zsh/Util/zsh-development-guide

------------------------------
GUIDELINES FOR ZSH DEVELOPMENT
------------------------------

Zsh is currently developed and maintained by the Zsh Development Group.
This development takes place by mailing list.  Check the META-FAQ for the
various zsh mailing lists and how to subscribe to them.  The development
is very open and anyone is welcomed and encouraged to join and contribute.
Because zsh is a very large package whose development can sometimes
be very rapid, I kindly ask that people observe a few guidelines when
contributing patches and feedback to the mailing list.  These guidelines
are very simple and hopefully should make for a more orderly development
of zsh.

Patches
-------

* Send all patches to the mailing list rather than directly to me.

* Send only context diffs "diff -c oldfile newfile" or unified diffs
  "diff -u oldfile newfile".  They are much easier to read and
  understand while also allowing the patch program to patch more
  intelligently.  Please make sure the filenames in the diff header
  are relative to the top-level directory of the zsh distribution; for
  example, it should say "Src/init.c" rather than "init.c" or
  "zsh/Src/init.c".

* Please put only one bug fix or feature enhancement in a single patch and
  only one patch per mail message.  This helps me to multiplex the many
  (possibly conflicting) patches that I receive for zsh.  You shouldn't
  needlessly split patches, but send them in the smallest LOGICAL unit.

* If a patch depends on other patches, then please say so.  Also please
  mention what version of zsh this patch is for.

* Please test your patch and make sure it applies cleanly. It takes
  considerably more time to manually merge a patch into the baseline code.

* There is now a zsh patch archive.  To have your patches appear in the
  archive, send them to the mailing list with a Subject: line starting
  with "PATCH:".

C coding style
--------------

* The primary language is ANSI C as defined by the 1989 standard, but the
  code should always be compatible with late K&R era compilers ("The C
  Programming Language" 1st edition, plus "void" and "enum").  There are
  many hacks to avoid the need to actually restrict the code to K&R C --
  check out the configure tests -- but always bear the compatibility
  requirements in mind.  In particular, preprocessing directives must
  have the "#" unindented, and string pasting is not available.

* Conversely, there are preprocessor macros to provide safe access to some
  language features not present in pure ANSI C, such as variable-length
  arrays.  Always use the macros if you want to use these facilities.

* Avoid writing code that generates warnings under gcc with the default
  options set by the configure script.  For example, write
  "if ((foo = bar))" rather than "if (foo = bar)".

* Please try not using lines longer than 79 characters.

* The indent/brace style is Kernighan and Ritchie with 4 characters
  indentations (with leading tab characters replacing sequences of
  8 spaces).  This means that the opening brace is the last character
  in the line of the if/while/for/do statement and the closing brace
  has its own line:

      if (foo) {
	  do that
      }

* Put only one simple statement on a line.  The body of an if/while/for/do
  statement has its own line with 4 characters indentation even if there
  are no braces.

* Do not use space between the function name and the opening parenthesis.
  Use space after if/for/while.  Use space after type casts.

* Do not use (unsigned char) casts since some compilers do not handle
  them properly.  Use the provided STOUC(X) macro instead.

* If you use emacs 19.30 or newer you can put the following line to your
  ~/.emacs file to make these formatting rules the default:

    (add-hook 'c-mode-common-hook (function (lambda () (c-set-style "BSD"))))

* Function declarations must look like this:

  /**/
  int
  foo(char *s, char **p)
  {
      function body
  }

  There must be an empty line, a line with "/**/", a line with the
  type of the function, and finally the name of the function with typed
  arguments.  These lines must not be indented.  The script generating
  function prototypes and the ansi2knr program depend on this format.
  If the function is not used outside the file it is defined in, it
  should be declared "static"; this keyword goes on the type line,
  before the return type.

* Global variable declarations must similarly be preceded by a
  line containing only "/**/", for the prototype generation script.
  The declaration itself should be all on one line (except for multi-line
  initialisers).

* Leave a blank line between the declarations and statements in a compound
  statement, if both are present.  Use blank lines elsewhere to separate
  groups of statements in the interests of clarity.  There should never
  be two consecutive blank lines.

Modules
-------

Modules are described by a file named `foo.mdd' for a module
`foo'. This file is actually a shell script that will sourced when zsh 
is build. To describe the module it can/should set the following shell 
variables:

  - moddeps         modules on which this module depends (default none)
  - nozshdep        non-empty indicates no dependence on the `zsh' pseudo-module
  - alwayslink      if non-empty, always link the module into the executable
  - autobins        builtins defined by the module, for autoloading
  - autoinfixconds  infix condition codes defined by the module, for
                    autoloading (without the leading `-')
  - autoprefixconds like autoinfixconds, but for prefix condition codes
  - autoparams      parameters defined by the module, for autoloading
  - objects         .o files making up this module (*must* be defined)
  - proto           .pro files for this module (default generated from $objects)
  - headers         extra headers for this module (default none)
  - hdrdeps         extra headers on which the .mdh depends (default none)
  - otherincs       extra headers that are included indirectly (default none)

Be sure to put the values in quotes. For further enlightenment have a
look at the `mkmakemod.sh' script in the Src directory of the
distribution.

Modules have to define four functions which will be called automatically
by the zsh core. The first one, named `setup_foo' for a module named
`foo', should set up any data needed in the module, at least any data
other modules may be interested in. The second one, named `boot_foo',
should register all builtins, conditional codes, and function wrappers
(i.e. anything that will be visible to the user) and will be called
after the `setup'-function. 
The third one, named `cleanup_foo' for module `foo' is called when the
user tries to unload a module and should de-register the builtins
etc. The last function, `finish_foo' is called when the module is
actually unloaded and should finalize all the data initialized in the 
`setup'-function. Since the last two functions are only executed when
the module is used as an dynamically loaded module you can surround
it with `#ifdef MODULE' and `#endif'.
In short, the `cleanup'-function should undo what the `boot'-function
did, and the `finish'-function should undo what the `setup'-function
did.
All of these functions should return zero if they succeeded and
non-zero otherwise.

Builtins are described in a table, for example:

  static struct builtin bintab[] = {
    BUILTIN("example", 0, bin_example, 0, -1, 0, "flags", NULL),
  };

Here `BUILTIN(...)' is a macro that simplifies the description. Its
arguments are:
  - the name of the builtin as a string
  - optional flags (see BINF_* in zsh.h)
  - the C-function implementing the builtin
  - the minimum number of arguments the builtin needs
  - the maximum number of arguments the builtin can handle or -1 if
    the builtin can get any number of arguments
  - an integer that is passed to the handler function and can be used
    to distinguish builtins if the same C-function is used to
    implement multiple builtins
  - the options the builtin accepts, given as a string containing the
    option characters (the above example makes the builtin accept the
    options `f', `l', `a', `g', and `s')
  - and finally a optional string containing option characters that
    will always be reported as set when calling the C-function (this,
    too, can be used when using one C-function to implement multiple
    builtins)

The definition of the handler function looks like:

  /**/
  static int
  bin_example(char *nam, char **args, char *ops, int func)
  {
    ...
  }

The special comment /**/ is used by the zsh Makefile to generate the
`*.pro' files. The arguments of the function are the number under
which this function was invoked (the name of the builtin, but for
functions that implement more than one builtin this information is
needed). The second argument is the array of arguments *excluding* the 
options that were defined in the struct and which are handled by the
calling code. These options are given as the third argument. It is an
array of 256 characters in which the n'th element is non-zero if the
option with ASCII-value n was set (i.e. you can easily test if an
option was used by `if (ops['f'])' etc.). The last argument is the
integer value from the table (the sixth argument to `BUILTIN(...)').
The integer return value by the function is the value returned by the
builtin in shell level.

To register builtins in zsh and thereby making them visible to the
user the function `addbuiltins()' is used:

  /**/
  int
  boot_example(Module m)
  {
    int ret;

    ret = addbuiltins(m->nam, bintab, sizeof(bintab)/sizeof(*bintab));
    ...
  }

The arguments are the name of the module (taken from the argument in
the example), the table of definitions and the number of entries in
this table.
The return value is 1 if everything went fine, 2 if at least one
builtin couldn't be defined, and 0 if none of the builtin could be
defined.

To de-register builtins use the function `deletebuiltins()':

  /**/
  int
  cleanup_example(Module m)
  {
    deletebuiltins(m->nam, bintab, sizeof(bintab)/sizeof(*bintab));
    ...
  }

The arguments and the return value are the same as for `addbuiltins()'

The definition of condition codes in modules is equally simple. First
we need a table with the descriptions:

  static struct conddef cotab[] = {
    CONDDEF("len", 0, cond_p_len, 1, 2, 0),
    CONDDEF("ex", CONDF_INFIX, cond_i_ex, 0, 0, 0),
  };

Again a macro is used, with the following arguments:

  - the name of the condition code without the leading hyphen
    (i.e. the example makes the condition codes `-len' and `-ex'
    usable in `[[...]]' constructs)
  - an optional flag which for now can only be CONDF_INFIX; if this is 
    given, an infix operator is created (i.e. the above makes
    `[[ -len str ]]' and `[[ s1 -ex s2 ]]' available)
  - the C-function implementing the conditional
  - for non-infix condition codes the next two arguments give the
    minimum and maximum number of string the conditional can handle
    (i.e. `-len' can get one or two strings); as with builtins giving
    -1 as the maximum number means that the conditional accepts any
    number of strings
  - finally as the last argument an integer that is passed to the
    handler function that can be used to distinguish different
    condition codes if the same C-function implements more than one of 
    them

The definition for the function looks like:

  /**/
  static int
  cond_p_len(char **a, int id)
  {
    ...
  }

The first argument is an array containing the strings (NULL-terminated
like the array of arguments for builtins), the second argument is the
integer value stored in the table (the last argument to `CONDDEF(...)').
The value returned by the function should be non-zero if the condition 
is true and zero otherwise.

Note that no preprocessing is done on the strings. This means that
no substitutions are performed on them and that they will be
tokenized. There are three helper functions available:

  - char *cond_str(args, num, raw)
    The first argument is the array of strings the handler function
    got as an argument and the second one is an index into this array.
    The return value is the num'th string from the array with
    substitutions performed. If the last argument is zero, the string
    will also be untokenized.
  - long cond_val(args, num)
    The arguments are the same as for cond_str(). The return value is
    the result of the mathematical evaluation of the num'th string
    form the array.
  - int cond_match(args, num, str)
    Again, the first two arguments are the same as for the other
    functions. The third argument is any string. The result of the
    function is non-zero if the the num'th string from the array taken 
    as a glob pattern matches the given string.

Registering and de-resgitering condition codes with the shell is
almost exactly the same as for builtins, using the functions
`addconddefs()' and `deleteconddefs()' instead:

  /**/
  int
  boot_example(Module m)
  {
    int ret;

    ret = addconddefs(m->nam, cotab, sizeof(cotab)/sizeof(*cotab));
    ...
  }

  /**/
  int
  cleanup_example(Module m)
  {
    deleteconddefs(m->nam, cotab, sizeof(cotab)/sizeof(*cotab));
    ...
  }

Arguments and return values are the same as for the functions for
builtins.

For defining parameters, a module can call `createparam()' directly or 
use a table to describe them, e.g.:

  static struct paramdef patab[] = {
    PARAMDEF("foo", PM_INTEGER, NULL, get_foo, set_foo, unset_foo),
    INTPARAMDEF("exint", &intparam),
    STRPARAMDEF("exstr", &strparam),
    ARRPARAMDEF("exarr", &arrparam),
  };

There are four macros used:

  - PARAMDEF() gets as arguments:
    - the name of the parameter
    - the parameter flags to set for it (from the PM_* flags defined
      in zsh.h)
    - optionally a pointer to a variable holding the value of the
      parameter
    - three functions that will be used to get the value of the
      parameter, store a value in the parameter, and unset the
      parameter
  - the other macros provide simple ways to define the most common
    types of parameters; they get the name of the parameter and a
    pointer to a variable holding the value as arguments; they are
    used to define integer-, scalar-, and array-parameters, so the
    variables whose addresses are given should be of type `long',
    `char *', and `char **', respectively

For a description of how to write functions for getting or setting the 
value of parameters, or how to write a function to unset a parameter,
see the description of the following functions in the `params.c' file:

  - `intvargetfn()' and `intvarsetfn()' for integer parameters
  - `strvargetfn()' and `strvarsetfn()' for scalar parameters
  - `arrvargetfn()' and `arrvarsetfn()' for array parameters
  - `stdunsetfn()' for unsetting parameters

Note that if one defines parameters using the last two macros (for
scalars and arrays), the variable holding the value should be
initialized to either `NULL' or to a a piece of memory created with
`zalloc()'. But this memory should *not* be freed in the
finish-function of the module because that will be taken care of by
the `deleteparamdefs()' function described below.

To register the parameters in the zsh core, the function
`addparamdefs()' is called as in:

  /**/
  int
  boot_example(Module m)
  {
    int ret;

    ret = addparamdefs(m->nam, patab, sizeof(patab)/sizeof(*patab))
    ...
  }

The arguments and the return value are as for the functions used to
add builtins and condition codes and like these, it should be called
in the boot-function of the module. To remove the parameters defined,
the function `deleteparamdefs()' should be called, again with the same 
arguments and the same return value as for the functions to remove
builtins and condition codes:

  /**/
  int
  cleanup_example(Module m)
  {
    deleteparamdefs(m->nam, patab, sizeof(patab)/sizeof(*patab));
    ...
  }

Modules can also define function hooks. Other modules can then add
functions to these hooks to make the first module call these functions
instead of the default.

Again, an array is used to define hooks:

  static struct hookdef foohooks[] = {
    HOOKDEF("foo", foofunc, 0),
  };

The first argument of the macro is the name of the hook. This name
is used whenever the hook is used. The second argument is the default
function for the hook or NULL if no default function exists. The
last argument is used to define flags for the hook. Currently only one
such flag is defined: `HOOKF_ALL'. If this flag is given and more than
one function was added to the hook, all functions will be called
(including the default function). Otherwise only the last function
added will be called.

The functions that can be used as default functions or that can be
added to a hook have to be defined like:

  /**/
  static int
  foofunc(Hookdef h, void *data)
  {
    ...
  }

The first argument is a pointer to the struct defining the hook. The
second argument is an arbitrary pointer that is given to the function
used to invoke hooks (see below).

The functions to register and de-register hooks look like those for
the other things that can be defined by modules:

  /**/
  int
  boot_foo(Module m)
  {
    int ret;

    ret = addhookdefs(m->nam, foohooks, sizeof(foohooks)/sizeof(*foohooks))
    ...
  }
  ...
  /**/
  int
  cleanup_foo(Module m)
  {
    deletehookdefs(m->nam, foohooks, sizeof(foohooks)/sizeof(*foohooks));
    ...
  }

Modules that define hooks can invoke the function(s) registered for
them by calling the function `runhook(name, data)'. The first argument 
is the name of the hook and the second one is the pointer given to the 
hook functions as their second argument. Hooks that have the `HOOKF_ALL' 
flag call all function defined for them until one returns non-zero.
The return value of `runhook()' is the return value of the last hook
function called or zero if none was called.

To add a function to a hook, the function `addhookfunc(name, func)' is 
called with the name of the hook and a hook function as arguments.
Deleting them is done by calling `deletehookfunc(name, func)' with the 
same arguments as for the corresponding call to `addhookfunc()'.

Alternative forms of the last three function are provided for hooks
that are changed or called very often. These functions,
`runhookdef(def, data)', `addhookdeffunc(def, func)', and
`deletehookdeffunc(def, func)' get a pointer to the `hookdef'
structure defining the hook instead of the name and otherwise behave
like their counterparts.

Modules can also define function hooks. Other modules can then add
functions to these hooks to make the first module call these functions
instead of the default.

Again, an array is used to define hooks:

  static struct hookdef foohooks[] = {
    HOOKDEF("foo", foofunc, 0),
  };

The first argument of the macro is the name of the hook. This name
is used whenever the hook is used. The second argument is the default
function for the hook or NULL if no default function exists. The
last argument is used to define flags for the hook. Currently only one
such flag is defined: `HOOKF_ALL'. If this flag is given and more than
one function was added to the hook, all functions will be called
(including the default function). Otherwise only the last function
added will be called.

The functions that can be used as default functions or that can be
added to a hook have to be defined like:

  /**/
  static int
  foofunc(Hookdef h, void *data)
  {
    ...
  }

The first argument is a pointer to the struct defining the hook. The
second argument is an arbitrary pointer that is given to the function
used to invoke hooks (see below).

The functions to register and de-register hooks look like those for
the other things that can be defined by modules:

  /**/
  int
  boot_foo(Module m)
  {
    int ret;

    ret = addhookdefs(m->nam, foohooks, sizeof(foohooks)/sizeof(*foohooks))
    ...
  }
  ...
  /**/
  int
  cleanup_foo(Module m)
  {
    deletehookdefs(m->nam, foohooks, sizeof(foohooks)/sizeof(*foohooks));
    ...
  }

Modules that define hooks can invoke the function(s) registered for
them by calling the function `runhook(name, data)'. The first argument 
is the name of the hook and the second one is the pointer given to the 
hook functions as their second argument. Hooks that have the `HOOKF_ALL' 
flag call all function defined for them until one returns non-zero.
The return value of `runhook()' is the return value of the last hook
function called or zero if none was called.

To add a function to a hook, the function `addhookfunc(name, func)' is 
called with the name of the hook and a hook function as arguments.
Deleting them is done by calling `deletehookfunc(name, func)' with the 
same arguments as for the corresponding call to `addhookfunc()'.

Alternative forms of the last three function are provided for hooks
that are changed or called very often. These functions,
`runhookdef(def, data)', `addhookdeffunc(def, func)', and
`deletehookdeffunc(def, func)' get a pointer to the `hookdef'
structure defining the hook instead of the name and otherwise behave
like their counterparts.

Finally, modules can define wrapper functions. These functions are
called whenever a shell function is to be executed.

The definition is simple:

  static struct funcwrap wrapper[] = {
    WRAPDEF(ex_wrapper),
  };

The macro `WRAPDEF(...)' gets the C-function as its only argument.
This function should be defined like:

  /**/
  static int
  ex_wrapper(List list, FuncWrap w, char *name)
  {
    ...
    runshfunc(list, w, name);
    ...
    return 0;
  }

The first two arguments should only be used to pass them to
`runshfunc()' which will execute the shell function. The last argument 
is the name of the function to be executed. The arguments passed to
the function can be accessed vie the global variable `pparams' (a
NULL-terminated array of strings).
The return value of the wrapper function should be zero if it calls
`runshfunc()' itself and non-zero otherwise. This can be used for
wrapper functions that only need to run under certain conditions or
that don't need to clean anything up after the shell function has
finished:

  /**/
  static int
  ex_wrapper(List list, FuncWrap w, char *name)
  {
    if (wrapper_need_to_run) {
      ...
      runshfunc(list, w, name);
      ...
      return 0;
    }
    return 1;
  }

Inside these wrapper functions the global variable `sfcontext' will be 
set to a vlue indicating the circumstances under which the shell
function was called. It can have any of the following values:

  - SFC_DIRECT:   the function was invoked directly by the user
  - SFC_SIGNAL:   the function was invoked as a signal handler
  - SFC_HOOK:     the function was automatically invoked as one of the
                  special functions known by the shell (like `chpwd')
  - SFC_WIDGET:   the function was called from the zsh line editor as a
                  user-defined widget
  - SFC_COMPLETE: the function was called from the completion code
                  (e.g. with `compctl -K func')

If a module invokes a shell function (e.g. as a hook function), the
value of this variable should only be changed temporarily and restored
to its previous value after the shell function has finished.

There is a problem when the user tries to unload a module that has
defined wrappers from a shell function. In this case the module can't
be unloaded immediately since the wrapper function is still on the
call stack. The zsh code delays unloading modules until all wrappers
from them have finished. To hide this from the user, the module's
cleanup function is run immediatly so that all builtins, condition
codes, and wrapper function defined by the module are
de-registered. But if there is some module-global state that has to be 
finalized (e.g. some memory that has to be freed) and that is used by
the wrapper functions finalizing this data in the cleanup function
won't work.
This is why ther are two functions each for the initialization and
finalization of modules. The `boot'- and `cleanup'-functions are run
whenever the user calls `zmodload' or `zmodload -u' and should only
register or de-register the module's interface that is visible to the
user. Anything else should be done in the `setup'- and
`finish'-functions. Otherwise modules that other modules depend upon
may destroy their state too early and wrapper functions in the latter
modules may stop working since the state they use is already destroyed.

Documentation
-------------

* Edit only the .yo files.  All other formats (man pages, TeXinfo, HTML,
  etc.) are automatically generated from the yodl source.

* Always use the correct markup.  em() is used for emphasis, and bf()
  for citations.  tt() marks text that is literal input to or output
  from the shell.  var() marks metasyntactic variables.

* In addition to appropriate markup, always use quotes (`') where
  appropriate.  Specifically, use quotes to mark text that is not a part
  of the actual text of the documentation (i.e., that it is being quoted).
  In principle, all combinations of quotes and markup are possible,
  because the purposes of the two devices are completely orthogonal.
  For example,

      Type `tt(xyzzy)' to let zsh know you have played tt(advent).
      Saying `plugh' aloud doesn't have much effect, however.

  In this case, "zsh" is normal text (a name), "advent" is a command name
  ocurring in the main text, "plugh" is a normal word that is being quoted
  (it's the user that says `plugh', not the documentation), and "xyzzy"
  is some text to be typed literally that is being quoted.

* For multiple-line pieces of text that should not be filled, there are
  various models.
  - If the text is pure example, i.e. with no metasyntactic variables etc.,
    it should be included within `example(...)'.  The text will be
    indented, will not be filled and will be put into a fixed width font.
  - If the text includes mixed fonts, it should be included within
    `indent(...)'.  The text is now filled unless `nofill(...)' is also
    used, and explicit font-changing commands are required inside.
  - If the text appears inside some other format, such as for example the
    `item()' list structure, then the instruction `nofill(...)', which
    simply turns off filling should be used; as with `indent(...)',
    explicit font changing commands are required.  This can be used
    without `indent()' when no identation is required, e.g. if the
    accumulated indentation would otherwise be too long.
  All the above should appear on their own, separated by newlines from the
  surrounding text.  No extra newlines after the opening or before the
  closing parenthesis are required.