Support non-ASCII correctly in character classes, test this. src/ChangeLog addition: 2012-04-21 Aidan Kehoe <kehoea@parhasard.net> Support non-ASCII correctly in character classes ([:alnum:] and friends). * regex.c: * regex.c (ISBLANK, ISUNIBYTE): New. Make these and friends independent of the locale, since we want them to be consistent in XEmacs. * regex.c (print_partial_compiled_pattern): Print the flags for charset_mule; don't print non-ASCII as the character values in ranges, this breaks with locales. * regex.c (enum): Define various flags the charset_mule and charset_mule_not opcodes can now take. * regex.c (CHAR_CLASS_MAX_LENGTH): Update this. * regex.c (re_iswctype, re_wctype): New, from GNU. * regex.c (re_wctype_can_match_non_ascii): New; used when deciding on whether to use charset_mule or the ASCII-only regex character set opcode. * regex.c (regex_compile): Error correctly on long, non-existent character class names. Break out the handling of charsets that can match non-ASCII into a separate clause. Use compile_char_class when compiling character classes. * regex.c (compile_char_class): New. Used in regex_compile when compiling character sets that may match non-ASCII. * regex.c (re_compile_fastmap): If there are flags set for charset_mule or charset_mule_not, we can't use the fastmap (since we need to check syntax table values that aren't available there). * regex.c (re_match_2_internal): Check the new flags passed to the charset_mule{,_not} opcode, observe them if appropriate. * regex.h: * regex.h (enum): Expose re_wctype_t here, imported from GNU. tests/ChangeLog addition: 2012-04-21 Aidan Kehoe <kehoea@parhasard.net> * automated/regexp-tests.el: * automated/regexp-tests.el (Assert-char-class): Check that #'string-match errors correctly with an over-long character class name. Add tests for character class functionality that supports non-ASCII characters. These tests expose bugs in GNU Emacs 24.0.94.2, but pass under current XEmacs.

@c -*-texinfo-*-
@setfilename ../../info/index.info

@c Indexing guidelines

@c I assume that all indexes will be combined.
@c Therefore, if a generated findex and permutations
@c cover the ways an index user would look up the entry,
@c then no cindex is added.
@c Concept index (cindex) entries will also be permuted.  Therefore, they
@c have no commas and few irrelevant connectives in them.

@c I tried to include words in a cindex that give the context of the entry,
@c particularly if there is more than one entry for the same concept.
@c For example, "nil in keymap"
@c Similarly for explicit findex and vindex entries, e.g. "print example".

@c Error codes are given cindex entries, e.g. "end-of-file error".

@c pindex is used for .el files and Unix programs

@node Index,  , Standard Hooks, Top
@unnumbered Index

@ignore
All variables, functions, keys, programs, files, and concepts are
in this one index.

All names and concepts are permuted, so they appear several times, one
for each permutation of the parts of the name.  For example,
@code{function-name} would appear as @b{function-name} and @b{name,
function-}.  Key entries are not permuted, however.
@end ignore

@c Print the indices

@printindex fn