state of the puzzle window, because it might be unexpectedly destroyed.

The back end provides functions to create and destroy `game_drawstate',
which means it can contain pointers to subsidiary allocated data if it
needs to. A common thing to want to allocate in a `game_drawstate' is a
`blitter'; see section 3.1.11 for more on this subject.

2.1.4. `game_ui'
----------------

`game_ui' contains whatever doesn't fit into the above three structures!

A new `game_ui' is created when the user begins playing a new instance
of a puzzle (i.e. during `New Game' or after entering a game ID etc). It
persists until the user finishes playing that game and begins another
one (or closes the window); in particular, `Restart Game' does _not_
destroy the `game_ui'.

`game_ui' is useful for implementing user-interface state which is not
part of `game_state'. Common examples are keyboard control (you wouldn't
want to have to separately Undo through every cursor motion) and mouse
dragging. See section 6.3.2 and section 6.3.3, respectively, for more
details.

Another use for `game_ui' is to store highly persistent data such as
the Mines death counter. This is conceptually rather different: where
the Net cursor position was _not important enough_ to preserve for the
player to restore by Undo, the Mines death counter is _too important_ to
permit the player to revert by Undo!

A final use for `game_ui' is to pass information to the redraw function
about recent changes to the game state. This is used in Mines, for
example, to indicate whether a requested `flash' should be a white flash
for victory or a red flash for defeat; see section 6.3.5.

2.2. Simple data in the back end
--------------------------------

In this section I begin to discuss each individual element in the back
end structure. To begin with, here are some simple self-contained data
elements.

2.2.1. `name'
-------------

  const char *name;

This is a simple ASCII string giving the name of the puzzle. This name
will be used in window titles, in game selection menus on monolithic
platforms, and anywhere else that the front end needs to know the name
of a game.

2.2.2. `winhelp_topic'
----------------------

  const char *winhelp_topic;

This member is used on Windows only, to provide online help. Although
the Windows front end provides a separate binary for each puzzle, it has
a single monolithic help file; so when a user selects `Help' from the
menu, the program needs to open the help file and jump to the chapter
describing that particular puzzle.

Therefore, each chapter in `puzzles.but' is labelled with a _help topic_
name, similar to this:

  \cfg{winhelp-topic}{games.net}

And then the corresponding game back end encodes the topic string (here
`games.net') in the `winhelp_topic' element of the game structure.

2.3. Handling game parameter sets
---------------------------------

In this section I present the various functions which handle the
`game_params' structure.

2.3.1. default_params()
-----------------------

  game_params *(*default_params)(void);

This function allocates a new `game_params' structure, fills it with the
default values, and returns a pointer to it.

2.3.2. fetch_preset()
---------------------

  int (*fetch_preset)(int i, char **name, game_params **params);

This function is used to populate the `Type' menu, which provides a list
of conveniently accessible preset parameters for most games.

The function is called with `i' equal to the index of the preset
required (numbering from zero). It returns FALSE if that preset does
not exist (if `i' is less than zero or greater than the largest preset
index). Otherwise, it sets `*params' to point at a newly allocated
`game_params' structure containing the preset information, sets `*name'
to point at a newly allocated C string containing the preset title (to
go on the `Type' menu), and returns TRUE.

If the game does not wish to support any presets at all, this function
is permitted to return FALSE always.

2.3.3. encode_params()
----------------------

  char *(*encode_params)(game_params *params, int full);

The job of this function is to take a `game_params', and encode it in
a string form for use in game IDs. The return value must be a newly
allocated C string, and _must_ not contain a colon or a hash (since
those characters are used to mark the end of the parameter section in a
game ID).

Ideally, it should also not contain any other potentially controversial
punctuation; bear in mind when designing a string parameter format
that it will probably be used on both Windows and Unix command lines
under a variety of exciting shell quoting and metacharacter rules.
Sticking entirely to alphanumerics is the safest thing; if you really
need punctuation, you can probably get away with commas, periods or
underscores without causing anybody any major inconvenience. If you
venture far beyond that, you're likely to irritate _somebody_.

(At the time of writing this, all existing games have purely
alphanumeric string parameter formats. Usually these involve a letter
denoting a parameter, followed optionally by a number giving the value
of that parameter, with a few mandatory parts at the beginning such as
numeric width and height separated by `x'.)

If the `full' parameter is TRUE, this function should encode absolutely
everything in the `game_params', such that a subsequent call to
decode_params() (section 2.3.4) will yield an identical structure.
If `full' is FALSE, however, you should leave out anything which
is not necessary to describe a _specific puzzle instance_, i.e.
anything which only takes effect when a new puzzle is _generated_.
For example, the Solo `game_params' includes a difficulty rating used
when constructing new puzzles; but a Solo game ID need not explicitly
include the difficulty, since to describe a puzzle once generated it's
sufficient to give the grid dimensions and the location and contents
of the clue squares. (Indeed, one might very easily type in a puzzle
out of a newspaper without _knowing_ what its difficulty level is in
Solo's terminology.) Therefore, Solo's encode_params() only encodes the
difficulty level if `full' is set.

2.3.4. decode_params()
----------------------

  void (*decode_params)(game_params *params, char const *string);

This function is the inverse of encode_params() (section 2.3.3). It
parses the supplied string and fills in the supplied `game_params'
structure. Note that the structure will _already_ have been allocated:
this function is not expected to create a _new_ `game_params', but to
modify an existing one.

This function can receive a string which only encodes a subset of the
parameters. The most obvious way in which this can happen is if the
string was constructed by encode_params() with its `full' parameter set
to FALSE; however, it could also happen if the user typed in a parameter
set manually and missed something out. Be prepared to deal with a wide
range of possibilities.

When dealing with a parameter which is not specified in the input
string, what to do requires a judgment call on the part of the
programmer. Sometimes it makes sense to adjust other parameters to bring
them into line with the new ones. In Mines, for example, you would
probably not want to keep the same mine count if the user dropped the
grid size and didn't specify one, since you might easily end up with
more mines than would actually fit in the grid! On the other hand,
sometimes it makes sense to leave the parameter alone: a Solo player
might reasonably expect to be able to configure size and difficulty
independently of one another.

This function currently has no direct means of returning an error if the
string cannot be parsed at all. However, the returned `game_params' is
almost always subsequently passed to validate_params() (section 2.3.10),
so if you really want to signal parse errors, you could always have a
`char *' in your parameters structure which stored an error message, and
have validate_params() return it if it is non-NULL.

2.3.5. free_params()
--------------------

  void (*free_params)(game_params *params);

This function frees a `game_params' structure, and any subsidiary
allocations contained within it.

2.3.6. dup_params()
-------------------

  game_params *(*dup_params)(game_params *params);

This function allocates a new `game_params' structure and initialises it
with an exact copy of the information in the one provided as input. It
returns a pointer to the new duplicate.

2.3.7. `can_configure'
----------------------

  int can_configure;

This boolean data element is set to TRUE if the back end supports
custom parameter configuration via a dialog box. If it is TRUE, then
the functions configure() and custom_params() are expected to work. See
section 2.3.8 and section 2.3.9 for more details.

2.3.8. configure()
------------------

  config_item *(*configure)(game_params *params);

This function is called when the user requests a dialog box for
custom parameter configuration. It returns a newly allocated array of
config_item structures, describing the GUI elements required in the
dialog box. The array should have one more element than the number of
controls, since it is terminated with a C_END marker (see below). Each
array element describes the control together with its initial value; the
front end will modify the value fields and return the updated array to
custom_params() (see section 2.3.9).

The config_item structure contains the following elements:

  char *name;
  int type;
  char *sval;
  int ival;

`name' is an ASCII string giving the textual label for a GUI control. It
is _not_ expected to be dynamically allocated.

`type' contains one of a small number of `enum' values defining what
type of control is being described. The meaning of the `sval' and `ival'
fields depends on the value in `type'. The valid values are:

`C_STRING'

    Describes a text input box. (This is also used for numeric input.
    The back end does not bother informing the front end that the box is
    numeric rather than textual; some front ends do have the capacity
    to take this into account, but I decided it wasn't worth the extra
    complexity in the interface.) For this type, `ival' is unused, and
    `sval' contains a dynamically allocated string representing the
    contents of the input box.

`C_BOOLEAN'

    Describes a simple checkbox. For this type, `sval' is unused, and
    `ival' is TRUE or FALSE.

`C_CHOICES'

    Describes a drop-down list presenting one of a small number of
    fixed choices. For this type, `sval' contains a list of strings
    describing the choices; the very first character of `sval' is
    used as a delimiter when processing the rest (so that the strings
    `:zero:one:two', `!zero!one!two' and `xzeroxonextwo' all define
    a three-element list containing `zero', `one' and `two'). `ival'
    contains the index of the currently selected element, numbering from
    zero (so that in the above example, 0 would mean `zero' and 2 would
    mean `two').

    Note that for this control type, `sval' is _not_ dynamically
    allocated, whereas it was for `C_STRING'.

`C_END'

    Marks the end of the array of `config_item's. All other fields are
    unused.

The array returned from this function is expected to have filled in the
initial values of all the controls according to the input `game_params'
structure.

If the game's `can_configure' flag is set to FALSE, this function is
never called and need not do anything at all.

2.3.9. custom_params()
----------------------

  game_params *(*custom_params)(config_item *cfg);

This function is the counterpart to configure() (section 2.3.8). It
receives as input an array of `config_item's which was originally
created by configure(), but in which the control values have since been
changed in accordance with user input. Its function is to read the new
values out of the controls and return a newly allocated `game_params'
structure representing the user's chosen parameter set.

(The front end will have modified the controls' _values_, but there will
still always be the same set of controls, in the same order, as provided
by configure(). It is not necessary to check the `name' and `type'
fields, although you could use assert() if you were feeling energetic.)

This function is not expected to (and indeed _must not_) free the input
`config_item' array. (If the parameters fail to validate, the dialog box
will stay open.)

If the game's `can_configure' flag is set to FALSE, this function is
never called and need not do anything at all.

2.3.10. validate_params()
-------------------------

  char *(*validate_params)(game_params *params, int full);

This function takes a `game_params' structure as input, and checks that
the parameters described in it fall within sensible limits. (At the very
least, grid dimensions should almost certainly be strictly positive, for
example.)

Return value is NULL if no problems were found, or alternatively a (non-
dynamically-allocated) ASCII string describing the error in human-
readable form.

If the `full' parameter is set, full validation should be performed: any
set of parameters which would not permit generation of a sensible puzzle
should be faulted. If `full' is _not_ set, the implication is that
these parameters are not going to be used for _generating_ a puzzle; so
parameters which can't even sensibly _describe_ a valid puzzle should
still be faulted, but parameters which only affect puzzle generation
should not be.

(The `full' option makes a difference when parameter combinations are
non-orthogonal. For example, Net has a boolean option controlling
whether it enforces a unique solution; it turns out that it's impossible
to generate a uniquely soluble puzzle with wrapping walls and width
2, so validate_params() will complain if you ask for one. However,
if the user had just been playing a unique wrapping puzzle of a more
sensible width, and then pastes in a game ID acquired from somebody else
which happens to describe a _non_-unique wrapping width-2 puzzle, then
validate_params() will be passed a `game_params' containing the width
and wrapping settings from the new game ID and the uniqueness setting
from the old one. This would be faulted, if it weren't for the fact that
`full' is not set during this call, so Net ignores the inconsistency.
The resulting `game_params' is never subsequently used to generate a
puzzle; this is a promise made by the mid-end when it asks for a non-
full validation.)

2.4. Handling game descriptions
-------------------------------

In this section I present the functions that deal with a textual
description of a puzzle, i.e. the part that comes after the colon in a
descriptive-format game ID.

2.4.1. new_desc()
-----------------

  char *(*new_desc)(game_params *params, random_state *rs,
                    char **aux, int interactive);

This function is where all the really hard work gets done. This is
the function whose job is to randomly generate a new puzzle, ensuring
solubility and uniqueness as appropriate.

As input it is given a `game_params' structure and a random state
(see section 5.1 for the random number API). It must invent a puzzle
instance, encode it in string form, and return a dynamically allocated C
string containing that encoding.

Additionally, it may return a second dynamically allocated string
in `*aux'. (If it doesn't want to, then it can leave that parameter
completely alone; it isn't required to set it to NULL, although doing
so is harmless.) That string, if present, will be passed to solve()
(section 2.7.4) later on; so if the puzzle is generated in such a way
that a solution is known, then information about that solution can be
saved in `*aux' for solve() to use.

The `interactive' parameter should be ignored by almost all puzzles.
Its purpose is to distinguish between generating a puzzle within a GUI
context for immediate play, and generating a puzzle in a command-line
context for saving to be played later. The only puzzle that currently
uses this distinction (and, I fervently hope, the only one which will
_ever_ need to use it) is Mines, which chooses a random first-click
location when generating puzzles non-interactively, but which waits
for the user to place the first click when interactive. If you think
you have come up with another puzzle which needs to make use of this
parameter, please think for at least ten minutes about whether there is
_any_ alternative!

Note that game description strings are not required to contain an
encoding of parameters such as grid size; a game description is
never separated from the `game_params' it was generated with, so any
information contained in that structure need not be encoded again in the
game description.

2.4.2. validate_desc()
----------------------

  char *(*validate_desc)(game_params *params, char *desc);

This function is given a game description, and its job is to validate
that it describes a puzzle which makes sense.

To some extent it's up to the user exactly how far they take the phrase
`makes sense'; there are no particularly strict rules about how hard the
user is permitted to shoot themself in the foot when typing in a bogus
game description by hand. (For example, Rectangles will not verify that
the sum of all the numbers in the grid equals the grid's area. So a user
could enter a puzzle which was provably not soluble, and the program
wouldn't complain; there just wouldn't happen to be any sequence of
moves which solved it.)

The one non-negotiable criterion is that any game description which
makes it through validate_desc() _must not_ subsequently cause a crash
or an assertion failure when fed to new_game() and thence to the rest of
the back end.

The return value is NULL on success, or a non-dynamically-allocated C
string containing an error message.

2.4.3. new_game()
-----------------

  game_state *(*new_game)(midend *me, game_params *params,
                          char *desc);

This function takes a game description as input, together with its
accompanying `game_params', and constructs a `game_state' describing the
initial state of the puzzle. It returns a newly allocated `game_state'
structure.

Almost all puzzles should ignore the `me' parameter. It is required by
Mines, which needs it for later passing to midend_supersede_game_desc()
(see section 2.11.2) once the user has placed the first click. I
fervently hope that no other puzzle will be awkward enough to require
it, so everybody else should ignore it. As with the `interactive'
parameter in new_desc() (section 2.4.1), if you think you have a reason
to need this parameter, please try very hard to think of an alternative
approach!

2.5. Handling game states
-------------------------

This section describes the functions which create and destroy
`game_state' structures.

(Well, except new_game(), which is in section 2.4.3 instead of under
here; but it deals with game descriptions _and_ game states and it had
to go in one section or the other.)

2.5.1. dup_game()
-----------------

  game_state *(*dup_game)(game_state *state);

This function allocates a new `game_state' structure and initialises it
with an exact copy of the information in the one provided as input. It
returns a pointer to the new duplicate.

2.5.2. free_game()
------------------

  void (*free_game)(game_state *state);

This function frees a `game_state' structure, and any subsidiary
allocations contained within it.

2.6. Handling `game_ui'
-----------------------

2.6.1. new_ui()
---------------

  game_ui *(*new_ui)(game_state *state);

This function allocates and returns a new `game_ui' structure for
playing a particular puzzle. It is passed a pointer to the initial
`game_state', in case it needs to refer to that when setting up the
initial values for the new game.

2.6.2. free_ui()
----------------

  void (*free_ui)(game_ui *ui);

This function frees a `game_ui' structure, and any subsidiary
allocations contained within it.

2.6.3. encode_ui()
------------------

  char *(*encode_ui)(game_ui *ui);

This function encodes any _important_ data in a `game_ui' structure in
string form. It is only called when saving a half-finished game to a
file.

It should be used sparingly. Almost all data in a `game_ui' is not
important enough to save. The location of the keyboard-controlled
cursor, for example, can be reset to a default position on reloading
the game without impacting the user experience. If the user should
somehow manage to save a game while a mouse drag was in progress, then
discarding that mouse drag would be an outright _feature_.

A typical thing that _would_ be worth encoding in this function is the
Mines death counter: it's in the `game_ui' rather than the `game_state'
because it's too important to allow the user to revert it by using Undo,
and therefore it's also too important to allow the user to revert it by
saving and reloading. (Of course, the user could edit the save file by
hand... But if the user is _that_ determined to cheat, they could just
as easily modify the game's source.)

2.6.4. decode_ui()
------------------

  void (*decode_ui)(game_ui *ui, char *encoding);

This function parses a string previously output by encode_ui(), and
writes the decoded data back into the provided `game_ui' structure.

2.6.5. changed_state()
----------------------

  void (*changed_state)(game_ui *ui, game_state *oldstate,
                        game_state *newstate);

This function is called by the mid-end whenever the current game state
changes, for any reason. Those reasons include:

 -  a fresh move being made by interpret_move() and execute_move()

 -  a solve operation being performed by solve() and execute_move()

 -  the user moving back and forth along the undo list by means of the
    Undo and Redo operations

 -  the user selecting Restart to go back to the initial game state.

The job of changed_state() is to update the `game_ui' for consistency
with the new game state, if any update is necessary. For example,
Same Game stores data about the currently selected tile group in its
`game_ui', and this data is intrinsically related to the game state it
was derived from. So it's very likely to become invalid when the game
state changes; thus, Same Game's changed_state() function clears the
current selection whenever it is called.

When anim_length() or flash_length() are called, you can be sure that
there has been a previous call to changed_state(). So changed_state()
can set up data in the `game_ui' which will be read by anim_length() and
flash_length(), and those functions will not have to worry about being
called without the data having been initialised.

2.7. Making moves
-----------------

This section describes the functions which actually make moves in
the game: that is, the functions which process user input and end up
producing new `game_state's.

2.7.1. interpret_move()
-----------------------

  char *(*interpret_move)(game_state *state, game_ui *ui,
                          game_drawstate *ds,
                          int x, int y, int button);

This function receives user input and processes it. Its input parameters
are the current `game_state', the current `game_ui' and the current
`game_drawstate', plus details of the input event. `button' is either
an ASCII value or a special code (listed below) indicating an arrow or
function key or a mouse event; when `button' is a mouse event, `x' and
`y' contain the pixel coordinates of the mouse pointer relative to the
top left of the puzzle's drawing area.

interpret_move() may return in three different ways:

 -  Returning NULL indicates that no action whatsoever occurred in
    response to the input event; the puzzle was not interested in it at
    all.

 -  Returning the empty string ("") indicates that the input event has
    resulted in a change being made to the `game_ui' which will require
    a redraw of the game window, but that no actual _move_ was made
    (i.e. no new `game_state' needs to be created).

 -  Returning anything else indicates that a move was made and that
    a new `game_state' must be created. However, instead of actually
    constructing a new `game_state' itself, this function is required to
    return a string description of the details of the move. This string
    will be passed to execute_move() (section 2.7.2) to actually create
    the new `game_state'. (Encoding moves as strings in this way means
    that the mid-end can keep the strings as well as the game states,
    and the strings can be written to disk when saving the game and fed
    to execute_move() again on reloading.)

The return value from interpret_move() is expected to be dynamically
allocated if and only if it is not either NULL _or_ the empty string.

After this function is called, the back end is permitted to rely on some
subsequent operations happening in sequence:

 -  execute_move() will be called to convert this move description into
    a new `game_state'

 -  changed_state() will be called with the new `game_state'.

This means that if interpret_move() needs to do updates to the `game_ui'
which are easier to perform by referring to the new `game_state', it can
safely leave them to be done in changed_state() and not worry about them
failing to happen.

(Note, however, that execute_move() may _also_ be called in other
circumstances. It is only interpret_move() which can rely on a
subsequent call to changed_state().)

The special key codes supported by this function are:

LEFT_BUTTON, MIDDLE_BUTTON, RIGHT_BUTTON

    Indicate that one of the mouse buttons was pressed down.

LEFT_DRAG, MIDDLE_DRAG, RIGHT_DRAG

    Indicate that the mouse was moved while one of the mouse buttons was
    still down. The mid-end guarantees that when one of these events is
    received, it will always have been preceded by a button-down event
    (and possibly other drag events) for the same mouse button, and no
    event involving another mouse button will have appeared in between.

LEFT_RELEASE, MIDDLE_RELEASE, RIGHT_RELEASE

    Indicate that a mouse button was released. The mid-end guarantees
    that when one of these events is received, it will always have been
    preceded by a button-down event (and possibly some drag events) for
    the same mouse button, and no event involving another mouse button
    will have appeared in between.

CURSOR_UP, CURSOR_DOWN, CURSOR_LEFT, CURSOR_RIGHT

    Indicate that an arrow key was pressed.

CURSOR_SELECT

    On platforms which have a prominent `select' button alongside their
    cursor keys, indicates that that button was pressed.

In addition, there are some modifiers which can be bitwise-ORed into the
`button' parameter:

MOD_CTRL, MOD_SHFT

    These indicate that the Control or Shift key was pressed alongside
    the key. They only apply to the cursor keys, not to mouse buttons or
    anything else.

MOD_NUM_KEYPAD

    This applies to some ASCII values, and indicates that the key code
    was input via the numeric keypad rather than the main keyboard. Some
    puzzles may wish to treat this differently (for example, a puzzle
    might want to use the numeric keypad as an eight-way directional
    pad), whereas others might not (a game involving numeric input
    probably just wants to treat the numeric keypad as numbers).

MOD_MASK

    This mask is the bitwise OR of all the available modifiers; you can
    bitwise-AND with ~MOD_MASK to strip all the modifiers off any input
    value.

2.7.2. execute_move()
---------------------

  game_state *(*execute_move)(game_state *state, char *move);

This function takes an input `game_state' and a move string as output
from interpret_move(). It returns a newly allocated `game_state' which
contains the result of applying the specified move to the input game
state.

This function may return NULL if it cannot parse the move string (and
this is definitely preferable to crashing or failing an assertion, since
one way this can happen is if loading a corrupt save file). However, it
must not return NULL for any move string that really was output from
interpret_move(): this is punishable by assertion failure in the mid-
end.

2.7.3. `can_solve'
------------------

  int can_solve;

This boolean field is set to TRUE if the game's solve() function does
something. If it's set to FALSE, the game will not even offer the
`Solve' menu option.

2.7.4. solve()
--------------

  char *(*solve)(game_state *orig, game_state *curr,
                 char *aux, char **error);

This function is called when the user selects the `Solve' option from
the menu.

It is passed two input game states: `orig' is the game state from the
very start of the puzzle, and `curr' is the current one. (Different
games find one or other or both of these convenient.) It is also passed
the `aux' string saved by new_desc() (section 2.4.1), in case that
encodes important information needed to provide the solution.

If this function is unable to produce a solution (perhaps, for example,
the game has no in-built solver so it can only solve puzzles it invented
internally and has an `aux' string for) then it may return NULL. If it
does this, it must also set `*error' to an error message to be presented
to the user (such as `Solution not known for this puzzle'); that error
message is not expected to be dynamically allocated.

If this function _does_ produce a solution, it returns a move string
suitable for feeding to execute_move() (section 2.7.2).

2.8. Drawing the game graphics
------------------------------

This section discusses the back end functions that deal with drawing.

2.8.1. new_drawstate()
----------------------

  game_drawstate *(*new_drawstate)(drawing *dr, game_state *state);

This function allocates and returns a new `game_drawstate' structure for
drawing a particular puzzle. It is passed a pointer to a `game_state',
in case it needs to refer to that when setting up any initial data.

This function may not rely on the puzzle having been newly started; a
new draw state can be constructed at any time if the front end requests
a forced redraw. For games like Pattern, in which initial game states
are much simpler than general ones, this might be important to keep in
mind.

The parameter `dr' is a drawing object (see chapter 3) which the
function might need to use to allocate blitters. (However, this isn't
recommended; it's usually more sensible to wait to allocate a blitter
until set_size() is called, because that way you can tailor it to the
scale at which the puzzle is being drawn.)

2.8.2. free_drawstate()
-----------------------

  void (*free_drawstate)(drawing *dr, game_drawstate *ds);

This function frees a `game_drawstate' structure, and any subsidiary
allocations contained within it.

The parameter `dr' is a drawing object (see chapter 3), which might be
required if you are freeing a blitter.

2.8.3. `preferred_tilesize'
---------------------------

  int preferred_tilesize;

Each game is required to define a single integer parameter which
expresses, in some sense, the scale at which it is drawn. This is
described in the APIs as `tilesize', since most puzzles are on a
square (or possibly triangular or hexagonal) grid and hence a sensible
interpretation of this parameter is to define it as the size of one grid
tile in pixels; however, there's no actual requirement that the `tile
size' be proportional to the game window size. Window size is required
to increase monotonically with `tile size', however.

The data element `preferred_tilesize' indicates the tile size which
should be used in the absence of a good reason to do otherwise (such as
the screen being too small, or the user explicitly requesting a resize
if that ever gets implemented).

2.8.4. compute_size()
---------------------

  void (*compute_size)(game_params *params, int tilesize,
                       int *x, int *y);

This function is passed a `game_params' structure and a tile size. It
returns, in `*x' and `*y', the size in pixels of the drawing area that
would be required to render a puzzle with those parameters at that tile
size.

2.8.5. set_size()
-----------------

  void (*set_size)(drawing *dr, game_drawstate *ds,
                   game_params *params, int tilesize);

This function is responsible for setting up a `game_drawstate' to draw
at a given tile size. Typically this will simply involve copying the
supplied `tilesize' parameter into a `tilesize' field inside the draw
state; for some more complex games it might also involve setting up
other dimension fields, or possibly allocating a blitter (see section
3.1.11).

The parameter `dr' is a drawing object (see chapter 3), which is
required if a blitter needs to be allocated.

Back ends may assume (and may enforce by assertion) that this function
will be called at most once for any `game_drawstate'. If a puzzle needs
to be redrawn at a different size, the mid-end will create a fresh
drawstate.

2.8.6. colours()
----------------

  float *(*colours)(frontend *fe, int *ncolours);

This function is responsible for telling the front end what colours the
puzzle will need to draw itself.

It returns the number of colours required in `*ncolours', and the return
value from the function itself is a dynamically allocated array of three
times that many `float's, containing the red, green and blue components
of each colour respectively as numbers in the range [0,1].

The second parameter passed to this function is a front end handle.
The only things it is permitted to do with this handle are to call the
front-end function called frontend_default_colour() (see section 4.31)
or the utility function called game_mkhighlight() (see section 5.4.7).
(The latter is a wrapper on the former, so front end implementors only
need to provide frontend_default_colour().) This allows colours() to
take local configuration into account when deciding on its own colour
allocations. Most games use the front end's default colour as their
background, apart from a few which depend on drawing relief highlights
so they adjust the background colour if it's too light for highlights to
show up against it.

Note that the colours returned from this function are for _drawing_,
not for printing. Printing has an entirely different colour allocation
policy.

2.8.7. anim_length()
--------------------

  float (*anim_length)(game_state *oldstate, game_state *newstate,
                       int dir, game_ui *ui);

This function is called when a move is made, undone or redone. It is
given the old and the new `game_state', and its job is to decide whether
the transition between the two needs to be animated or can be instant.

`oldstate' is the state that was current until this call; `newstate'
is the state that will be current after it. `dir' specifies the
chronological order of those states: if it is positive, then the
transition is the result of a move or a redo (and so `newstate' is the
later of the two moves), whereas if it is negative then the transition
is the result of an undo (so that `newstate' is the _earlier_ move).

If this function decides the transition should be animated, it returns
the desired length of the animation in seconds. If not, it returns zero.

State changes as a result of a Restart operation are never animated; the
mid-end will handle them internally and never consult this function at
all. State changes as a result of Solve operations are also not animated
by default, although you can change this for a particular game by
setting a flag in `flags' (section 2.10.6).

The function is also passed a pointer to the local `game_ui'. It may
refer to information in here to help with its decision (see section
6.3.7 for an example of this), and/or it may _write_ information about
the nature of the animation which will be read later by redraw().

When this function is called, it may rely on changed_state() having been
called previously, so if anim_length() needs to refer to information in
the `game_ui', then changed_state() is a reliable place to have set that
information up.

Move animations do not inhibit further input events. If the user
continues playing before a move animation is complete, the animation
will be abandoned and the display will jump straight to the final state.

2.8.8. flash_length()
---------------------

  float (*flash_length)(game_state *oldstate, game_state *newstate,
                        int dir, game_ui *ui);

This function is called when a move is completed. (`Completed'
means that not only has the move been made, but any animation which
accompanied it has finished.) It decides whether the transition from
`oldstate' to `newstate' merits a `flash'.

A flash is much like a move animation, but it is _not_ interrupted by
further user interface activity; it runs to completion in parallel with
whatever else might be going on on the display. The only thing which
will rush a flash to completion is another flash.

The purpose of flashes is to indicate that the game has been completed.
They were introduced as a separate concept from move animations because
of Net: the habit of most Net players (and certainly me) is to rotate a
tile into place and immediately lock it, then move on to another tile.
When you make your last move, at the instant the final tile is rotated
into place the screen starts to flash to indicate victory - but if you
then press the lock button out of habit, then the move animation is
cancelled, and the victory flash does not complete. (And if you _don't_
press the lock button, the completed grid will look untidy because there
will be one unlocked square.) Therefore, I introduced a specific concept
of a `flash' which is separate from a move animation and can proceed in
parallel with move animations and any other display activity, so that
the victory flash in Net is not cancelled by that final locking move.

The input parameters to flash_length() are exactly the same as the ones
to anim_length().

Just like anim_length(), when this function is called, it may rely on
changed_state() having been called previously, so if it needs to refer
to information in the `game_ui' then changed_state() is a reliable place
to have set that information up.

(Some games use flashes to indicate defeat as well as victory; Mines,
for example, flashes in a different colour when you tread on a mine from
the colour it uses when you complete the game. In order to achieve this,
its flash_length() function has to store a flag in the `game_ui' to
indicate which flash type is required.)

2.8.9. redraw()
---------------

  void (*redraw)(drawing *dr, game_drawstate *ds,
                 game_state *oldstate, game_state *newstate, int dir,
                 game_ui *ui, float anim_time, float flash_time);

This function is responsible for actually drawing the contents of
the game window, and for redrawing every time the game state or the
`game_ui' changes.

The parameter `dr' is a drawing object which may be passed to the
drawing API functions (see chapter 3 for documentation of the drawing
API). This function may not save `dr' and use it elsewhere; it must only
use it for calling back to the drawing API functions within its own
lifetime.

`ds' is the local `game_drawstate', of course, and `ui' is the local
`game_ui'.

`newstate' is the semantically-current game state, and is always non-
NULL. If `oldstate' is also non-NULL, it means that a move has recently
been made and the game is still in the process of displaying an
animation linking the old and new states; in this situation, `anim_time'
will give the length of time (in seconds) that the animation has already
been running. If `oldstate' is NULL, then `anim_time' is unused (and
will hopefully be set to zero to avoid confusion).

`flash_time', if it is is non-zero, denotes that the game is in the
middle of a flash, and gives the time since the start of the flash. See
section 2.8.8 for general discussion of flashes.

The very first time this function is called for a new `game_drawstate',
it is expected to redraw the _entire_ drawing area. Since this often
involves drawing visual furniture which is never subsequently altered,
it is often simplest to arrange this by having a special `first time'
flag in the draw state, and resetting it after the first redraw.

When this function (or any subfunction) calls the drawing API, it is
expected to pass colour indices which were previously defined by the
colours() function.

2.9. Printing functions
-----------------------

This section discusses the back end functions that deal with printing
puzzles out on paper.

2.9.1. `can_print'
------------------

  int can_print;

This flag is set to TRUE if the puzzle is capable of printing itself
on paper. (This makes sense for some puzzles, such as Solo, which can
be filled in with a pencil. Other puzzles, such as Twiddle, inherently
involve moving things around and so would not make sense to print.)

If this flag is FALSE, then the functions print_size() and print() will
never be called.

2.9.2. `can_print_in_colour'
----------------------------

  int can_print_in_colour;

This flag is set to TRUE if the puzzle is capable of printing itself
differently when colour is available. For example, Map can actually
print coloured regions in different _colours_ rather than resorting to
cross-hatching.

If the `can_print' flag is FALSE, then this flag will be ignored.

2.9.3. print_size()
-------------------

  void (*print_size)(game_params *params, float *x, float *y);

This function is passed a `game_params' structure and a tile size. It
returns, in `*x' and `*y', the preferred size in _millimetres_ of that
puzzle if it were to be printed out on paper.

If the `can_print' flag is FALSE, this function will never be called.

2.9.4. print()
--------------

  void (*print)(drawing *dr, game_state *state, int tilesize);

This function is called when a puzzle is to be printed out on paper. It
should use the drawing API functions (see chapter 3) to print itself.

This function is separate from redraw() because it is often very
different:

 -  The printing function may not depend on pixel accuracy, since
    printer resolution is variable. Draw as if your canvas had infinite
    resolution.

 -  The printing function sometimes needs to display things in a
    completely different style. Net, for example, is very different as
    an on-screen puzzle and as a printed one.

 -  The printing function is often much simpler since it has no need to
    deal with repeated partial redraws.

However, there's no reason the printing and redraw functions can't share
some code if they want to.

When this function (or any subfunction) calls the drawing API, the
colour indices it passes should be colours which have been allocated by
the print_*_colour() functions within this execution of print(). This is
very different from the fixed small number of colours used in redraw(),
because printers do not have a limitation on the total number of colours
that may be used. Some puzzles' printing functions might wish to
allocate only one `ink' colour and use it for all drawing; others might
wish to allocate _more_ colours than are used on screen.

One possible colour policy worth mentioning specifically is that a
puzzle's printing function might want to allocate the _same_ colour
indices as are used by the redraw function, so that code shared between
drawing and printing does not have to keep switching its colour indices.
In order to do this, the simplest thing is to make use of the fact that
colour indices returned from print_*_colour() are guaranteed to be in
increasing order from zero. So if you have declared an `enum' defining
three colours COL_BACKGROUND, COL_THIS and COL_THAT, you might then
write

  int c;
  c = print_mono_colour(dr, 1); assert(c == COL_BACKGROUND);
  c = print_mono_colour(dr, 0); assert(c == COL_THIS);
  c = print_mono_colour(dr, 0); assert(c == COL_THAT);

If the `can_print' flag is FALSE, this function will never be called.

2.10. Miscellaneous
-------------------

2.10.1. `can_format_as_text'
----------------------------

  int can_format_as_text;

This boolean field is TRUE if the game supports formatting a game state
as ASCII text (typically ASCII art) for copying to the clipboard and
pasting into other applications. If it is FALSE, front ends will not
offer the `Copy' command at all.

If this field is FALSE, the function text_format() (section 2.10.2) is
not expected to do anything at all.

2.10.2. text_format()
---------------------

  char *(*text_format)(game_state *state);

This function is passed a `game_state', and returns a newly allocated C
string containing an ASCII representation of that game state. It is used
to implement the `Copy' operation in many front ends.

This function should only be called if the back end field
`can_format_as_text' (section 2.10.1) is TRUE.

The returned string may contain line endings (and will probably want
to), using the normal C internal `\n' convention. For consistency
between puzzles, all multi-line textual puzzle representations should
_end_ with a newline as well as containing them internally. (There are
currently no puzzles which have a one-line ASCII representation, so
there's no precedent yet for whether that should come with a newline or
not.)

2.10.3. wants_statusbar()
-------------------------

  int wants_statusbar;

This boolean field is set to TRUE if the puzzle has a use for a textual
status line (to display score, completion status, currently active
tiles, etc).

2.10.4. `is_timed'
------------------

  int is_timed;

This boolean field is TRUE if the puzzle is time-critical. If so, the
mid-end will maintain a game timer while the user plays.

If this field is FALSE, then timing_state() will never be called and
need not do anything.

2.10.5. timing_state()
----------------------

  int (*timing_state)(game_state *state, game_ui *ui);

This function is passed the current `game_state' and the local
`game_ui'; it returns TRUE if the game timer should currently be
running.

A typical use for the `game_ui' in this function is to note when the
game was first completed (by setting a flag in changed_state() - see
section 2.6.5), and freeze the timer thereafter so that the user can
undo back through their solution process without altering their time.

2.10.6. `flags'
---------------

  int flags;

This field contains miscellaneous per-backend flags. It consists of the
bitwise OR of some combination of the following:

BUTTON_BEATS(x,y)

    Given any x and y from the set {LEFT_BUTTON, MIDDLE_BUTTON,
    RIGHT_BUTTON}, this macro evaluates to a bit flag which indicates
    that when buttons x and y are both pressed simultaneously, the mid-
    end should consider x to have priority. (In the absence of any such
    flags, the mid-end will always consider the most recently pressed
    button to have priority.)

SOLVE_ANIMATES

    This flag indicates that moves generated by solve() (section 2.7.4)
    are candidates for animation just like any other move. For most
    games, solve moves should not be animated, so the mid-end doesn't
    even bother calling anim_length() (section 2.8.7), thus saving some
    special-case code in each game. On the rare occasion that animated
    solve moves are actually required, you can set this flag.

REQUIRE_RBUTTON

    This flag indicates that the puzzle cannot be usefully played
    without the use of mouse buttons other than the left one. On some
    PDA platforms, this flag is used by the front end to enable right-
    button emulation through an appropriate gesture. Note that a puzzle
    is not required to set this just because it _uses_ the right button,
    but only if its use of the right button is critical to playing the
    game. (Slant, for example, uses the right button to cycle through
    the three square states in the opposite order from the left button,
    and hence can manage fine without it.)

REQUIRE_NUMPAD

    This flag indicates that the puzzle cannot be usefully played
    without the use of number-key input. On some PDA platforms it
    causes an emulated number pad to appear on the screen. Similarly to
    REQUIRE_RBUTTON, a puzzle need not specify this simply if its use of
    the number keys is not critical.

2.11. Things a back end may do on its own initiative
----------------------------------------------------

This section describes a couple of things that a back end may choose
to do by calling functions elsewhere in the program, which would not
otherwise be obvious.

2.11.1. Create a random state
-----------------------------

If a back end needs random numbers at some point during normal play, it
can create a fresh `random_state' by first calling `get_random_seed'
(section 4.27) and then passing the returned seed data to random_new().

This is likely not to be what you want. If a puzzle needs randomness in
the middle of play, it's likely to be more sensible to store some sort
of random state within the `game_state', so that the random numbers are
tied to the particular game state and hence the player can't simply keep
undoing their move until they get numbers they like better.

This facility is currently used only in Net, to implement the `jumble'
command, which sets every unlocked tile to a new random orientation.
This randomness _is_ a reasonable use of the feature, because it's non-
adversarial - there's no advantage to the user in getting different
random numbers.

2.11.2. Supersede its own game description
------------------------------------------

In response to a move, a back end is (reluctantly) permitted to call
midend_supersede_game_desc():

  void midend_supersede_game_desc(midend *me,
                                  char *desc, char *privdesc);

When the user selects `New Game', the mid-end calls new_desc()
(section 2.4.1) to get a new game description, and (as well as using
that to generate an initial game state) stores it for the save file
and for telling to the user. The function above overwrites that
game description, and also splits it in two. `desc' becomes the new
game description which is provided to the user on request, and is
also the one used to construct a new initial game state if the user
selects `Restart'. `privdesc' is a `private' game description, used to
reconstruct the game's initial state when reloading.

The distinction between the two, as well as the need for this function
at all, comes from Mines. Mines begins with a blank grid and no
idea of where the mines actually are; new_desc() does almost no
work in interactive mode, and simply returns a string encoding the
`random_state'. When the user first clicks to open a tile, _then_ Mines
generates the mine positions, in such a way that the game is soluble
from that starting point. Then it uses this function to supersede the
random-state game description with a proper one. But it needs two: one
containing the initial click location (because that's what you want to
happen if you restart the game, and also what you want to send to a
friend so that they play _the same game_ as you), and one without the
initial click location (because when you save and reload the game, you
expect to see the same blank initial state as you had before saving).

I should stress again that this function is a horrid hack. Nobody should
use it if they're not Mines; if you think you need to use it, think
again repeatedly in the hope of finding a better way to do whatever it
was you needed to do.

3. The drawing API
------------------

The back end function redraw() (section 2.8.9) is required to draw the
puzzle's graphics on the window's drawing area, or on paper if the
puzzle is printable. To do this portably, it is provided with a drawing
API allowing it to talk directly to the front end. In this chapter I
document that API, both for the benefit of back end authors trying to
use it and for front end authors trying to implement it.

The drawing API as seen by the back end is a collection of global
functions, each of which takes a pointer to a `drawing' structure (a
`drawing object'). These objects are supplied as parameters to the back
end's redraw() and print() functions.

In fact these global functions are not implemented directly by the front
end; instead, they are implemented centrally in `drawing.c' and form a
small piece of middleware. The drawing API as supplied by the front end
is a structure containing a set of function pointers, plus a `void *'
handle which is passed to each of those functions. This enables a single
front end to switch between multiple implementations of the drawing API
if necessary. For example, the Windows API supplies a printing mechanism
integrated into the same GDI which deals with drawing in windows, and
therefore the same API implementation can handle both drawing and
printing; but on Unix, the most common way for applications to print
is by producing PostScript output directly, and although it would be
_possible_ to write a single (say) draw_rect() function which checked
a global flag to decide whether to do GTK drawing operations or output
PostScript to a file, it's much nicer to have two separate functions and
switch between them as appropriate.

When drawing, the puzzle window is indexed by pixel coordinates, with
the top left pixel defined as (0,0) and the bottom right pixel (w-1,h-
1), where `w' and `h' are the width and height values returned by the
back end function compute_size() (section 2.8.4).

When printing, the puzzle's print area is indexed in exactly the same
way (with an arbitrary tile size provided by the printing module
`printing.c'), to facilitate sharing of code between the drawing and
printing routines. However, when printing, puzzles may no longer assume
that the coordinate unit has any relationship to a pixel; the printer's
actual resolution might very well not even be known at print time, so
the coordinate unit might be smaller or larger than a pixel. Puzzles'
print functions should restrict themselves to drawing geometric shapes
rather than fiddly pixel manipulation.

_Puzzles' redraw functions may assume that the surface they draw on is
persistent_. It is the responsibility of every front end to preserve
the puzzle's window contents in the face of GUI window expose issues
and similar. It is not permissible to request that the back end redraw
any part of a window that it has already drawn, unless something has
actually changed as a result of making moves in the puzzle.

Most front ends accomplish this by having the drawing routines draw on a
stored bitmap rather than directly on the window, and copying the bitmap
to the window every time a part of the window needs to be redrawn.
Therefore, it is vitally important that whenever the back end does any
drawing it informs the front end of which parts of the window it has
accessed, and hence which parts need repainting. This is done by calling
draw_update() (section 3.1.9).

In the following sections I first discuss the drawing API as seen by the
back end, and then the _almost_ identical function-pointer form seen by
the front end.

3.1. Drawing API as seen by the back end
----------------------------------------

This section documents the back-end drawing API, in the form of
functions which take a `drawing' object as an argument.

3.1.1. draw_rect()
------------------

  void draw_rect(drawing *dr, int x, int y, int w, int h,
                 int colour);

Draws a filled rectangle in the puzzle window.

`x' and `y' give the coordinates of the top left pixel of the rectangle.
`w' and `h' give its width and height. Thus, the horizontal extent of
the rectangle runs from `x' to `x+w-1' inclusive, and the vertical
extent from `y' to `y+h-1' inclusive.

`colour' is an integer index into the colours array returned by the back
end function colours() (section 2.8.6).

There is no separate pixel-plotting function. If you want to plot a
single pixel, the approved method is to use draw_rect() with width and
height set to 1.

Unlike many of the other drawing functions, this function is guaranteed
to be pixel-perfect: the rectangle will be sharply defined and not anti-
aliased or anything like that.

This function may be used for both drawing and printing.

3.1.2. draw_rect_outline()
--------------------------

  void draw_rect_outline(drawing *dr, int x, int y, int w, int h,
                         int colour);

Draws an outline rectangle in the puzzle window.

`x' and `y' give the coordinates of the top left pixel of the rectangle.
`w' and `h' give its width and height. Thus, the horizontal extent of
the rectangle runs from `x' to `x+w-1' inclusive, and the vertical
extent from `y' to `y+h-1' inclusive.

`colour' is an integer index into the colours array returned by the back
end function colours() (section 2.8.6).

From a back end perspective, this function may be considered to be part
of the drawing API. However, front ends are not required to implement
it, since it is actually implemented centrally (in misc.c) as a wrapper
on draw_polygon().

This function may be used for both drawing and printing.

3.1.3. draw_line()
------------------

  void draw_line(drawing *dr, int x1, int y1, int x2, int y2,
                 int colour);

Draws a straight line in the puzzle window.

`x1' and `y1' give the coordinates of one end of the line. `x2' and `y2'
give the coordinates of the other end. The line drawn includes both
those points.

`colour' is an integer index into the colours array returned by the back
end function colours() (section 2.8.6).

Some platforms may perform anti-aliasing on this function. Therefore, do
not assume that you can erase a line by drawing the same line over it
in the background colour; anti-aliasing might lead to perceptible ghost
artefacts around the vanished line.

This function may be used for both drawing and printing.

3.1.4. draw_polygon()
---------------------

  void draw_polygon(drawing *dr, int *coords, int npoints,
                    int fillcolour, int outlinecolour);

Draws an outlined or filled polygon in the puzzle window.

`coords' is an array of (2*npoints) integers, containing the `x' and `y'
coordinates of `npoints' vertices.

`fillcolour' and `outlinecolour' are integer indices into the colours
array returned by the back end function colours() (section 2.8.6).
`fillcolour' may also be -1 to indicate that the polygon should be
outlined only.

The polygon defined by the specified list of vertices is first filled in
`fillcolour', if specified, and then outlined in `outlinecolour'.

`outlinecolour' may _not_ be -1; it must be a valid colour (and front
ends are permitted to enforce this by assertion). This is because
different platforms disagree on whether a filled polygon should include
its boundary line or not, so drawing _only_ a filled polygon would
have non-portable effects. If you want your filled polygon not to
have a visible outline, you must set `outlinecolour' to the same as
`fillcolour'.

Some platforms may perform anti-aliasing on this function. Therefore, do
not assume that you can erase a polygon by drawing the same polygon over
it in the background colour. Also, be prepared for the polygon to extend
a pixel beyond its obvious bounding box as a result of this; if you
really need it not to do this to avoid interfering with other delicate
graphics, you should probably use clip() (section 3.1.7).

This function may be used for both drawing and printing.

3.1.5. draw_circle()
--------------------

  void draw_circle(drawing *dr, int cx, int cy, int radius,
                   int fillcolour, int outlinecolour);

Draws an outlined or filled circle in the puzzle window.

`cx' and `cy' give the coordinates of the centre of the circle. `radius'
gives its radius. The total horizontal pixel extent of the circle is
from `cx-radius+1' to `cx+radius-1' inclusive, and the vertical extent
similarly around `cy'.

`fillcolour' and `outlinecolour' are integer indices into the colours
array returned by the back end function colours() (section 2.8.6).
`fillcolour' may also be -1 to indicate that the circle should be
outlined only.

The circle is first filled in `fillcolour', if specified, and then
outlined in `outlinecolour'.

`outlinecolour' may _not_ be -1; it must be a valid colour (and front
ends are permitted to enforce this by assertion). This is because
different platforms disagree on whether a filled circle should include
its boundary line or not, so drawing _only_ a filled circle would
have non-portable effects. If you want your filled circle not to
have a visible outline, you must set `outlinecolour' to the same as
`fillcolour'.

Some platforms may perform anti-aliasing on this function. Therefore, do
not assume that you can erase a circle by drawing the same circle over
it in the background colour. Also, be prepared for the circle to extend
a pixel beyond its obvious bounding box as a result of this; if you
really need it not to do this to avoid interfering with other delicate
graphics, you should probably use clip() (section 3.1.7).

This function may be used for both drawing and printing.

3.1.6. draw_text()
------------------

  void draw_text(drawing *dr, int x, int y, int fonttype,
                 int fontsize, int align, int colour, char *text);

Draws text in the puzzle window.

`x' and `y' give the coordinates of a point. The relation of this point
to the location of the text is specified by `align', which is a bitwise
OR of horizontal and vertical alignment flags:

ALIGN_VNORMAL

    Indicates that `y' is aligned with the baseline of the text.

ALIGN_VCENTRE

    Indicates that `y' is aligned with the vertical centre of the
    text. (In fact, it's aligned with the vertical centre of normal
    _capitalised_ text: displaying two pieces of text with ALIGN_VCENTRE
    at the same y-coordinate will cause their baselines to be aligned
    with one another, even if one is an ascender and the other a
    descender.)

ALIGN_HLEFT

    Indicates that `x' is aligned with the left-hand end of the text.

ALIGN_HCENTRE

    Indicates that `x' is aligned with the horizontal centre of the
    text.

ALIGN_HRIGHT

    Indicates that `x' is aligned with the right-hand end of the text.

`fonttype' is either FONT_FIXED or FONT_VARIABLE, for a monospaced
or proportional font respectively. (No more detail than that may be
specified; it would only lead to portability issues between different
platforms.)

`fontsize' is the desired size, in pixels, of the text. This size
corresponds to the overall point size of the text, not to any internal
dimension such as the cap-height.

`colour' is an integer index into the colours array returned by the back
end function colours() (section 2.8.6).

This function may be used for both drawing and printing.

3.1.7. clip()
-------------

  void clip(drawing *dr, int x, int y, int w, int h);

Establishes a clipping rectangle in the puzzle window.

`x' and `y' give the coordinates of the top left pixel of the clipping
rectangle. `w' and `h' give its width and height. Thus, the horizontal
extent of the rectangle runs from `x' to `x+w-1' inclusive, and the
vertical extent from `y' to `y+h-1' inclusive. (These are exactly the
same semantics as draw_rect().)

After this call, no drawing operation will affect anything outside the
specified rectangle. The effect can be reversed by calling unclip()
(section 3.1.8).

Back ends should not assume that a clipping rectangle will be
automatically cleared up by the front end if it's left lying around;
that might work on current front ends, but shouldn't be relied upon.
Always explicitly call unclip().

This function may be used for both drawing and printing.

3.1.8. unclip()
---------------

  void unclip(drawing *dr);

Reverts the effect of a previous call to clip(). After this call, all
drawing operations will be able to affect the entire puzzle window
again.

This function may be used for both drawing and printing.

3.1.9. draw_update()
--------------------

  void draw_update(drawing *dr, int x, int y, int w, int h);

Informs the front end that a rectangular portion of the puzzle window
has been drawn on and needs to be updated.

`x' and `y' give the coordinates of the top left pixel of the update
rectangle. `w' and `h' give its width and height. Thus, the horizontal
extent of the rectangle runs from `x' to `x+w-1' inclusive, and the
vertical extent from `y' to `y+h-1' inclusive. (These are exactly the
same semantics as draw_rect().)

The back end redraw function _must_ call this function to report any
changes it has made to the window. Otherwise, those changes may not
become immediately visible, and may then appear at an unpredictable
subsequent time such as the next time the window is covered and re-
exposed.

This function is only important when drawing. It may be called when
printing as well, but doing so is not compulsory, and has no effect.
(So if you have a shared piece of code between the drawing and printing
routines, that code may safely call draw_update().)

3.1.10. status_bar()
--------------------

  void status_bar(drawing *dr, char *text);

Sets the text in the game's status bar to `text'. The text is copied
from the supplied buffer, so the caller is free to deallocate or modify
the buffer after use.

(This function is not exactly a _drawing_ function, but it shares with
the drawing API the property that it may only be called from within the
back end redraw function, so this is as good a place as any to document
it.)

The supplied text is filtered through the mid-end for optional rewriting
before being passed on to the front end; the mid-end will prepend the
current game time if the game is timed (and may in future perform other
rewriting if it seems like a good idea).

This function is for drawing only; it must never be called during
printing.

3.1.11. Blitter functions
-------------------------

This section describes a group of related functions which save and
restore a section of the puzzle window. This is most commonly used to
implement user interfaces involving dragging a puzzle element around the
window: at the end of each call to redraw(), if an object is currently
being dragged, the back end saves the window contents under that
location and then draws the dragged object, and at the start of the next
redraw() the first thing it does is to restore the background.

The front end defines an opaque type called a `blitter', which is
capable of storing a rectangular area of a specified size.

Blitter functions are for drawing only; they must never be called during
printing.

3.1.11.1. blitter_new()
-----------------------

  blitter *blitter_new(drawing *dr, int w, int h);

Creates a new blitter object which stores a rectangle of size `w' by `h'
pixels. Returns a pointer to the blitter object.

Blitter objects are best stored in the `game_drawstate'. A good time to
create them is in the set_size() function (section 2.8.5), since it is
at this point that you first know how big a rectangle they will need to
save.

3.1.11.2. blitter_free()
------------------------

  void blitter_free(drawing *dr, blitter *bl);

Disposes of a blitter object. Best called in free_drawstate(). (However,
check that the blitter object is not NULL before attempting to free it;
it is possible that a draw state might be created and freed without ever
having set_size() called on it in between.)

3.1.11.3. blitter_save()
------------------------

  void blitter_save(drawing *dr, blitter *bl, int x, int y);

This is a true drawing API function, in that it may only be called from
within the game redraw routine. It saves a rectangular portion of the
puzzle window into the specified blitter object.

`x' and `y' give the coordinates of the top left corner of the saved
rectangle. The rectangle's width and height are the ones specified when
the blitter object was created.

This function is required to cope and do the right thing if `x' and `y'
are out of range. (The right thing probably means saving whatever part
of the blitter rectangle overlaps with the visible area of the puzzle
window.)

3.1.11.4. blitter_load()
------------------------

  void blitter_load(drawing *dr, blitter *bl, int x, int y);

This is a true drawing API function, in that it may only be called from
within the game redraw routine. It restores a rectangular portion of the
puzzle window from the specified blitter object.

`x' and `y' give the coordinates of the top left corner of the rectangle
to be restored. The rectangle's width and height are the ones specified
when the blitter object was created.

Alternatively, you can specify both `x' and `y' as the special value
BLITTER_FROMSAVED, in which case the rectangle will be restored to
exactly where it was saved from. (This is probably what you want to do
almost all the time, if you're using blitters to implement draggable
puzzle elements.)

This function is required to cope and do the right thing if `x' and
`y' (or the equivalent ones saved in the blitter) are out of range.
(The right thing probably means restoring whatever part of the blitter
rectangle overlaps with the visible area of the puzzle window.)

If this function is called on a blitter which had previously been saved
from a partially out-of-range rectangle, then the parts of the saved
bitmap which were not visible at save time are undefined. If the blitter
is restored to a different position so as to make those parts visible,
the effect on the drawing area is undefined.

3.1.12. print_mono_colour()
---------------------------

  int print_mono_colour(drawing *dr, int grey);

This function allocates a colour index for a simple monochrome colour
during printing.

`grey' must be 0 or 1. If `grey' is 0, the colour returned is black; if
`grey' is 1, the colour is white.

3.1.13. print_grey_colour()
---------------------------

  int print_grey_colour(drawing *dr, float grey);

This function allocates a colour index for a grey-scale colour during
printing.

`grey' may be any number between 0 (black) and 1 (white); for example,
0.5 indicates a medium grey.

The chosen colour will be rendered to the limits of the printer's
halftoning capability.

3.1.14. print_hatched_colour()
------------------------------

  int print_hatched_colour(drawing *dr, int hatch);

This function allocates a colour index which does not represent a
literal _colour_. Instead, regions shaded in this colour will be hatched
with parallel lines. The `hatch' parameter defines what type of hatching
should be used in place of this colour:

HATCH_SLASH

    This colour will be hatched by lines slanting to the right at 45
    degrees.

HATCH_BACKSLASH

    This colour will be hatched by lines slanting to the left at 45
    degrees.

HATCH_HORIZ

    This colour will be hatched by horizontal lines.

HATCH_VERT

    This colour will be hatched by vertical lines.

HATCH_PLUS

    This colour will be hatched by criss-crossing horizontal and
    vertical lines.

HATCH_X

    This colour will be hatched by criss-crossing diagonal lines.

Colours defined to use hatching may not be used for drawing lines or
text; they may only be used for filling areas. That is, they may be
used as the `fillcolour' parameter to draw_circle() and draw_polygon(),
and as the colour parameter to draw_rect(), but may not be used as the
`outlinecolour' parameter to draw_circle() or draw_polygon(), or with
draw_line() or draw_text().

3.1.15. print_rgb_mono_colour()
-------------------------------

  int print_rgb_mono_colour(drawing *dr, float r, float g,
                            float b, float grey);

This function allocates a colour index for a fully specified RGB colour
during printing.

`r', `g' and `b' may each be anywhere in the range from 0 to 1.

If printing in black and white only, these values will be ignored, and
either pure black or pure white will be used instead, according to the
`grey' parameter. (The fallback colour is the same as the one which
would be allocated by print_mono_colour(grey).)

3.1.16. print_rgb_grey_colour()
-------------------------------

  int print_rgb_grey_colour(drawing *dr, float r, float g,
                            float b, float grey);

This function allocates a colour index for a fully specified RGB colour
during printing.

`r', `g' and `b' may each be anywhere in the range from 0 to 1.

If printing in black and white only, these values will be ignored, and
a shade of grey given by the `grey' parameter will be used instead.
(The fallback colour is the same as the one which would be allocated by
print_grey_colour(grey).)

3.1.17. print_rgb_hatched_colour()
----------------------------------

  int print_rgb_hatched_colour(drawing *dr, float r, float g,
                               float b, float hatched);

This function allocates a colour index for a fully specified RGB colour
during printing.

`r', `g' and `b' may each be anywhere in the range from 0 to 1.

If printing in black and white only, these values will be ignored, and
a form of cross-hatching given by the `hatch' parameter will be used
instead; see section 3.1.14 for the possible values of this parameter.
(The fallback colour is the same as the one which would be allocated by
print_hatched_colour(hatch).)

3.1.18. print_line_width()
--------------------------

  void print_line_width(drawing *dr, int width);

This function is called to set the thickness of lines drawn during
printing. It is meaningless in drawing: all lines drawn by draw_line(),
draw_circle and draw_polygon() are one pixel in thickness. However, in
printing there is no clear definition of a pixel and so line widths must
be explicitly specified.

The line width is specified in the usual coordinate system. Note,
however, that it is a hint only: the central printing system may choose
to vary line thicknesses at user request or due to printer capabilities.

3.2. The drawing API as implemented by the front end
----------------------------------------------------

This section describes the drawing API in the function-pointer form in
which it is implemented by a front end.

(It isn't only platform-specific front ends which implement this API;
the platform-independent module `ps.c' also provides an implementation
of it which outputs PostScript. Thus, any platform which wants to do PS
printing can do so with minimum fuss.)

The following entries all describe function pointer fields in a
structure called `drawing_api'. Each of the functions takes a `void *'
context pointer, which it should internally cast back to a more useful
type. Thus, a drawing _object_ (`drawing *)' suitable for passing to
the back end redraw or printing functions is constructed by passing a
`drawing_api' and a `void *' to the function drawing_new() (see section
3.3.1).

3.2.1. draw_text()
------------------

  void (*draw_text)(void *handle, int x, int y, int fonttype,
                    int fontsize, int align, int colour, char *text);

This function behaves exactly like the back end draw_text() function;
see section 3.1.6.

3.2.2. draw_rect()
------------------

  void (*draw_rect)(void *handle, int x, int y, int w, int h,
                    int colour);

This function behaves exactly like the back end draw_rect() function;
see section 3.1.1.

3.2.3. draw_line()
------------------

  void (*draw_line)(void *handle, int x1, int y1, int x2, int y2,
                    int colour);

This function behaves exactly like the back end draw_line() function;
see section 3.1.3.

3.2.4. draw_polygon()
---------------------

  void (*draw_polygon)(void *handle, int *coords, int npoints,
                       int fillcolour, int outlinecolour);

This function behaves exactly like the back end draw_polygon() function;
see section 3.1.4.

3.2.5. draw_circle()
--------------------

  void (*draw_circle)(void *handle, int cx, int cy, int radius,
                      int fillcolour, int outlinecolour);

This function behaves exactly like the back end draw_circle() function;
see section 3.1.5.

3.2.6. draw_update()
--------------------

  void (*draw_update)(void *handle, int x, int y, int w, int h);

This function behaves exactly like the back end draw_update() function;
see section 3.1.9.

An implementation of this API which only supports printing is permitted
to define this function pointer to be NULL rather than bothering to
define an empty function. The middleware in drawing.c will notice and
avoid calling it.

3.2.7. clip()
-------------

  void (*clip)(void *handle, int x, int y, int w, int h);

This function behaves exactly like the back end clip() function; see
section 3.1.7.

3.2.8. unclip()
---------------

  void (*unclip)(void *handle);

This function behaves exactly like the back end unclip() function; see
section 3.1.8.

3.2.9. start_draw()
-------------------

  void (*start_draw)(void *handle);

This function is called at the start of drawing. It allows the front end
to initialise any temporary data required to draw with, such as device
contexts.

Implementations of this API which do not provide drawing services may
define this function pointer to be NULL; it will never be called unless
drawing is attempted.

3.2.10. end_draw()
------------------

  void (*end_draw)(void *handle);

This function is called at the end of drawing. It allows the front end
to do cleanup tasks such as deallocating device contexts and scheduling
appropriate GUI redraw events.

Implementations of this API which do not provide drawing services may
define this function pointer to be NULL; it will never be called unless
drawing is attempted.

3.2.11. status_bar()
--------------------

  void (*status_bar)(void *handle, char *text);

This function behaves exactly like the back end status_bar() function;
see section 3.1.10.

Front ends implementing this function need not worry about it
being called repeatedly with the same text; the middleware code in
status_bar() will take care of this.

Implementations of this API which do not provide drawing services may
define this function pointer to be NULL; it will never be called unless
drawing is attempted.

3.2.12. blitter_new()
---------------------

  blitter *(*blitter_new)(void *handle, int w, int h);

This function behaves exactly like the back end blitter_new() function;
see section 3.1.11.1.

Implementations of this API which do not provide drawing services may
define this function pointer to be NULL; it will never be called unless
drawing is attempted.

3.2.13. blitter_free()
----------------------

  void (*blitter_free)(void *handle, blitter *bl);

This function behaves exactly like the back end blitter_free() function;
see section 3.1.11.2.

Implementations of this API which do not provide drawing services may
define this function pointer to be NULL; it will never be called unless
drawing is attempted.

3.2.14. blitter_save()
----------------------

  void (*blitter_save)(void *handle, blitter *bl, int x, int y);

This function behaves exactly like the back end blitter_save() function;
see section 3.1.11.3.

Implementations of this API which do not provide drawing services may
define this function pointer to be NULL; it will never be called unless
drawing is attempted.

3.2.15. blitter_load()
----------------------

  void (*blitter_load)(void *handle, blitter *bl, int x, int y);

This function behaves exactly like the back end blitter_load() function;
see section 3.1.11.4.

Implementations of this API which do not provide drawing services may
define this function pointer to be NULL; it will never be called unless
drawing is attempted.

3.2.16. begin_doc()
-------------------

  void (*begin_doc)(void *handle, int pages);

This function is called at the beginning of a printing run. It gives the
front end an opportunity to initialise any required printing subsystem.
It also provides the number of pages in advance.

Implementations of this API which do not provide printing services may
define this function pointer to be NULL; it will never be called unless
printing is attempted.

3.2.17. begin_page()
--------------------

  void (*begin_page)(void *handle, int number);

This function is called during printing, at the beginning of each page.
It gives the page number (numbered from 1 rather than 0, so suitable for
use in user-visible contexts).

Implementations of this API which do not provide printing services may
define this function pointer to be NULL; it will never be called unless
printing is attempted.

3.2.18. begin_puzzle()
----------------------

  void (*begin_puzzle)(void *handle, float xm, float xc,
                       float ym, float yc, int pw, int ph, float wmm);

This function is called during printing, just before printing a single
puzzle on a page. It specifies the size and location of the puzzle on
the page.

`xm' and `xc' specify the horizontal position of the puzzle on the page,
as a linear function of the page width. The front end is expected to
multiply the page width by `xm', add `xc' (measured in millimetres), and
use the resulting x-coordinate as the left edge of the puzzle.

Similarly, `ym' and `yc' specify the vertical position of the puzzle as
a function of the page height: the page height times `ym', plus `yc'
millimetres, equals the desired distance from the top of the page to the
top of the puzzle.

(This unwieldy mechanism is required because not all printing systems
can communicate the page size back to the software. The PostScript back
end, for example, writes out PS which determines the page size at print
time by means of calling `clippath', and centres the puzzles within
that. Thus, exactly the same PS file works on A4 or on US Letter paper
without needing local configuration, which simplifies matters.)

pw and ph give the size of the puzzle in drawing API coordinates. The
printing system will subsequently call the puzzle's own print function,
which will in turn call drawing API functions in the expectation that an
area pw by ph units is available to draw the puzzle on.

Finally, wmm gives the desired width of the puzzle in millimetres. (The
aspect ratio is expected to be preserved, so if the desired puzzle
height is also needed then it can be computed as wmm*ph/pw.)

Implementations of this API which do not provide printing services may
define this function pointer to be NULL; it will never be called unless
printing is attempted.

3.2.19. end_puzzle()
--------------------

  void (*end_puzzle)(void *handle);

This function is called after the printing of a specific puzzle is
complete.

Implementations of this API which do not provide printing services may
define this function pointer to be NULL; it will never be called unless
printing is attempted.

3.2.20. end_page()
------------------

  void (*end_page)(void *handle, int number);

This function is called after the printing of a page is finished.

Implementations of this API which do not provide printing services may
define this function pointer to be NULL; it will never be called unless
printing is attempted.

3.2.21. end_doc()
-----------------

  void (*end_doc)(void *handle);

This function is called after the printing of the entire document is
finished. This is the moment to close files, send things to the print
spooler, or whatever the local convention is.

Implementations of this API which do not provide printing services may
define this function pointer to be NULL; it will never be called unless
printing is attempted.

3.2.22. line_width()
--------------------

  void (*line_width)(void *handle, float width);

This function is called to set the line thickness, during printing only.
Note that the width is a float here, where it was an int as seen by the
back end. This is because drawing.c may have scaled it on the way past.

However, the width is still specified in the same coordinate system as
the rest of the drawing.

Implementations of this API which do not provide printing services may
define this function pointer to be NULL; it will never be called unless
printing is attempted.

3.3. The drawing API as called by the front end
-----------------------------------------------

There are a small number of functions provided in drawing.c which the
front end needs to _call_, rather than helping to implement. They are
described in this section.

3.3.1. drawing_new()
--------------------

  drawing *drawing_new(const drawing_api *api, midend *me,
                       void *handle);

This function creates a drawing object. It is passed a `drawing_api',
which is a structure containing nothing but function pointers; and also
a `void *' handle. The handle is passed back to each function pointer
when it is called.

The `midend' parameter is used for rewriting the status bar contents:
status_bar() (see section 3.1.10) has to call a function in the mid-
end which might rewrite the status bar text. If the drawing object
is to be used only for printing, or if the game is known not to call
status_bar(), this parameter may be NULL.

3.3.2. drawing_free()
---------------------

  void drawing_free(drawing *dr);

This function frees a drawing object. Note that the `void *' handle is
not freed; if that needs cleaning up it must be done by the front end.

3.3.3. print_get_colour()
-------------------------

  void print_get_colour(drawing *dr, int colour, int printincolour,
                        int *hatch, float *r, float *g, float *b)

This function is called by the implementations of the drawing API
functions when they are called in a printing context. It takes a colour
index as input, and returns the description of the colour as requested
by the back end.

`printincolour' is TRUE iff the implementation is printing in colour.
This will alter the results returned if the colour in question was
specified with a black-and-white fallback value.

If the colour should be rendered by hatching, `*hatch' is filled with
the type of hatching desired. See section 3.1.13 for details of the
values this integer can take.

If the colour should be rendered as solid colour, `*hatch' is given a
negative value, and `*r', `*g' and `*b' are filled with the RGB values
of the desired colour (if printing in colour), or all filled with the
grey-scale value (if printing in black and white).

4. The API provided by the mid-end
----------------------------------

This chapter documents the API provided by the mid-end to be called by
the front end. You probably only need to read this if you are a front
end implementor, i.e. you are porting Puzzles to a new platform. If
you're only interested in writing new puzzles, you can safely skip this
chapter.

All the persistent state in the mid-end is encapsulated within a
`midend' structure, to facilitate having multiple mid-ends in any
port which supports multiple puzzle windows open simultaneously. Each
`midend' is intended to handle the contents of a single puzzle window.

4.1. midend_new()
-----------------

  midend *midend_new(frontend *fe, const game *ourgame,
                     const drawing_api *drapi, void *drhandle)

Allocates and returns a new mid-end structure.

The `fe' argument is stored in the mid-end. It will be used when calling
back to functions such as activate_timer() (section 4.28), and will be
passed on to the back end function colours() (section 2.8.6).

The parameters `drapi' and `drhandle' are passed to drawing_new()
(section 3.3.1) to construct a drawing object which will be passed to
the back end function redraw() (section 2.8.9). Hence, all drawing-
related function pointers defined in `drapi' can expect to be called
with `drhandle' as their first argument.

The `ourgame' argument points to a container structure describing a game
back end. The mid-end thus created will only be capable of handling that
one game. (So even in a monolithic front end containing all the games,
this imposes the constraint that any individual puzzle window is tied to
a single game. Unless, of course, you feel brave enough to change the
mid-end for the window without closing the window...)

4.2. midend_free()
------------------

  void midend_free(midend *me);

Frees a mid-end structure and all its associated data.

4.3. midend_set_params()
------------------------

  void midend_set_params(midend *me, game_params *params);

Sets the current game parameters for a mid-end. Subsequent games
generated by midend_new_game() (section 4.6) will use these parameters
until further notice.

The usual way in which the front end will have an actual `game_params'
structure to pass to this function is if it had previously got it from
midend_fetch_preset() (section 4.14). Thus, this function is usually
called in response to the user making a selection from the presets menu.

4.4. midend_get_params()
------------------------

  game_params *midend_get_params(midend *me);

Returns the current game parameters stored in this mid-end.

The returned value is dynamically allocated, and should be freed when
finished with by passing it to the game's own free_params() function
(see section 2.3.5).

4.5. midend_size()
------------------

  void midend_size(midend *me, int *x, int *y, int user_size);

Tells the mid-end to figure out its window size.

On input, `*x' and `*y' should contain the maximum or requested size
for the window. (Typically this will be the size of the screen that the
window has to fit on, or similar.) The mid-end will repeatedly call the
back end function compute_size() (section 2.8.4), searching for a tile
size that best satisfies the requirements. On exit, `*x' and `*y' will
contain the size needed for the puzzle window's drawing area. (It is
of course up to the front end to adjust this for any additional window
furniture such as menu bars and window borders, if necessary. The status
bar is also not included in this size.)

Use `user_size' to indicate whether `*x' and `*y' are a requested size,
or just a maximum size.

If `user_size' is set to TRUE, the mid-end will treat the input size as
a request, and will pick a tile size which approximates it _as closely
as possible_, going over the game's preferred tile size if necessary to
achieve this. The mid-end will also use the resulting tile size as its
preferred one until further notice, on the assumption that this size was
explicitly requested by the user. Use this option if you want your front
end to support dynamic resizing of the puzzle window with automatic
scaling of the puzzle to fit.

If `user_size' is set to FALSE, then the game's tile size will never go
over its preferred one, although it may go under in order to fit within
the maximum bounds specified by `*x' and `*y'. This is the recommended
approach when opening a new window at default size: the game will use
its preferred size unless it has to use a smaller one to fit on the
screen. If the tile size is shrunk for this reason, the change will not
persist; if a smaller grid is subsequently chosen, the tile size will
recover.

The mid-end will try as hard as it can to return a size which is
less than or equal to the input size, in both dimensions. In extreme
circumstances it may fail (if even the lowest possible tile size gives
window dimensions greater than the input), in which case it will return
a size greater than the input size. Front ends should be prepared
for this to happen (i.e. don't crash or fail an assertion), but may
handle it in any way they see fit: by rejecting the game parameters
which caused the problem, by opening a window larger than the screen
regardless of inconvenience, by introducing scroll bars on the window,
by drawing on a large bitmap and scaling it into a smaller window, or by
any other means you can think of. It is likely that when the tile size
is that small the game will be unplayable anyway, so don't put _too_
much effort into handling it creatively.

If your platform has no limit on window size (or if you're planning to
use scroll bars for large puzzles), you can pass dimensions of INT_MAX
as input to this function. You should probably not do that _and_ set the
`user_size' flag, though!

4.6. midend_new_game()
----------------------

  void midend_new_game(midend *me);

Causes the mid-end to begin a new game. Normally the game will be a
new randomly generated puzzle. However, if you have previously called
midend_game_id() or midend_set_config(), the game generated might be
dictated by the results of those functions. (In particular, you _must_
call midend_new_game() after calling either of those functions, or else
no immediate effect will be visible.)

You will probably need to call midend_size() after calling this
function, because if the game parameters have been changed since the
last new game then the window size might need to change. (If you know
the parameters _haven't_ changed, you don't need to do this.)

This function will create a new `game_drawstate', but does not actually
perform a redraw (since you often need to call midend_size() before
the redraw can be done). So after calling this function and after
calling midend_size(), you should then call midend_redraw(). (It is not
necessary to call midend_force_redraw(); that will discard the draw
state and create a fresh one, which is unnecessary in this case since
there's a fresh one already. It would work, but it's usually excessive.)

4.7. midend_restart_game()
--------------------------

  void midend_restart_game(midend *me);

This function causes the current game to be restarted. This is done by
placing a new copy of the original game state on the end of the undo
list (so that an accidental restart can be undone).

This function automatically causes a redraw, i.e. the front end can
expect its drawing API to be called from _within_ a call to this
function.

4.8. midend_force_redraw()
--------------------------

  void midend_force_redraw(midend *me);

Forces a complete redraw of the puzzle window, by means of discarding
the current `game_drawstate' and creating a new one from scratch before
calling the game's redraw() function.

The front end can expect its drawing API to be called from within a call
to this function.

4.9. midend_redraw()
--------------------

  void midend_redraw(midend *me);

Causes a partial redraw of the puzzle window, by means of simply calling
the game's redraw() function. (That is, the only things redrawn will be
things that have changed since the last redraw.)

The front end can expect its drawing API to be called from within a call
to this function.

4.10. midend_process_key()
--------------------------

  int midend_process_key(midend *me, int x, int y, int button);

The front end calls this function to report a mouse or keyboard event.
The parameters `x', `y' and `button' are almost identical to the ones
passed to the back end function interpret_move() (section 2.7.1), except
that the front end is _not_ required to provide the guarantees about
mouse event ordering. The mid-end will sort out multiple simultaneous
button presses and changes of button; the front end's responsibility
is simply to pass on the mouse events it receives as accurately as
possible.

(Some platforms may need to emulate absent mouse buttons by means of
using a modifier key such as Shift with another mouse button. This tends
to mean that if Shift is pressed or released in the middle of a mouse
drag, the mid-end will suddenly stop receiving, say, LEFT_DRAG events
and start receiving RIGHT_DRAGs, with no intervening button release or
press events. This too is something which the mid-end will sort out for
you; the front end has no obligation to maintain sanity in this area.)

The front end _should_, however, always eventually send some kind of
button release. On some platforms this requires special effort: Windows,
for example, requires a call to the system API function SetCapture() in
order to ensure that your window receives a mouse-up event even if the
pointer has left the window by the time the mouse button is released.
On any platform that requires this sort of thing, the front end _is_
responsible for doing it.

Calling this function is very likely to result in calls back to the
front end's drawing API and/or activate_timer() (section 4.28).

4.11. midend_colours()
----------------------

  float *midend_colours(midend *me, int *ncolours);

Returns an array of the colours required by the game, in exactly
the same format as that returned by the back end function colours()
(section 2.8.6). Front ends should call this function rather than
calling the back end's version directly, since the mid-end adds standard
customisation facilities. (At the time of writing, those customisation
facilities are implemented hackily by means of environment variables,
but it's not impossible that they may become more full and formal in
future.)

4.12. midend_timer()
--------------------

  void midend_timer(midend *me, float tplus);

If the mid-end has called activate_timer() (section 4.28) to request
regular callbacks for purposes of animation or timing, this is the
function the front end should call on a regular basis. The argument
`tplus' gives the time, in seconds, since the last time either this
function was called or activate_timer() was invoked.

One of the major purposes of timing in the mid-end is to perform move
animation. Therefore, calling this function is very likely to result in
calls back to the front end's drawing API.

4.13. midend_num_presets()
--------------------------

  int midend_num_presets(midend *me);

Returns the number of game parameter presets supplied by this game.
Front ends should use this function and midend_fetch_preset() to
configure their presets menu rather than calling the back end directly,
since the mid-end adds standard customisation facilities. (At the time
of writing, those customisation facilities are implemented hackily by
means of environment variables, but it's not impossible that they may
become more full and formal in future.)

4.14. midend_fetch_preset()
---------------------------

  void midend_fetch_preset(midend *me, int n,
                           char **name, game_params **params);

Returns one of the preset game parameter structures for the game.
On input `n' must be a non-negative integer and less than the value
returned from midend_num_presets(). On output, `*name' is set to an
ASCII string suitable for entering in the game's presets menu, and
`*params' is set to the corresponding `game_params' structure.

Both of the two output values are dynamically allocated, but they are
owned by the mid-end structure: the front end should not ever free them
directly, because they will be freed automatically during midend_free().

4.15. midend_which_preset()
---------------------------

  int midend_which_preset(midend *me);

Returns the numeric index of the preset game parameter structure which
matches the current game parameters, or a negative number if no preset
matches. Front ends could use this to maintain a tick beside one of the
items in the menu (or tick the `Custom' option if the return value is
less than zero).

4.16. midend_wants_statusbar()
------------------------------

  int midend_wants_statusbar(midend *me);

This function returns TRUE if the puzzle has a use for a textual status
line (to display score, completion status, currently active tiles, time,
or anything else).

Front ends should call this function rather than talking directly to the
back end.

4.17. midend_get_config()
-------------------------

  config_item *midend_get_config(midend *me, int which,
                                 char **wintitle);

Returns a dialog box description for user configuration.

On input, which should be set to one of three values, which select which
of the various dialog box descriptions is returned:

CFG_SETTINGS

    Requests the GUI parameter configuration box generated by the puzzle
    itself. This should be used when the user selects `Custom' from the
    game types menu (or equivalent). The mid-end passes this request on
    to the back end function configure() (section 2.3.8).

CFG_DESC

    Requests a box suitable for entering a descriptive game ID (and
    viewing the existing one). The mid-end generates this dialog box
    description itself. This should be used when the user selects
    `Specific' from the game menu (or equivalent).

CFG_SEED

    Requests a box suitable for entering a random-seed game ID (and
    viewing the existing one). The mid-end generates this dialog box
    description itself. This should be used when the user selects
    `Random Seed' from the game menu (or equivalent).

The returned value is an array of config_items, exactly as described
in section 2.3.8. Another returned value is an ASCII string giving a
suitable title for the configuration window, in `*wintitle'.

Both returned values are dynamically allocated and will need to be
freed. The window title can be freed in the obvious way; the config_item
array is a slightly complex structure, so a utility function free_cfg()
is provided to free it for you. See section 5.2.6.

(Of course, you will probably not want to free the config_item array
until the dialog box is dismissed, because before then you will probably
need to pass it to midend_set_config.)

4.18. midend_set_config()
-------------------------

  char *midend_set_config(midend *me, int which,
                          config_item *cfg);

Passes the mid-end the results of a configuration dialog box. `which'
should have the same value which it had when midend_get_config() was
called; `cfg' should be the array of `config_item's returned from
midend_get_config(), modified to contain the results of the user's
editing operations.

This function returns NULL on success, or otherwise (if the
configuration data was in some way invalid) an ASCII string containing
an error message suitable for showing to the user.

If the function succeeds, it is likely that the game parameters will
have been changed and it is certain that a new game will be requested.
The front end should therefore call midend_new_game(), and probably also
re-think the window size using midend_size() and eventually perform a
refresh using midend_redraw().

4.19. midend_game_id()
----------------------

  char *midend_game_id(midend *me, char *id);

Passes the mid-end a string game ID (of any of the valid forms `params',
`params:description' or `params#seed') which the mid-end will process
and use for the next generated game.

This function returns NULL on success, or otherwise (if the
configuration data was in some way invalid) an ASCII string containing
an error message (not dynamically allocated) suitable for showing to the
user. In the event of an error, the mid-end's internal state will be
left exactly as it was before the call.

If the function succeeds, it is likely that the game parameters will
have been changed and it is certain that a new game will be requested.
The front end should therefore call midend_new_game(), and probably
also re-think the window size using midend_size() and eventually case a
refresh using midend_redraw().

4.20. midend_get_game_id()
--------------------------

  char *midend_get_game_id(midend *me)

Returns a descriptive game ID (i.e. one in the form
`params:description') describing the game currently active in the mid-
end. The returned string is dynamically allocated.

4.21. midend_text_format()
--------------------------

  char *midend_text_format(midend *me);

Formats the current game's current state as ASCII text suitable for
copying to the clipboard. The returned string is dynamically allocated.

You should not call this function if the game's `can_format_as_text'
flag is FALSE.

If the returned string contains multiple lines (which is likely), it
will use the normal C line ending convention (\n only). On platforms
which use a different line ending convention for data in the clipboard,
it is the front end's responsibility to perform the conversion.

4.22. midend_solve()
--------------------

  char *midend_solve(midend *me);

Requests the mid-end to perform a Solve operation.

On success, NULL is returned. On failure, an error message (not
dynamically allocated) is returned, suitable for showing to the user.

The front end can expect its drawing API and/or activate_timer() to be
called from within a call to this function.

4.23. midend_serialise()
------------------------

  void midend_serialise(midend *me,
                        void (*write)(void *ctx, void *buf, int len),
                        void *wctx);

Calling this function causes the mid-end to convert its entire internal
state into a long ASCII text string, and to pass that string (piece by
piece) to the supplied `write' function.

Desktop implementations can use this function to save a game in any
state (including half-finished) to a disk file, by supplying a `write'
function which is a wrapper on fwrite() (or local equivalent). Other
implementations may find other uses for it, such as compressing the
large and sprawling mid-end state into a manageable amount of memory
when a palmtop application is suspended so that another one can run; in
this case write might want to write to a memory buffer rather than a
file. There may be other uses for it as well.

This function will call back to the supplied `write' function a number
of times, with the first parameter (`ctx') equal to `wctx', and the
other two parameters pointing at a piece of the output string.

4.24. midend_deserialise()
--------------------------

  char *midend_deserialise(midend *me,
                           int (*read)(void *ctx, void *buf, int len),
                           void *rctx);

This function is the counterpart to midend_serialise(). It calls the
supplied read function repeatedly to read a quantity of data, and
attempts to interpret that data as a serialised mid-end as output by
midend_serialise().

The read function is called with the first parameter (`ctx') equal
to `rctx', and should attempt to read `len' bytes of data into the
buffer pointed to by `buf'. It should return FALSE on failure or TRUE
on success. It should not report success unless it has filled the
entire buffer; on platforms which might be reading from a pipe or other
blocking data source, `read' is responsible for looping until the whole
buffer has been filled.

If the de-serialisation operation is successful, the mid-end's internal
data structures will be replaced by the results of the load, and NULL
will be returned. Otherwise, the mid-end's state will be completely
unchanged and an error message (typically some variation on `save file
is corrupt') will be returned. As usual, the error message string is not
dynamically allocated.

If this function succeeds, it is likely that the game parameters will
have been changed. The front end should therefore probably re-think the
window size using midend_size(), and probably cause a refresh using
midend_redraw().

Because each mid-end is tied to a specific game back end, this function
will fail if you attempt to read in a save file generated by a different
game from the one configured in this mid-end, even if your application
is a monolithic one containing all the puzzles. (It would be pretty easy
to write a function which would look at a save file and determine which
game it was for; any front end implementor who needs such a function can
probably be accommodated.)

4.25. Direct reference to the back end structure by the front end
-----------------------------------------------------------------

Although _most_ things the front end needs done should be done by
calling the mid-end, there are a few situations in which the front end
needs to refer directly to the game back end structure.

The most obvious of these is

 -  passing the game back end as a parameter to midend_new().

There are a few other back end features which are not wrapped by the
mid-end because there didn't seem much point in doing so:

 -  fetching the `name' field to use in window titles and similar

 -  reading the `can_configure', `can_solve' and `can_format_as_text'
    fields to decide whether to add those items to the menu bar or
    equivalent

 -  reading the `winhelp_topic' field (Windows only)

 -  the GTK front end provides a `--generate' command-line option which
    directly calls the back end to do most of its work. This is not
    really part of the main front end code, though, and I'm not sure it
    counts.

In order to find the game back end structure, the front end does one of
two things:

 -  If the particular front end is compiling a separate binary per game,
    then the back end structure is a global variable with the standard
    name `thegame':

      extern const game thegame;

 -  If the front end is compiled as a monolithic application containing
    all the puzzles together (in which case the preprocessor symbol
    COMBINED must be defined when compiling most of the code base), then
    there will be two global variables defined:

      extern const game *gamelist[];
      extern const int gamecount;

    `gamelist' will be an array of `gamecount' game structures, declared
    in the automatically constructed source module `list.c'. The
    application should search that array for the game it wants, probably
    by reaching into each game structure and looking at its `name'
    field.

4.26. Mid-end to front-end calls
--------------------------------

This section describes the small number of functions which a front end
must provide to be called by the mid-end or other standard utility
modules.

4.27. get_random_seed()
-----------------------

  void get_random_seed(void **randseed, int *randseedsize);

This function is called by a new mid-end, and also occasionally by game
back ends. Its job is to return a piece of data suitable for using as a
seed for initialisation of a new `random_state'.

On exit, `*randseed' should be set to point at a newly allocated piece
of memory containing some seed data, and `*randseedsize' should be set
to the length of that data.

A simple and entirely adequate implementation is to return a piece of
data containing the current system time at the highest conveniently
available resolution.

4.28. activate_timer()
----------------------

  void activate_timer(frontend *fe);

This is called by the mid-end to request that the front end begin
calling it back at regular intervals.

The timeout interval is left up to the front end; the finer it is, the
smoother move animations will be, but the more CPU time will be used.
Current front ends use values around 20ms (i.e. 50Hz).

After this function is called, the mid-end will expect to receive calls
to midend_timer() on a regular basis.

4.29. deactivate_timer()
------------------------

  void deactivate_timer(frontend *fe);

This is called by the mid-end to request that the front end stop calling
midend_timer().

4.30. fatal()
-------------

  void fatal(char *fmt, ...);

This is called by some utility functions if they encounter a genuinely
fatal error such as running out of memory. It is a variadic function
in the style of printf(), and is expected to show the formatted error
message to the user any way it can and then terminate the application.
It must not return.

4.31. frontend_default_colour()
-------------------------------

  void frontend_default_colour(frontend *fe, float *output);

This function expects to be passed a pointer to an array of three
floats. It returns the platform's local preferred background colour
in those three floats, as red, green and blue values (in that order)
ranging from 0.0 to 1.0.

This function should only ever be called by the back end function
colours() (section 2.8.6). (Thus, it isn't a _midend_-to-frontend
function as such, but there didn't seem to be anywhere else particularly
good to put it. Sorry.)

5. Utility APIs
---------------

This chapter documents a variety of utility APIs provided for the
general use of the rest of the Puzzles code.

5.1. Random number generation
-----------------------------

Platforms' local random number generators vary widely in quality and
seed size. Puzzles therefore supplies its own high-quality random number
generator, with the additional advantage of giving the same results if
fed the same seed data on different platforms. This allows game random
seeds to be exchanged between different ports of Puzzles and still
generate the same games.

Unlike the ANSI C rand() function, the Puzzles random number generator
has an _explicit_ state object called a `random_state'. One of these
is managed by each mid-end, for example, and passed to the back end to
generate a game with.

5.1.1. random_new()
-------------------

  random_state *random_new(char *seed, int len);

Allocates, initialises and returns a new `random_state'. The input data
is used as the seed for the random number stream (i.e. using the same
seed at a later time will generate the same stream).

The seed data can be any data at all; there is no requirement to use
printable ASCII, or NUL-terminated strings, or anything like that.

5.1.2. random_copy()
--------------------

  random_state *random_copy(random_state *tocopy);

Allocates a new `random_state', copies the contents of another
`random_state' into it, and returns the new state. If exactly the
same sequence of functions is subseqently called on both the copy and
the original, the results will be identical. This may be useful for
speculatively performing some operation using a given random state, and
later replaying that operation precisely.

5.1.3. random_free()
--------------------

  void random_free(random_state *state);

Frees a `random_state'.

5.1.4. random_bits()
--------------------

  unsigned long random_bits(random_state *state, int bits);

Returns a random number from 0 to 2^bits-1 inclusive. `bits' should be
between 1 and 32 inclusive.

5.1.5. random_upto()
--------------------

  unsigned long random_upto(random_state *state, unsigned long limit);

Returns a random number from 0 to limit-1 inclusive.

5.1.6. random_state_encode()
----------------------------

  char *random_state_encode(random_state *state);

Encodes the entire contents of a `random_state' in printable ASCII.
Returns a dynamically allocated string containing that encoding. This
can subsequently be passed to random_state_decode() to reconstruct the
same `random_state'.

5.1.7. random_state_decode()
----------------------------

  random_state *random_state_decode(char *input);

Decodes a string generated by random_state_encode() and reconstructs an
equivalent `random_state' to the one encoded, i.e. it should produce the
same stream of random numbers.

This function has no error reporting; if you pass it an invalid string
it will simply generate an arbitrary random state, which may turn out to
be noticeably non-random.

5.1.8. shuffle()
----------------

  void shuffle(void *array, int nelts, int eltsize, random_state *rs);

Shuffles an array into a random order. The interface is much like ANSI C
qsort(), except that there's no need for a compare function.

`array' is a pointer to the first element of the array. `nelts' is the
number of elements in the array; `eltsize' is the size of a single
element (typically measured using `sizeof'). `rs' is a `random_state'
used to generate all the random numbers for the shuffling process.

5.2. Memory allocation
----------------------

Puzzles has some central wrappers on the standard memory allocation
functions, which provide compile-time type checking, and run-time error
checking by means of quitting the application if it runs out of memory.
This doesn't provide the best possible recovery from memory shortage,
but on the other hand it greatly simplifies the rest of the code,
because nothing else anywhere needs to worry about NULL returns from
allocation.

5.2.1. snew()
-------------

  var = snew(type);

This macro takes a single argument which is a _type name_. It allocates
space for one object of that type. If allocation fails it will call
fatal() and not return; so if it does return, you can be confident that
its return value is non-NULL.

The return value is cast to the specified type, so that the compiler
will type-check it against the variable you assign it into. Thus, this
ensures you don't accidentally allocate memory the size of the wrong
type and assign it into a variable of the right one (or vice versa!).

5.2.2. snewn()
--------------

  var = snewn(n, type);

This macro is the array form of snew(). It takes two arguments; the
first is a number, and the second is a type name. It allocates space
for that many objects of that type, and returns a type-checked non-NULL
pointer just as snew() does.

5.2.3. sresize()
----------------

  var = sresize(var, n, type);

This macro is a type-checked form of realloc(). It takes three
arguments: an input memory block, a new size in elements, and a type.
It re-sizes the input memory block to a size sufficient to contain that
many elements of that type. It returns a type-checked non-NULL pointer,
like snew() and snewn().

The input memory block can be NULL, in which case this function will
behave exactly like snewn(). (In principle any ANSI-compliant realloc()
implementation ought to cope with this, but I've never quite trusted it
to work everywhere.)

5.2.4. sfree()
--------------

  void sfree(void *p);

This function is pretty much equivalent to free(). It is provided with a
dynamically allocated block, and frees it.

The input memory block can be NULL, in which case this function will do
nothing. (In principle any ANSI-compliant free() implementation ought to
cope with this, but I've never quite trusted it to work everywhere.)

5.2.5. dupstr()
---------------

  char *dupstr(const char *s);

This function dynamically allocates a duplicate of a C string. Like the
snew() functions, it guarantees to return non-NULL or not return at all.

(Many platforms provide the function strdup(). As well as guaranteeing
never to return NULL, my version has the advantage of being defined
_everywhere_, rather than inconveniently not quite everywhere.)

5.2.6. free_cfg()
-----------------

  void free_cfg(config_item *cfg);

This function correctly frees an array of `config_item's, including
walking the array until it gets to the end and freeing precisely those
`sval' fields which are expected to be dynamically allocated.

(See section 2.3.8 for details of the `config_item' structure.)

5.3. Sorted and counted tree functions
--------------------------------------

Many games require complex algorithms for generating random puzzles, and
some require moderately complex algorithms even during play. A common
requirement during these algorithms is for a means of maintaining sorted
or unsorted lists of items, such that items can be removed and added
conveniently.

For general use, Puzzles provides the following set of functions which
maintain 2-3-4 trees in memory. (A 2-3-4 tree is a balanced tree
structure, with the property that all lookups, insertions, deletions,
splits and joins can be done in O(log N) time.)

All these functions expect you to be storing a tree of `void *'
pointers. You can put anything you like in those pointers.

By the use of per-node element counts, these tree structures have the
slightly unusual ability to look elements up by their numeric index
within the list represented by the tree. This means that they can be
used to store an unsorted list (in which case, every time you insert a
new element, you must explicitly specify the position where you wish to
insert it). They can also do numeric lookups in a sorted tree, which
might be useful for (for example) tracking the median of a changing data
set.

As well as storing sorted lists, these functions can be used for storing
`maps' (associative arrays), by defining each element of a tree to be a
(key, value) pair.

5.3.1. newtree234()
-------------------

  tree234 *newtree234(cmpfn234 cmp);

Creates a new empty tree, and returns a pointer to it.

The parameter `cmp' determines the sorting criterion on the tree. Its
prototype is

  typedef int (*cmpfn234)(void *, void *);

If you want a sorted tree, you should provide a function matching this
prototype, which returns like strcmp() does (negative if the first
argument is smaller than the second, positive if it is bigger, zero if
they compare equal). In this case, the function addpos234() will not be
usable on your tree (because all insertions must respect the sorting
order).

If you want an unsorted tree, pass NULL. In this case you will not be
able to use either add234() or del234(), or any other function such
as find234() which depends on a sorting order. Your tree will become
something more like an array, except that it will efficiently support
insertion and deletion as well as lookups by numeric index.

5.3.2. freetree234()
--------------------

  void freetree234(tree234 *t);

Frees a tree. This function will not free the _elements_ of the tree
(because they might not be dynamically allocated, or you might be
storing the same set of elements in more than one tree); it will just
free the tree structure itself. If you want to free all the elements of
a tree, you should empty it before passing it to freetree234(), by means
of code along the lines of

  while ((element = delpos234(tree, 0)) != NULL)
      sfree(element); /* or some more complicated free function */

5.3.3. add234()
---------------

  void *add234(tree234 *t, void *e);

Inserts a new element `e' into the tree `t'. This function expects the
tree to be sorted; the new element is inserted according to the sort
order.

If an element comparing equal to `e' is already in the tree, then the
insertion will fail, and the return value will be the existing element.
Otherwise, the insertion succeeds, and `e' is returned.

5.3.4. addpos234()
------------------

  void *addpos234(tree234 *t, void *e, int index);

Inserts a new element into an unsorted tree. Since there is no sorting
order to dictate where the new element goes, you must specify where you
want it to go. Setting `index' to zero puts the new element right at the
start of the list; setting `index' to the current number of elements in
the tree puts the new element at the end.

Return value is `e', in line with add234() (although this function
cannot fail except by running out of memory, in which case it will bomb
out and die rather than returning an error indication).

5.3.5. index234()
-----------------

  void *index234(tree234 *t, int index);

Returns a pointer to the `index'th element of the tree, or NULL if
`index' is out of range. Elements of the tree are numbered from zero.

5.3.6. find234()
----------------

  void *find234(tree234 *t, void *e, cmpfn234 cmp);

Searches for an element comparing equal to `e' in a sorted tree.

If `cmp' is NULL, the tree's ordinary comparison function will be used
to perform the search. However, sometimes you don't want that; suppose,
for example, each of your elements is a big structure containing a
`char *' name field, and you want to find the element with a given name.
You _could_ achieve this by constructing a fake element structure,
setting its name field appropriately, and passing it to find234(),
but you might find it more convenient to pass _just_ a name string to
find234(), supplying an alternative comparison function which expects
one of its arguments to be a bare name and the other to be a large
structure containing a name field.

Therefore, if `cmp' is not NULL, then it will be used to compare `e' to
elements of the tree. The first argument passed to `cmp' will always be
`e'; the second will be an element of the tree.

(See section 5.3.1 for the definition of the `cmpfn234' function pointer
type.)

The returned value is the element found, or NULL if the search is
unsuccessful.

5.3.7. findrel234()
-------------------

  void *findrel234(tree234 *t, void *e, cmpfn234 cmp, int relation);

This function is like find234(), but has the additional ability to do a
_relative_ search. The additional parameter `relation' can be one of the
following values:

REL234_EQ

    Find only an element that compares equal to `e'. This is exactly the
    behaviour of find234().

REL234_LT

    Find the greatest element that compares strictly less than `e'. `e'
    may be NULL, in which case it finds the greatest element in the
    whole tree (which could also be done by index234(t, count234(t)-1)).

REL234_LE

    Find the greatest element that compares less than or equal to `e'.
    (That is, find an element that compares equal to `e' if possible,
    but failing that settle for something just less than it.)

REL234_GT

    Find the smallest element that compares strictly greater than `e'.
    `e' may be NULL, in which case it finds the smallest element in the
    whole tree (which could also be done by index234(t, 0)).

REL234_GE

    Find the smallest element that compares greater than or equal
    to `e'. (That is, find an element that compares equal to `e' if
    possible, but failing that settle for something just bigger than
    it.)

Return value, as before, is the element found or NULL if no element
satisfied the search criterion.

5.3.8. findpos234()
-------------------

  void *findpos234(tree234 *t, void *e, cmpfn234 cmp, int *index);

This function is like find234(), but has the additional feature of
returning the index of the element found in the tree; that index is
written to `*index' in the event of a successful search (a non-NULL
return value).

`index' may be NULL, in which case this function behaves exactly like
find234().

5.3.9. findrelpos234()
----------------------

  void *findrelpos234(tree234 *t, void *e, cmpfn234 cmp, int relation,
                      int *index);

This function combines all the features of findrel234() and
findpos234().

5.3.10. del234()
----------------

  void *del234(tree234 *t, void *e);

Finds an element comparing equal to `e' in the tree, deletes it, and
returns it.

The input tree must be sorted.

The element found might be `e' itself, or might merely compare equal to
it.

Return value is NULL if no such element is found.

5.3.11. delpos234()
-------------------

  void *delpos234(tree234 *t, int index);

Deletes the element at position `index' in the tree, and returns it.

Return value is NULL if the index is out of range.

5.3.12. count234()
------------------

  int count234(tree234 *t);

Returns the number of elements currently in the tree.

5.3.13. splitpos234()
---------------------

  tree234 *splitpos234(tree234 *t, int index, int before);

Splits the input tree into two pieces at a given position, and creates a
new tree containing all the elements on one side of that position.

If `before' is TRUE, then all the items at or after position `index' are
left in the input tree, and the items before that point are returned in
the new tree. Otherwise, the reverse happens: all the items at or after
`index' are moved into the new tree, and those before that point are
left in the old one.

If `index' is equal to 0 or to the number of elements in the input tree,
then one of the two trees will end up empty (and this is not an error
condition). If `index' is further out of range in either direction, the
operation will fail completely and return NULL.

This operation completes in O(log N) time, no matter how large the tree
or how balanced or unbalanced the split.

5.3.14. split234()
------------------

  tree234 *split234(tree234 *t, void *e, cmpfn234 cmp, int rel);

Splits a sorted tree according to its sort order.

`rel' can be any of the relation constants described in section 5.3.7,
_except_ for REL234_EQ. All the elements having that relation to `e'
will be transferred into the new tree; the rest will be left in the old
one.

The parameter `cmp' has the same semantics as it does in find234(): if
it is not NULL, it will be used in place of the tree's own comparison
function when comparing elements to `e', in such a way that `e' itself
is always the first of its two operands.

Again, this operation completes in O(log N) time, no matter how large
the tree or how balanced or unbalanced the split.

5.3.15. join234()
-----------------

  tree234 *join234(tree234 *t1, tree234 *t2);

Joins two trees together by concatenating the lists they represent. All
the elements of `t2' are moved into `t1', in such a way that they appear
_after_ the elements of `t1'. The tree `t2' is freed; the return value
is `t1'.

If you apply this function to a sorted tree and it violates the sort
order (i.e. the smallest element in `t2' is smaller than or equal to the
largest element in `t1'), the operation will fail and return NULL.

This operation completes in O(log N) time, no matter how large the trees
being joined together.

5.3.16. join234r()
------------------

  tree234 *join234r(tree234 *t1, tree234 *t2);

Joins two trees together in exactly the same way as join234(), but this
time the combined tree is returned in `t2', and `t1' is destroyed. The
elements in `t1' still appear before those in `t2'.

Again, this operation completes in O(log N) time, no matter how large
the trees being joined together.

5.3.17. copytree234()
---------------------

  tree234 *copytree234(tree234 *t, copyfn234 copyfn,
                       void *copyfnstate);

Makes a copy of an entire tree.

If `copyfn' is NULL, the tree will be copied but the elements will not
be; i.e. the new tree will contain pointers to exactly the same physical
elements as the old one.

If you want to copy each actual element during the operation, you can
instead pass a function in `copyfn' which makes a copy of each element.
That function has the prototype

  typedef void *(*copyfn234)(void *state, void *element);

and every time it is called, the `state' parameter will be set to the
value you passed in as `copyfnstate'.

5.4. Miscellaneous utility functions and macros
-----------------------------------------------

This section contains all the utility functions which didn't sensibly
fit anywhere else.

5.4.1. TRUE and FALSE
---------------------

The main Puzzles header file defines the macros TRUE and FALSE, which
are used throughout the code in place of 1 and 0 (respectively) to
indicate that the values are in a boolean context. For code base
consistency, I'd prefer it if submissions of new code followed this
convention as well.

5.4.2. max() and min()
----------------------

The main Puzzles header file defines the pretty standard macros max()
and min(), each of which is given two arguments and returns the one
which compares greater or less respectively.

These macros may evaluate their arguments multiple times. Avoid side
effects.

5.4.3. PI
---------

The main Puzzles header file defines a macro PI which expands to a
floating-point constant representing pi.

(I've never understood why ANSI's <math.h> doesn't define this. It'd be
so useful!)

5.4.4. obfuscate_bitmap()
-------------------------

  void obfuscate_bitmap(unsigned char *bmp, int bits, int decode);

This function obscures the contents of a piece of data, by cryptographic
methods. It is useful for games of hidden information (such as Mines,
Guess or Black Box), in which the game ID theoretically reveals all the
information the player is supposed to be trying to guess. So in order
that players should be able to send game IDs to one another without
accidentally spoiling the resulting game by looking at them, these games
obfuscate their game IDs using this function.

Although the obfuscation function is cryptographic, it cannot properly
be called encryption because it has no key. Therefore, anybody motivated
enough can re-implement it, or hack it out of the Puzzles source,
and strip the obfuscation off one of these game IDs to see what lies
beneath. (Indeed, they could usually do it much more easily than that,
by entering the game ID into their own copy of the puzzle and hitting
Solve.) The aim is not to protect against a determined attacker; the aim
is simply to protect people who wanted to play the game honestly from
_accidentally_ spoiling their own fun.

The input argument `bmp' points at a piece of memory to be obfuscated.
`bits' gives the length of the data. Note that that length is in _bits_
rather than bytes: if you ask for obfuscation of a partial number of
bytes, then you will get it. Bytes are considered to be used from the
top down: thus, for example, setting `bits' to 10 will cover the whole
of bmp[0] and the _top two_ bits of bmp[1]. The remainder of a partially
used byte is undefined (i.e. it may be corrupted by the function).

The parameter `decode' is FALSE for an encoding operation, and TRUE
for a decoding operation. Each is the inverse of the other. (There's
no particular reason you shouldn't obfuscate by decoding and restore
cleartext by encoding, if you really wanted to; it should still work.)

The input bitmap is processed in place.

5.4.5. bin2hex()
----------------

  char *bin2hex(const unsigned char *in, int inlen);

This function takes an input byte array and converts it into an
ASCII string encoding those bytes in (lower-case) hex. It returns a
dynamically allocated string containing that encoding.

This function is useful for encoding the result of obfuscate_bitmap() in
printable ASCII for use in game IDs.

5.4.6. hex2bin()
----------------

  unsigned char *hex2bin(const char *in, int outlen);

This function takes an ASCII string containing hex digits, and converts
it back into a byte array of length `outlen'. If there aren't enough
hex digits in the string, the contents of the resulting array will be
undefined.

This function is the inverse of bin2hex().

5.4.7. game_mkhighlight()
-------------------------

  void game_mkhighlight(frontend *fe, float *ret,
                        int background, int highlight, int lowlight);

It's reasonably common for a puzzle game's graphics to use highlights
and lowlights to indicate `raised' or `lowered' sections. Fifteen,
Sixteen and Twiddle are good examples of this.

Puzzles using this graphical style are running a risk if they just use
whatever background colour is supplied to them by the front end, because
that background colour might be too light to see any highlights on at
all. (In particular, it's not unheard of for the front end to specify a
default background colour of white.)

Therefore, such puzzles can call this utility function from their
colours() routine (section 2.8.6). You pass it your front end handle, a
pointer to the start of your return array, and three colour indices. It
will:

 -  call frontend_default_colour() (section 4.31) to fetch the front
    end's default background colour

 -  alter the brightness of that colour if it's unsuitable

 -  define brighter and darker variants of the colour to be used as
    highlights and lowlights

 -  write those results into the relevant positions in the `ret' array.

Thus, ret[background*3] to ret[background*3+2] will be set to RGB values
defining a sensible background colour, and similary `highlight' and
`lowlight' will be set to sensible colours.

6. How to write a new puzzle
----------------------------

This chapter gives a guide to how to actually write a new puzzle: where
to start, what to do first, how to solve common problems.

The previous chapters have been largely composed of facts. This one is
mostly advice.

6.1. Choosing a puzzle
----------------------

Before you start writing a puzzle, you have to choose one. Your taste
in puzzle games is up to you, of course; and, in fact, you're probably
reading this guide because you've _already_ thought of a game you want
to write. But if you want to get it accepted into the official Puzzles
distribution, then there's a criterion it has to meet.

The current Puzzles editorial policy is that all games should be _fair_.
A fair game is one which a player can only fail to complete through
demonstrable lack of skill - that is, such that a better player in the
same situation would have _known_ to do something different.

For a start, that means every game presented to the user must have _at
least one solution_. Giving the unsuspecting user a puzzle which is
actually impossible is not acceptable. (There is an exception: if the
user has selected some non-default option which is clearly labelled as
potentially unfair, _then_ you're allowed to generate possibly insoluble
puzzles, because the user isn't unsuspecting any more. Same Game and
Mines both have options of this type.)

Also, this actually _rules out_ games such as Klondike, or the normal
form of Mahjong Solitaire. Those games have the property that even if
there is a solution (i.e. some sequence of moves which will get from
the start state to the solved state), the player doesn't necessarily
have enough information to _find_ that solution. In both games, it is
possible to reach a dead end because you had an arbitrary choice to make
and made it the wrong way. This violates the fairness criterion, because
a better player couldn't have known they needed to make the other
choice.

(GNOME has a variant on Mahjong Solitaire which makes it fair: there
is a Shuffle operation which randomly permutes all the remaining tiles
without changing their positions, which allows you to get out of a
sticky situation. Using this operation adds a 60-second penalty to your
solution time, so it's to the player's advantage to try to minimise
the chance of having to use it. It's still possible to render the game
uncompletable if you end up with only two tiles vertically stacked,
but that's easy to foresee and avoid using a shuffle operation. This
form of the game _is_ fair. Implementing it in Puzzles would require
an infrastructure change so that the back end could communicate time
penalties to the mid-end, but that would be easy enough.)

Providing a _unique_ solution is a little more negotiable; it depends
on the puzzle. Solo would have been of unacceptably low quality if it
didn't always have a unique solution, whereas Twiddle inherently has
multiple solutions by its very nature and it would have been meaningless
to even _suggest_ making it uniquely soluble. Somewhere in between, Flip
could reasonably be made to have unique solutions (by enforcing a zero-
dimension kernel in every generated matrix) but it doesn't seem like a
serious quality problem that it doesn't.

Of course, you don't _have_ to care about all this. There's nothing
stopping you implementing any puzzle you want to if you're happy to
maintain your puzzle yourself, distribute it from your own web site,
fork the Puzzles code completely, or anything like that. It's free
software; you can do what you like with it. But any game that you want
to be accepted into _my_ Puzzles code base has to satisfy the fairness
criterion, which means all randomly generated puzzles must have a
solution (unless the user has deliberately chosen otherwise) and it must
be possible _in theory_ to find that solution without having to guess.

6.2. Getting started
--------------------

The simplest way to start writing a new puzzle is to copy `nullgame.c'.
This is a template puzzle source file which does almost nothing, but
which contains all the back end function prototypes and declares the
back end data structure correctly. It is built every time the rest of
Puzzles is built, to ensure that it doesn't get out of sync with the
code and remains buildable.

So start by copying `nullgame.c' into your new source file. Then you'll
gradually add functionality until the very boring Null Game turns into
your real game.

Next you'll need to add your puzzle to the Makefiles, in order to
compile it conveniently. _Do not edit the Makefiles_: they are created
automatically by the script `mkfiles.pl', from the file called `Recipe'.
Edit `Recipe', and then re-run `mkfiles.pl'.

Also, don't forget to add your puzzle to `list.c': if you don't, then it
will still run fine on platforms which build each puzzle separately, but
Mac OS X and other monolithic platforms will not include your new puzzle
in their single binary.

Once your source file is building, you can move on to the fun bit.

6.2.1. Puzzle generation
------------------------

Randomly generating instances of your puzzle is almost certain to be
the most difficult part of the code, and also the task with the highest
chance of turning out to be completely infeasible. Therefore I strongly
recommend doing it _first_, so that if it all goes horribly wrong you
haven't wasted any more time than you absolutely had to. What I usually
do is to take an unmodified `nullgame.c', and start adding code to
new_game_desc() which tries to generate a puzzle instance and print it
out using printf(). Once that's working, _then_ I start connecting it up
to the return value of new_game_desc(), populating other structures like
`game_params', and generally writing the rest of the source file.

There are many ways to generate a puzzle which is known to be soluble.
In this section I list all the methods I currently know of, in case any
of them can be applied to your puzzle. (Not all of these methods will
work, or in some cases even make sense, for all puzzles.)

Some puzzles are mathematically tractable, meaning you can work out in
advance which instances are soluble. Sixteen, for example, has a parity
constraint in some settings which renders exactly half the game space
unreachable, but it can be mathematically proved that any position
not in that half _is_ reachable. Therefore, Sixteen's grid generation
simply consists of selecting at random from a well defined subset of the
game space. Cube in its default state is even easier: _every_ possible
arrangement of the blue squares and the cube's starting position is
soluble!

Another option is to redefine what you mean by `soluble'. Black Box
takes this approach. There are layouts of balls in the box which are
completely indistinguishable from one another no matter how many beams
you fire into the box from which angles, which would normally be grounds
for declaring those layouts unfair; but fortunately, detecting that
indistinguishability is computationally easy. So Black Box doesn't
demand that your ball placements match its own; it merely demands
that your ball placements be _indistinguishable_ from the ones it was
thinking of. If you have an ambiguous puzzle, then any of the possible
answers is considered to be a solution. Having redefined the rules in
that way, any puzzle is soluble again.

Those are the simple techniques. If they don't work, you have to get
cleverer.

One way to generate a soluble puzzle is to start from the solved state
and make inverse moves until you reach a starting state. Then you know
there's a solution, because you can just list the inverse moves you made
and make them in the opposite order to return to the solved state.

This method can be simple and effective for puzzles where you get to
decide what's a starting state and what's not. In Pegs, for example,
the generator begins with one peg in the centre of the board and makes
inverse moves until it gets bored; in this puzzle, valid inverse moves
are easy to detect, and _any_ state that's reachable from the solved
state by inverse moves is a reasonable starting position. So Pegs just
continues making inverse moves until the board satisfies some criteria
about extent and density, and then stops and declares itself done.

For other puzzles, it can be a lot more difficult. Same Game uses
this strategy too, and it's lucky to get away with it at all: valid
inverse moves aren't easy to find (because although it's easy to insert
additional squares in a Same Game position, it's difficult to arrange
that _after_ the insertion they aren't adjacent to any other squares of
the same colour), so you're constantly at risk of running out of options
and having to backtrack or start again. Also, Same Game grids never
start off half-empty, which means you can't just stop when you run out
of moves - you have to find a way to fill the grid up _completely_.

The other way to generate a puzzle that's soluble is to start from the
other end, and actually write a _solver_. This tends to ensure that a
puzzle has a _unique_ solution over and above having a solution at all,
so it's a good technique to apply to puzzles for which that's important.

One theoretical drawback of generating soluble puzzles by using a solver
is that your puzzles are restricted in difficulty to those which the
solver can handle. (Most solvers are not fully general: many sets of
puzzle rules are NP-complete or otherwise nasty, so most solvers can
only handle a subset of the theoretically soluble puzzles.) It's been
my experience in practice, however, that this usually isn't a problem;
computers are good at very different things from humans, and what the
computer thinks is nice and easy might still be pleasantly challenging
for a human. For example, when solving Dominosa puzzles I frequently
find myself using a variety of reasoning techniques that my solver
doesn't know about; in principle, therefore, I should be able to solve
the puzzle using only those techniques it _does_ know about, but this
would involve repeatedly searching the entire grid for the one simple
deduction I can make. Computers are good at this sort of exhaustive
search, but it's been my experience that human solvers prefer to do more
complex deductions than to spend ages searching for simple ones. So in
many cases I don't find my own playing experience to be limited by the
restrictions on the solver.

(This isn't _always_ the case. Solo is a counter-example; generating
Solo puzzles using a simple solver does lead to qualitatively easier
puzzles. Therefore I had to make the Solo solver rather more advanced
than most of them.)

There are several different ways to apply a solver to the problem of
generating a soluble puzzle. I list a few of them below.

The simplest approach is brute force: randomly generate a puzzle, use
the solver to see if it's soluble, and if not, throw it away and try
again until you get lucky. This is often a viable technique if all
else fails, but it tends not to scale well: for many puzzle types, the
probability of finding a uniquely soluble instance decreases sharply
as puzzle size goes up, so this technique might work reasonably fast
for small puzzles but take (almost) forever at larger sizes. Still, if
there's no other alternative it can be usable: Pattern and Dominosa
both use this technique. (However, Dominosa has a means of tweaking the
randomly generated grids to increase the _probability_ of them being
soluble, by ruling out one of the most common ambiguous cases. This
improved generation speed by over a factor of 10 on the highest preset!)

An approach which can be more scalable involves generating a grid and
then tweaking it to make it soluble. This is the technique used by Mines
and also by Net: first a random puzzle is generated, and then the solver
is run to see how far it gets. Sometimes the solver will get stuck;
when that happens, examine the area it's having trouble with, and make
a small random change in that area to allow it to make more progress.
Continue solving (possibly even without restarting the solver), tweaking
as necessary, until the solver finishes. Then restart the solver from
the beginning to ensure that the tweaks haven't caused new problems in
the process of solving old ones (which can sometimes happen).

This strategy works well in situations where the usual solver failure
mode is to get stuck in an easily localised spot. Thus it works well
for Net and Mines, whose most common failure mode tends to be that most
of the grid is fine but there are a few widely separated ambiguous
sections; but it would work less well for Dominosa, in which the way you
get stuck is to have scoured the whole grid and not found anything you
can deduce _anywhere_. Also, it relies on there being a low probability
that tweaking the grid introduces a new problem at the same time as
solving the old one; Mines and Net also have the property that most of
their deductions are local, so that it's very unlikely for a tweak to
affect something half way across the grid from the location where it was
applied. In Dominosa, by contrast, a lot of deductions use information
about half the grid (`out of all the sixes, only one is next to a
three', which can depend on the values of up to 32 of the 56 squares in
the default setting!), so this tweaking strategy would be rather less
likely to work well.

A more specialised strategy is that used in Solo and Slant. These
puzzles have the property that they derive their difficulty from not
presenting all the available clues. (In Solo's case, if all the possible
clues were provided then the puzzle would already be solved; in Slant
it would still require user action to fill in the lines, but it would
present no challenge at all). Therefore, a simple generation technique
is to leave the decision of which clues to provide until the last
minute. In other words, first generate a random _filled_ grid with all
possible clues present, and then gradually remove clues for as long as
the solver reports that it's still soluble. Unlike the methods described
above, this technique _cannot_ fail - once you've got a filled grid,
nothing can stop you from being able to convert it into a viable puzzle.
However, it wouldn't even be meaningful to apply this technique to (say)
Pattern, in which clues can never be left out, so the only way to affect
the set of clues is by altering the solution.

(Unfortunately, Solo is complicated by the need to provide puzzles at
varying difficulty levels. It's easy enough to generate a puzzle of
_at most_ a given level of difficulty; you just have a solver with
configurable intelligence, and you set it to a given level and apply the
above technique, thus guaranteeing that the resulting grid is solvable
by someone with at most that much intelligence. However, generating a
puzzle of _at least_ a given level of difficulty is rather harder; if
you go for _at most_ Intermediate level, you're likely to find that
you've accidentally generated a Trivial grid a lot of the time, because
removing just one number is sufficient to take the puzzle from Trivial
straight to Ambiguous. In that situation Solo has no remaining options
but to throw the puzzle away and start again.)

A final strategy is to use the solver _during_ puzzle construction:
lay out a bit of the grid, run the solver to see what it allows you to
deduce, and then lay out a bit more to allow the solver to make more
progress. There are articles on the web that recommend constructing
Sudoku puzzles by this method (which is completely the opposite way
round to how Solo does it); for Sudoku it has the advantage that you
get to specify your clue squares in advance (so you can have them make
pretty patterns).

Rectangles uses a strategy along these lines. First it generates a grid
by placing the actual rectangles; then it has to decide where in each
rectangle to place a number. It uses a solver to help it place the
numbers in such a way as to ensure a unique solution. It does this by
means of running a test solver, but it runs the solver _before_ it's
placed any of the numbers - which means the solver must be capable of
coping with uncertainty about exactly where the numbers are! It runs
the solver as far as it can until it gets stuck; then it narrows down
the possible positions of a number in order to allow the solver to make
more progress, and so on. Most of the time this process terminates with
the grid fully solved, at which point any remaining number-placement
decisions can be made at random from the options not so far ruled out.
Note that unlike the Net/Mines tweaking strategy described above, this
algorithm does not require a checking run after it completes: if it
finishes successfully at all, then it has definitely produced a uniquely
soluble puzzle.

Most of the strategies described above are not 100% reliable. Each
one has a failure rate: every so often it has to throw out the whole
grid and generate a fresh one from scratch. (Solo's strategy would
be the exception, if it weren't for the need to provide configurable
difficulty levels.) Occasional failures are not a fundamental problem in
this sort of work, however: it's just a question of dividing the grid
generation time by the success rate (if it takes 10ms to generate a
candidate grid and 1/5 of them work, then it will take 50ms on average
to generate a viable one), and seeing whether the expected time taken
to _successfully_ generate a puzzle is unacceptably slow. Dominosa's
generator has a very low success rate (about 1 out of 20 candidate grids
turn out to be usable, and if you think _that's_ bad then go and look
at the source code and find the comment showing what the figures were
before the generation-time tweaks!), but the generator itself is very
fast so this doesn't matter. Rectangles has a slower generator, but
fails well under 50% of the time.

So don't be discouraged if you have an algorithm that doesn't always
work: if it _nearly_ always works, that's probably good enough. The one
place where reliability is important is that your algorithm must never
produce false positives: it must not claim a puzzle is soluble when it
isn't. It can produce false negatives (failing to notice that a puzzle
is soluble), and it can fail to generate a puzzle at all, provided it
doesn't do either so often as to become slow.

One last piece of advice: for grid-based puzzles, when writing and
testing your generation algorithm, it's almost always a good idea _not_
to test it initially on a grid that's square (i.e. w==h), because if the
grid is square then you won't notice if you mistakenly write `h' instead
of `w' (or vice versa) somewhere in the code. Use a rectangular grid for
testing, and any size of grid will be likely to work after that.

6.2.2. Designing textual description formats
--------------------------------------------

Another aspect of writing a puzzle which is worth putting some thought
into is the design of the various text description formats: the format
of the game parameter encoding, the game description encoding, and the
move encoding.

The first two of these should be reasonably intuitive for a user to type
in; so provide some flexibility where possible. Suppose, for example,
your parameter format consists of two numbers separated by an `x' to
specify the grid dimensions (`10x10' or `20x15'), and then has some
suffixes to specify other aspects of the game type. It's almost always a
good idea in this situation to arrange that decode_params() can handle
the suffixes appearing in any order, even if encode_params() only ever
generates them in one order.

These formats will also be expected to be reasonably stable: users will
expect to be able to exchange game IDs with other users who aren't
running exactly the same version of your game. So make them robust and
stable: don't build too many assumptions into the game ID format which
will have to be changed every time something subtle changes in the
puzzle code.

6.3. Common how-to questions
----------------------------

This section lists some common things people want to do when writing a
puzzle, and describes how to achieve them within the Puzzles framework.

6.3.1. Drawing objects at only one position
-------------------------------------------

A common phenomenon is to have an object described in the `game_state'
or the `game_ui' which can only be at one position. A cursor - probably
specified in the `game_ui' - is a good example.

In the `game_ui', it would _obviously_ be silly to have an array
covering the whole game grid with a boolean flag stating whether the
cursor was at each position. Doing that would waste space, would make
it difficult to find the cursor in order to do anything with it, and
would introduce the potential for synchronisation bugs in which you
ended up with two cursors or none. The obviously sensible way to store a
cursor in the `game_ui' is to have fields directly encoding the cursor's
coordinates.

However, it is a mistake to assume that the same logic applies to the
`game_drawstate'. If you replicate the cursor position fields in the
draw state, the redraw code will get very complicated. In the draw
state, in fact, it _is_ probably the right thing to have a cursor flag
for every position in the grid. You probably have an array for the whole
grid in the drawstate already (stating what is currently displayed in
the window at each position); the sensible approach is to add a `cursor'
flag to each element of that array. Then the main redraw loop will look
something like this (pseudo-code):

  for (y = 0; y < h; y++) {
      for (x = 0; x < w; x++) {
          int value = state->symbol_at_position[y][x];
          if (x == ui->cursor_x && y == ui->cursor_y)
              value |= CURSOR;
          if (ds->symbol_at_position[y][x] != value) {
              symbol_drawing_subroutine(dr, ds, x, y, value);
              ds->symbol_at_position[y][x] = value;
          }
      }
  }

This loop is very simple, pretty hard to get wrong, and _automatically_
deals both with erasing the previous cursor and drawing the new one,
with no special case code required.

This type of loop is generally a sensible way to write a redraw
function, in fact. The best thing is to ensure that the information
stored in the draw state for each position tells you _everything_ about
what was drawn there. A good way to ensure that is to pass precisely
the same information, and _only_ that information, to a subroutine that
does the actual drawing; then you know there's no additional information
which affects the drawing but which you don't notice changes in.

6.3.2. Implementing a keyboard-controlled cursor
------------------------------------------------

It is often useful to provide a keyboard control method in a basically
mouse-controlled game. A keyboard-controlled cursor is best implemented
by storing its location in the `game_ui' (since if it were in the
`game_state' then the user would have to separately undo every cursor
move operation). So the procedure would be:

 -  Put cursor position fields in the `game_ui'.

 -  interpret_move() responds to arrow keys by modifying the cursor
    position fields and returning "".

 -  interpret_move() responds to some sort of fire button by actually
    performing a move based on the current cursor location.

 -  You might want an additional `game_ui' field stating whether the
    cursor is currently visible, and having it disappear when a mouse
    action occurs (so that it doesn't clutter the display when not
    actually in use).

 -  You might also want to automatically hide the cursor in
    changed_state() when the current game state changes to one in
    which there is no move to make (which is the case in some types of
    completed game).

 -  redraw() draws the cursor using the technique described in section
    6.3.1.

6.3.3. Implementing draggable sprites
-------------------------------------

Some games have a user interface which involves dragging some sort of
game element around using the mouse. If you need to show a graphic
moving smoothly over the top of other graphics, use a blitter (see
section 3.1.11 for the blitter API) to save the background underneath
it. The typical scenario goes:

 -  Have a blitter field in the `game_drawstate'.

 -  Set the blitter field to NULL in the game's new_drawstate()
    function, since you don't yet know how big the piece of saved
    background needs to be.

 -  In the game's set_size() function, once you know the size of the
    object you'll be dragging around the display and hence the required
    size of the blitter, actually allocate the blitter.

 -  In free_drawstate(), free the blitter if it's not NULL.

 -  In interpret_move(), respond to mouse-down and mouse-drag events by
    updating some fields in the game_ui which indicate that a drag is in
    progress.

 -  At the _very end_ of redraw(), after all other drawing has been
    done, draw the moving object if there is one. First save the
    background under the object in the blitter; then set a clip
    rectangle covering precisely the area you just saved (just in case
    anti-aliasing or some other error causes your drawing to go beyond
    the area you saved). Then draw the object, and call unclip().
    Finally, set a flag in the game_drawstate that indicates that the
    blitter needs restoring.

 -  At the very start of redraw(), before doing anything else at all,
    check the flag in the game_drawstate, and if it says the blitter
    needs restoring then restore it. (Then clear the flag, so that this
    won't happen again in the next redraw if no moving object is drawn
    this time.)

This way, you will be able to write the rest of the redraw function
completely ignoring the dragged object, as if it were floating above
your bitmap and being completely separate.

6.3.4. Sharing large invariant data between all game states
-----------------------------------------------------------

In some puzzles, there is a large amount of data which never changes
between game states. The array of numbers in Dominosa is a good example.

You _could_ dynamically allocate a copy of that array in every
`game_state', and have dup_game() make a fresh copy of it for every new
`game_state'; but it would waste memory and time. A more efficient way
is to use a reference-counted structure.

 -  Define a structure type containing the data in question, and also
    containing an integer reference count.

 -  Have a field in `game_state' which is a pointer to this structure.

 -  In new_game(), when creating a fresh game state at the start of a
    new game, create an instance of this structure, initialise it with
    the invariant data, and set its reference count to 1.

 -  In dup_game(), rather than making a copy of the structure for the
    new game state, simply set the new game state to point at the same
    copy of the structure, and increment its reference count.

 -  In free_game(), decrement the reference count in the structure
    pointed to by the game state; if the count reaches zero, free the
    structure.

This way, the invariant data will persist for only as long as it's
genuinely needed; _as soon_ as the last game state for a particular
puzzle instance is freed, the invariant data for that puzzle will
vanish as well. Reference counting is a very efficient form of garbage
collection, when it works at all. (Which it does in this instance, of
course, because there's no possibility of circular references.)

6.3.5. Implementing multiple types of flash
-------------------------------------------

In some games you need to flash in more than one different way. Mines,
for example, flashes white when you win, and flashes red when you tread
on a mine and die.

The simple way to do this is:

 -  Have a field in the `game_ui' which describes the type of flash.

 -  In flash_length(), examine the old and new game states to decide
    whether a flash is required and what type. Write the type of flash
    to the `game_ui' field whenever you return non-zero.

 -  In redraw(), when you detect that `flash_time' is non-zero, examine
    the field in `game_ui' to decide which type of flash to draw.

redraw() will never be called with `flash_time' non-zero unless
flash_length() was first called to tell the mid-end that a flash was
required; so whenever redraw() notices that `flash_time' is non-zero,
you can be sure that the field in `game_ui' is correctly set.

6.3.6. Animating game moves
---------------------------

A number of puzzle types benefit from a quick animation of each move you
make.

For some games, such as Fifteen, this is particularly easy. Whenever
redraw() is called with `oldstate' non-NULL, Fifteen simply compares the
position of each tile in the two game states, and if the tile is not in
the same place then it draws it some fraction of the way from its old
position to its new position. This method copes automatically with undo.

Other games are less obvious. In Sixteen, for example, you can't just
draw each tile a fraction of the way from its old to its new position:
if you did that, the end tile would zip very rapidly past all the others
to get to the other end and that would look silly. (Worse, it would look
inconsistent if the end tile was drawn on top going one way and on the
bottom going the other way.)

A useful trick here is to define a field or two in the game state that
indicates what the last move was.

 -  Add a `last move' field to the `game_state' (or two or more fields
    if the move is complex enough to need them).

 -  new_game() initialises this field to a null value for a new game
    state.

 -  execute_move() sets up the field to reflect the move it just
    performed.

 -  redraw() now needs to examine its `dir' parameter. If `dir' is
    positive, it determines the move being animated by looking at the
    last-move field in `newstate'; but if `dir' is negative, it has to
    look at the last-move field in `oldstate', and invert whatever move
    it finds there.

Note also that Sixteen needs to store the _direction_ of the move,
because you can't quite determine it by examining the row or column in
question. You can in almost all cases, but when the row is precisely
two squares long it doesn't work since a move in either direction looks
the same. (You could argue that since moving a 2-element row left and
right has the same effect, it doesn't matter which one you animate; but
in fact it's very disorienting to click the arrow left and find the row
moving right, and almost as bad to undo a move to the right and find the
game animating _another_ move to the right.)

6.3.7. Animating drag operations
--------------------------------

In Untangle, moves are made by dragging a node from an old position to a
new position. Therefore, at the time when the move is initially made, it
should not be animated, because the node has already been dragged to the
right place and doesn't need moving there. However, it's nice to animate
the same move if it's later undone or redone. This requires a bit of
fiddling.

The obvious approach is to have a flag in the `game_ui' which inhibits
move animation, and to set that flag in interpret_move(). The question
is, when would the flag be reset again? The obvious place to do so
is changed_state(), which will be called once per move. But it will
be called _before_ anim_length(), so if it resets the flag then
anim_length() will never see the flag set at all.

The solution is to have _two_ flags in a queue.

 -  Define two flags in `game_ui'; let's call them `current' and `next'.

 -  Set both to FALSE in `new_ui()'.

 -  When a drag operation completes in interpret_move(), set the `next'
    flag to TRUE.

 -  Every time changed_state() is called, set the value of `current' to
    the value in `next', and then set the value of `next' to FALSE.

 -  That way, `current' will be TRUE _after_ a call to changed_state()
    if and only if that call to changed_state() was the result of a
    drag operation processed by interpret_move(). Any other call to
    changed_state(), due to an Undo or a Redo or a Restart or a Solve,
    will leave `current' FALSE.

 -  So now anim_length() can request a move animation if and only if the
    `current' flag is _not_ set.

6.3.8. Inhibiting the victory flash when Solve is used
------------------------------------------------------

Many games flash when you complete them, as a visual congratulation for
having got to the end of the puzzle. It often seems like a good idea to
disable that flash when the puzzle is brought to a solved state by means
of the Solve operation.

This is easily done:

 -  Add a `cheated' flag to the `game_state'.

 -  Set this flag to FALSE in new_game().

 -  Have solve() return a move description string which clearly
    identifies the move as a solve operation.

 -  Have execute_move() respond to that clear identification by setting
    the `cheated' flag in the returned `game_state'. The flag will
    then be propagated to all subsequent game states, even if the user
    continues fiddling with the game after it is solved.

 -  flash_length() now returns non-zero if `oldstate' is not completed
    and `newstate' is, _and_ neither state has the `cheated' flag set.

6.4. Things to test once your puzzle is written
-----------------------------------------------

Puzzle implementations written in this framework are self-testing as far
as I could make them.

Textual game and move descriptions, for example, are generated and
parsed as part of the normal process of play. Therefore, if you can make
moves in the game _at all_ you can be reasonably confident that the
mid-end serialisation interface will function correctly and you will
be able to save your game. (By contrast, if I'd stuck with a single
make_move() function performing the jobs of both interpret_move() and
execute_move(), and had separate functions to encode and decode a game
state in string form, then those functions would not be used during
normal play; so they could have been completely broken, and you'd never
know it until you tried to save the game - which would have meant you'd
have to test game saving _extensively_ and make sure to test every
possible type of game state. As an added bonus, doing it the way I did
leads to smaller save files.)

There is one exception to this, which is the string encoding of the
`game_ui'. Most games do not store anything permanent in the `game_ui',
and hence do not need to put anything in its encode and decode
functions; but if there is anything in there, you do need to test game
loading and saving to ensure those functions work properly.

It's also worth testing undo and redo of all operations, to ensure that
the redraw and the animations (if any) work properly. Failing to animate
undo properly seems to be a common error.

Other than that, just use your common sense.