view src/compiler.h @ 4690:257b468bf2ca

Move the #'query-coding-region implementation to C. This is necessary because there is no reasonable way to access the corresponding mswindows-multibyte functionality from Lisp, and we need such functionality if we're going to have a reliable and portable #'query-coding-region implementation. However, this change doesn't yet provide #'query-coding-region for the mswindow-multibyte coding systems, there should be no functional differences between an XEmacs with this change and one without it. src/ChangeLog addition: 2009-09-19 Aidan Kehoe <kehoea@parhasard.net> Move the #'query-coding-region implementation to C. This is necessary because there is no reasonable way to access the corresponding mswindows-multibyte functionality from Lisp, and we need such functionality if we're going to have a reliable and portable #'query-coding-region implementation. However, this change doesn't yet provide #'query-coding-region for the mswindow-multibyte coding systems, there should be no functional differences between an XEmacs with this change and one without it. * mule-coding.c (struct fixed_width_coding_system): Add a new coding system type, fixed_width, and implement it. It uses the CCL infrastructure but has a much simpler creation API, and its own query_method, formerly in lisp/mule/mule-coding.el. * unicode.c: Move the Unicode query method implementation here from unicode.el. * lisp.h: Declare Fmake_coding_system_internal, Fcopy_range_table here. * intl-win32.c (complex_vars_of_intl_win32): Use Fmake_coding_system_internal, not Fmake_coding_system. * general-slots.h: Add Qsucceeded, Qunencodable, Qinvalid_sequence here. * file-coding.h (enum coding_system_variant): Add fixed_width_coding_system here. (struct coding_system_methods): Add query_method and query_lstream_method to the coding system methods. Provide flags for the query methods. Declare the default query method; initialise it correctly in INITIALIZE_CODING_SYSTEM_TYPE. * file-coding.c (default_query_method): New function, the default query method for coding systems that do not set it. Moved from coding.el. (make_coding_system_1): Accept new elements in PROPS in #'make-coding-system; aliases, a list of aliases; safe-chars and safe-charsets (these were previously accepted but not saved); and category. (Fmake_coding_system_internal): New function, what used to be #'make-coding-system--on Mule builds, we've now moved some of the functionality of this to Lisp. (Fcoding_system_canonical_name_p): Move this earlier in the file, since it's now called from within make_coding_system_1. (Fquery_coding_region): Move the implementation of this here, from coding.el. (complex_vars_of_file_coding): Call Fmake_coding_system_internal, not Fmake_coding_system; specify safe-charsets properties when we're a mule build. * extents.h (mouse_highlight_priority, Fset_extent_priority, Fset_extent_face, Fmap_extents): Make these available to other C files. lisp/ChangeLog addition: 2009-09-19 Aidan Kehoe <kehoea@parhasard.net> Move the #'query-coding-region implementation to C. * coding.el: Consolidate code that depends on the presence or absence of Mule at the end of this file. (default-query-coding-region, query-coding-region): Move these functions to C. (default-query-coding-region-safe-charset-skip-chars-map): Remove this variable, the corresponding C variable is Vdefault_query_coding_region_chartab_cache in file-coding.c. (query-coding-string): Update docstring to reflect actual multiple values, be more careful about not modifying a range table that we're currently mapping over. (encode-coding-char): Make the implementation of this simpler. (featurep 'mule): Autoload #'make-coding-system from mule/make-coding-system.el if we're a mule build; provide an appropriate compiler macro. Do various non-mule compatibility things if we're not a mule build. * update-elc.el (additional-dump-dependencies): Add mule/make-coding-system as a dump time dependency if we're a mule build. * unicode.el (ccl-encode-to-ucs-2): (decode-char): (encode-char): Move these earlier in the file, for the sake of some byte compile warnings. (unicode-query-coding-region): Move this to unicode.c * mule/make-coding-system.el: New file, not dumped. Contains the functionality to rework the arguments necessary for fixed-width coding systems, and contains the implementation of #'make-coding-system, which now calls #'make-coding-system-internal. * mule/vietnamese.el (viscii): * mule/latin.el (iso-8859-2): (windows-1250): (iso-8859-3): (iso-8859-4): (iso-8859-14): (iso-8859-15): (iso-8859-16): (iso-8859-9): (macintosh): (windows-1252): * mule/hebrew.el (iso-8859-8): * mule/greek.el (iso-8859-7): (windows-1253): * mule/cyrillic.el (iso-8859-5): (koi8-r): (koi8-u): (windows-1251): (alternativnyj): (koi8-ru): (koi8-t): (koi8-c): (koi8-o): * mule/arabic.el (iso-8859-6): (windows-1256): Move all these coding systems to being of type fixed-width, not of type CCL. This allows the distinct query-coding-region for them to be in C, something which will eventually allow us to implement query-coding-region for the mswindows-multibyte coding systems. * mule/general-late.el (posix-charset-to-coding-system-hash): Document why we're pre-emptively persuading the byte compiler that the ELC for this file needs to be written using escape-quoted. Call #'set-unicode-query-skip-chars-args, now the Unicode query-coding-region implementation is in C. * mule/thai-xtis.el (tis-620): Don't bother checking whether we're XEmacs or not here. * mule/mule-coding.el: Move the eight bit fixed-width functionality from this file to make-coding-system.el. tests/ChangeLog addition: 2009-09-19 Aidan Kehoe <kehoea@parhasard.net> * automated/mule-tests.el: Check a coding system's type, not an 8-bit-fixed property, for whether that coding system should be treated as a fixed-width coding system. * automated/query-coding-tests.el: Don't test the query coding functionality for mswindows-multibyte coding systems, it's not yet implemented.
author Aidan Kehoe <kehoea@parhasard.net>
date Sat, 19 Sep 2009 22:53:13 +0100
parents f386b9b92417
children aa5ed11f473b
line wrap: on
line source

/* Compiler-specific definitions for XEmacs.
   Copyright (C) 1998-1999, 2003 Free Software Foundation, Inc.
   Copyright (C) 1994 Richard Mlynarik.
   Copyright (C) 1995, 1996, 2000-2004 Ben Wing.

This file is part of XEmacs.

XEmacs is free software; you can redistribute it and/or modify it
under the terms of the GNU General Public License as published by the
Free Software Foundation; either version 2, or (at your option) any
later version.

XEmacs is distributed in the hope that it will be useful, but WITHOUT
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
for more details.

You should have received a copy of the GNU General Public License
along with XEmacs; see the file COPYING.  If not, write to
the Free Software Foundation, Inc., 59 Temple Place - Suite 330,
Boston, MA 02111-1307, USA.  */

/* Synched up with: not in FSF. */

/* Authorship:

   NOT_REACHED, DOESNT_RETURN, PRINTF_ARGS by Richard Mlynarik, c. 1994.
   RETURN_SANS_WARNING by Martin buchholz, 1998 or 1999.
   Many changes and improvements by Jerry James, 2003.
     Split out of lisp.h, reorganized, and modernized.
     {BEGIN,END}_C_DECLS, NEED_GCC, GCC_VERSION
     ATTRIBUTE_MALLOC, ATTRIBUTE_CONST, ATTRIBUTE_PURE, UNUSED
*/

#ifndef INCLUDED_compiler_h
#define INCLUDED_compiler_h

/* Define min() and max(). (Some compilers put them in strange places that
   won't be referenced by include files used by XEmacs, such as `macros.h'
   under Solaris.) */

#ifndef min
# define min(a,b) (((a) <= (b)) ? (a) : (b))
#endif
#ifndef max
# define max(a,b) (((a) > (b)) ? (a) : (b))
#endif

/* Regular C complains about possible clobbering of local vars NOT declared
   as volatile if there's a longjmp() in a function.  C++ complains if such
   vars ARE volatile; or more correctly, sans volatile no problem even when
   you longjmp, avec volatile you get unfixable compile errors like

/src/xemacs/lilfix/src/process-unix.c: In function `void
   unix_send_process(Lisp_Object, lstream*)':
/src/xemacs/lilfix/src/process-unix.c:1577: no matching function for call to `
   Lisp_Object::Lisp_Object(volatile Lisp_Object&)'
/src/xemacs/lilfix/src/lisp-union.h:32: candidates are:
   Lisp_Object::Lisp_Object(const Lisp_Object&)
*/

#ifdef __cplusplus
# define VOLATILE_IF_NOT_CPP
#else
# define VOLATILE_IF_NOT_CPP volatile
#endif

/* Avoid indentation problems when XEmacs sees the curly braces */
#ifndef BEGIN_C_DECLS
# ifdef __cplusplus
#  define BEGIN_C_DECLS extern "C" {
#  define END_C_DECLS }
# else
#  define BEGIN_C_DECLS
#  define END_C_DECLS
# endif
#endif

/* Guard against older gccs that did not define all of these symbols */
#ifdef __GNUC__
# ifndef __GNUC_MINOR__
#  define __GNUC_MINOR__      0
# endif
# ifndef __GNUC_PATCHLEVEL__
#  define __GNUC_PATCHLEVEL__ 0
# endif
#endif /* __GNUC__ */

/* Simplify testing for specific GCC versions.  For non-GNU compilers,
   GCC_VERSION evaluates to zero. */
#ifndef NEED_GCC
# define NEED_GCC(major,minor,patch) (major * 1000000 + minor * 1000 + patch)
#endif /* NEED_GCC */
#ifndef GCC_VERSION
# ifdef __GNUC__
#  define GCC_VERSION NEED_GCC (__GNUC__, __GNUC_MINOR__, __GNUC_PATCHLEVEL__)
# else
#  define GCC_VERSION 0
# endif /* __GNUC__ */
#endif /* GCC_VERSION */

#ifdef _MSC_VER
#define MSC_VERSION _MSC_VER
#else
#define MSC_VERSION 0
#endif

/* GCC < 2.6.0 could only declare one attribute per function.  In that case,
   we define DOESNT_RETURN in preference to PRINTF_ARGS, which is only used
   for checking args against the string spec. */
#ifndef PRINTF_ARGS
# if (GCC_VERSION >= NEED_GCC (2, 6, 0))
#  define PRINTF_ARGS(string_index,first_to_check) \
          __attribute__ ((format (printf, string_index, first_to_check)))
# else
#  define PRINTF_ARGS(string_index,first_to_check)
# endif /* GNUC */
#endif

#ifndef DOESNT_RETURN_TYPE
# if (GCC_VERSION > NEED_GCC (0, 0, 0))
#  if (GCC_VERSION >= NEED_GCC (2, 5, 0))
#   ifndef __INTEL_COMPILER
#    define RETURN_NOT_REACHED(value) DO_NOTHING
#   endif
#   define DOESNT_RETURN_TYPE(rettype) rettype
#   define DECLARE_DOESNT_RETURN_TYPE(rettype,decl) rettype decl \
	   __attribute__ ((noreturn))
#  else /* GCC_VERSION < NEED_GCC (2, 5, 0) */
#   define DOESNT_RETURN_TYPE(rettype) rettype volatile
#   define DECLARE_DOESNT_RETURN_TYPE(rettype,decl) rettype volatile decl
#  endif /* GCC_VERSION >= NEED_GCC (2, 5, 0) */
# elif (MSC_VERSION >= 1200)
/* MSVC 6.0 has a mechanism to declare functions which never return */
#  define DOESNT_RETURN_TYPE(rettype) __declspec(noreturn) rettype
#  define DECLARE_DOESNT_RETURN_TYPE(rettype,decl) \
  __declspec(noreturn) rettype XCDECL decl
#  if (MSC_VERSION >= 1300)
/* VC++ 7 issues warnings about return statements in __declspec(noreturn)
   functions; this problem didn't exist under VC++ 6 */
#   define RETURN_NOT_REACHED(value) DO_NOTHING
#  endif
# else /* not gcc, VC++ */
#  define DOESNT_RETURN_TYPE(rettype) rettype
#  define DECLARE_DOESNT_RETURN_TYPE(rettype,decl) rettype decl
# endif /* GCC_VERSION > NEED_GCC (0, 0, 0) */
#endif /* DOESNT_RETURN_TYPE */
#ifndef DOESNT_RETURN
# define DOESNT_RETURN DOESNT_RETURN_TYPE (void)
# define DECLARE_DOESNT_RETURN(decl) DECLARE_DOESNT_RETURN_TYPE (void, decl)
#endif /* DOESNT_RETURN */

/* Another try to fix SunPro C compiler warnings */
/* "end-of-loop code not reached" */
/* "statement not reached */
#if defined __SUNPRO_C || defined __USLC__
# define RETURN_SANS_WARNINGS if (1) return
# define RETURN_NOT_REACHED(value) DO_NOTHING
#endif

/* More ways to shut up compiler.  This works in Fcommand_loop_1(),
   where there's an infinite loop in a function returning a Lisp object.
*/
#if (defined (_MSC_VER) && MSC_VERSION < 1300) || defined (__SUNPRO_C) || \
  defined (__SUNPRO_CC) || (defined (DEC_ALPHA) && defined (OSF1))
# define DO_NOTHING_DISABLING_NO_RETURN_WARNINGS if (0) return Qnil
#else
# define DO_NOTHING_DISABLING_NO_RETURN_WARNINGS DO_NOTHING
#endif

#ifndef RETURN_NOT_REACHED
# define RETURN_NOT_REACHED(value) return (value)
#endif

#ifndef RETURN_SANS_WARNINGS
# define RETURN_SANS_WARNINGS return
#endif

#ifndef DO_NOTHING
# define DO_NOTHING do {} while (0)
#endif

#ifndef DECLARE_NOTHING
# define DECLARE_NOTHING struct nosuchstruct
#endif

#ifndef ATTRIBUTE_MALLOC
# if (GCC_VERSION >= NEED_GCC (2, 96, 0))
#  define ATTRIBUTE_MALLOC __attribute__ ((__malloc__))
# else
#  define ATTRIBUTE_MALLOC
# endif /* GCC_VERSION >= NEED_GCC (2, 96, 0) */
#endif /* ATTRIBUTE_MALLOC */

#ifndef ATTRIBUTE_PURE
# if (GCC_VERSION >= NEED_GCC (2, 96, 0))
#  define ATTRIBUTE_PURE __attribute__ ((pure))
# else
#  define ATTRIBUTE_PURE
# endif /* GCC_VERSION >= NEED_GCC (2, 96, 0) */
#endif /* ATTRIBUTE_PURE */

#ifndef ATTRIBUTE_CONST
# if (GCC_VERSION >= NEED_GCC (2, 5, 0))
#  define ATTRIBUTE_CONST __attribute__ ((const))
#  define CONST_FUNC
# else
#  define ATTRIBUTE_CONST
#  define CONST_FUNC const
# endif /* GCC_VERSION >= NEED_GCC (2, 5, 0) */
#endif /* ATTRIBUTE_CONST */

/*
   NOTE:  These macros MUST be named UNUSED (exactly) or something
   prefixed with USED_IF_, or DEFUN docstrings will be parsed incorrectly.
   See comments in make_docfile.c (write_c_args).  You'd think that this
   wouldn't happen, but unfortunately we do indeed have some arguments
   of DEFUNs unused for GNU compatibility or because features are missing.

   #### At one time, __attribute__ ((unused)) confused G++.  We don't know
   which versions.  Please report problems and fix conditionals.
   #### A similar issue arose with the Intel CC.  We know that v7 didn't
   work and v9 does.  Let us know if v8 works or not, please.
   See <m34plsmh88.fsf@jerrypc.cs.usu.edu>.
*/
#ifndef UNUSED_ARG
# define UNUSED_ARG(decl) unused_##decl
#endif
#ifndef UNUSED
# if defined(__GNUC__) && (!defined(__INTEL_COMPILER) || __INTEL_COMPILER >= 800)
#  define ATTRIBUTE_UNUSED __attribute__ ((unused))
# else
#  define ATTRIBUTE_UNUSED
# endif
# define UNUSED(decl) UNUSED_ARG (decl) ATTRIBUTE_UNUSED
# ifdef MULE
#  define USED_IF_MULE(decl) decl
# else
#  define USED_IF_MULE(decl) UNUSED (decl)
# endif
# if defined (MULE) || defined (ERROR_CHECK_TEXT)
#  define USED_IF_MULE_OR_CHECK_TEXT(decl) decl
# else
#  define USED_IF_MULE_OR_CHECK_TEXT(decl) UNUSED (decl)
# endif
# ifdef USE_XFT
#  define USED_IF_XFT(decl) decl
# else
#  define USED_IF_XFT(decl) UNUSED (decl)
# endif
#endif /* UNUSED */

#ifdef DEBUG_XEMACS
# define REGISTER
# define register
#else
# define REGISTER register
#endif

#if defined(HAVE_MS_WINDOWS) && defined(HAVE_SHLIB)
# ifdef EMACS_MODULE
#  define MODULE_API __declspec(dllimport)
# else
#  define MODULE_API __declspec(dllexport)
# endif
#else
# define MODULE_API
#endif

/* Under "strict-aliasing" assumptions, you're not necessarily allowed to
   access the same memory address as two different types.  The proper way
   around that is with a union.  The macros below help out, e.g. the
   definition of XE_MAKEPOINTS(val) is

   ANSI_ALIASING_TYPEDEF (POINTS, POINTS);
   #define XE_MAKEPOINTS(l)       ANSI_ALIASING_CAST (POINTS, l)

   replacing

   BAD!!! #define XE_MAKEPOINTS(l)       (* (POINTS *) &(l))

   On the other hand, if you are just casting from one pointer to the other
   in order to pass a pointer to another function, it's probably OK to just
   trick GCC by inserting an intermediate cast to (void *), to avoid
   warnings about "dereferencing type-punned pointer".  #### I don't know
   how kosher this is, but do strict-aliasing rules really apply across
   functions?

   Note that the input to e.g. VOIDP_CAST must be an lvalue (i.e. not
   &(something)), but the value of the macro is also an lvalue, so in place
   of `(void **) &foo' you could write `& VOIDP_CAST (foo)' if you are
   subsequently dereferencing the value or don't feel comfortable doing a
   trick like `(void **) (void *) &foo'.

   Unfortunately, it does not work to just define the union type on the fly in
   the cast -- otherwise, we could avoid the need for a typedef.  Or rather,
   it does work under gcc but not under Visual C++.

   --ben
 */

#define ANSI_ALIASING_TYPEDEF(name, type) typedef union { char c; type p; } *ANSI_ALIASING_##name
#define ANSI_ALIASING_CAST(name, val) (((ANSI_ALIASING_##name) &(val))->p)
ANSI_ALIASING_TYPEDEF (voidp, void *);
/* VOIDP_CAST: Cast an lvalue to (void *) in a way that is ANSI-aliasing
   safe and will not result in GCC warnings.  The result is still an
   lvalue, so you can assign to it or take its address. */
#define VOIDP_CAST(l)  ANSI_ALIASING_CAST (voidp, l)

#endif /* INCLUDED_compiler_h */