* src/bytecode.c (exec_byte_code): Simplify, mostly by moving
initializers into decls, and by omitting some unnecessary
changes to ‘top’. This improves performance a bit on x86-64,
as it happens.
This improves performance overall on my benchmark on x86-64,
since the interpreted program-counter resides in a machine
register rather than in RAM.
* etc/DEBUG, src/.gdbinit: Remove xbytecode GDB command, as there
is no longer a byte stack to decode.
* src/bytecode.c (struct byte_stack, byte_stack_list)
(relocate_byte_stack): Remove. All uses removed.
(FETCH): Simplify now that pc is now local (typically, in a
register) and no longer needs to be relocated.
(CHECK_RANGE): Remove. All uses now done inline, in a different way.
(BYTE_CODE_QUIT): Remove; now done by op_relative_branch.
(exec_byte_code): Allocate a copy of the function’s bytecode,
so that there is no problem if GC moves it.
* src/lisp.h (struct handler): Remove byte_stack member.
All uses removed.
(SAFE_ALLOCA_LISP_EXTRA): New macro, a generalization of
SAFE_ALLOCA_LISP.
(SAFE_ALLOCA_LISP): Use it.
* src/bytecode.c (exec_byte_code): Remove MAX_ALLOCA-based limit
on bytecode maxdepth, by using SAFE_ALLOCA_LISP instead of alloca.
pipeline is fuller.
* src/bytecode.c (BYTE_CODE_QUIT): Check for GC, too. Do the
check only once every 256 times. This should be good enough, and
improves performance significantly on x86-64 as branch-prediction
typically assumes checking will not be done so the instruction
pipeline stays fuller.
(exec_byte_code): Set up the quit counter. Don’t call maybe_gc
directly, as BYTE_CODE_QUIT does that now.
* src/bytecode.c (BYTE_CODE_SAFE): Default to false, so that it
can be used outside #if. All uses of ‘defined BYTE_CODE_SAFE’
changed to ‘BYTE_CODE_SAFE’. Use BYTE_CODE_SAFE in plain
expressions instead of #if expressions when this is easy.
(struct byte_stack) [BYTE_CODE_SAFE]: Remove member ‘bottom’,
as it is no longer needed.
(exec_byte_code): Omit #if BYTE_CODE_SAFE when this is easy.
Simplify stack-overflow checking when BYTE_CODE_SAFE.
BYTE_CODE_METER hasn’t worked since 2013, and nobody seems to have
noticed. Perhaps we should remove it?
* src/bytecode.c (METER_2) [BYTE_CODE_METER]:
Use *aref_addr instead of AREF, since it needs to be an lvalue.
* src/xdisp.c (try_scrolling): Handle the case where the last
visible screen line of a window displays a before- or after-string
that takes up the whole screen line, and therefore there's no
place to display the cursor, even though the window does seem to
include the position of point. (Bug#24179)
* src/eval.c (call_debugger): Bind inhibit-changing-match-data to nil so
that debugger code that needs to do regexp match won't break
(Bug #23949, Bug #24166, Bug#16294).
d841a03 ; Spelling fix
a6ae479 Post AppDefined events from the main thread ONLY (bug#23934)
d35d398 Update to the AUTHORS file for Bob Weiner
4d2f4df Revert "Fix local printer set to left aligned string formatter."
cd1b4d6 Revert "Fix ses-delete-blanks to delete only blanks + documen...
f7ceb8e Revert "Fix English."
baa7abd Improve doc strings of 'gud-gdb' and 'gdb'
aa4271a Fix doc string of 'minibuffer-message-timeout'
b275cc7 Fix English.
3c97b0f Fix ses-delete-blanks to delete only blanks + documentation.
272391f profiler: document prefix arg for tree expansion
442cc39 Clarify usage of eshell-visual-options
b443c3c Fix comment in files-in-below-directory
8a38e94 Fix local printer set to left aligned string formatter.
0f0b191 ; Fix typos in NEWS
6bdf687 ; * etc/NEWS: Remove temporary marks
d4c6774 Fix missing point information in undo
3a9d629 Avoid crashes when buffer modification hooks clobber match data
178b2f5 Note combine-and-quote-strings doesn't shell quote
dec7567 Explain when package-initialize isn't called
113d1e2 Fix escaping in sh-indent-after-continuation docstr
80e2044 ; * etc/NEWS: Improve previous change.
5bb9e6c ; * etc/NEWS: Document how to avoid horizontal scroll bars.
38f4b8e Clarify the documentation of back-references in replacements
850ba44 Clarify lexical binding with symbol args behavior
f981b31 * lisp/net/tramp-sh.el (tramp-remote-path): Doc fix.
68fc964 * lisp/net/tramp-sh.el (tramp-remote-path): Doc fix. (Bug#23...
21110af Avoid assertion violations when rendering some fonts
6192b6c Document more details of package activation
272ddc6 Fixup warning message regarding HOME a bit more
43206d6 * lisp/leim/quail/indian.el ("bengali-probhat"): Change indic...
d41f7ff Fix input method "probhat" for Bengali
c150a64 ; Fix typo in commit before last
ebf0472 Add to elisp-completion-at-point's docstring
fd9fad0 Give more helpful warning about setting HOME
ea6b01d ; * lisp/term.el (term-mode): FIXME comment about bidi reorde...
# Conflicts:
# lisp/term.el
* src/gnutls.c [HAVE_GNUTLS]: Don’t include gnutls/gnutls.h
a second time; although it doesn’t hurt, it’s not needed.
* src/sysdep.c [!HAVE_GNUTLS]: Don’t include gnutls/crypto.h,
as it may not be available. Problem reported by Glenn Morris in:
http://lists.gnu.org/archive/html/emacs-devel/2016-08/msg00100.html
Although the problem is reported for OS X (Bug#23982), it seems to
be possible on other POSIXish platforms.
* src/gnutls.c (emacs_gnutls_nonblock_errno) [!WINDOWSNT]:
New function.
(emacs_gnutls_handshake) [!WINDOWSNT]:
Use it as the errno function, if non-blocking.
(Fgnutls_boot): Use GNUTLS_NONBLOCK if non-blocking.
* configure.ac, etc/NEWS: Require GnuTLS 2.12.2 or later,
instead of merely 2.6.6 or later (Bug#23982#20).
* src/gnutls.c (init_gnutls_functions, emacs_gnutls_handshake)
[WINDOWSNT]: Don’t bother with gnutls_check_version or
gnutls_transport_set_lowat, as they are no longer needed now that
we require gnutls 2.12.2 or later.
* src/fns.c (mapcar1): Return number of elements computed,
which can be less than LENI if the function alters the list.
All callers changed. (Bug#24118)
(sanitize_window_sizes): Adjust to new calling convention of
window--sanitize-window-sizes and only call it if defined.
(Fwindow__sanitize_window_sizes): Remove.
(syms_of_window): Don't defsubr it.
This fixes the following compilation warning:
regex.c:516:0: warning: macro "STREQ" is not used [-Wunused-macros]
#define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
^
* src/regex.c (STREQ): Remove unused macro. It should have been removed
in a [4538a5e: Refactor regex character class parsing in [:name:]]
commit but was mistakenly left out.
Emacs only ever uses its own regex syntax so support for other syntaxes
is never used. Hardcode the syntax so that the compilar can detect such
dead code and remove it from compiled code.
The only exception is RE_NO_POSIX_BACKTRACKING which can be separatelly
specified. Handle this separatelly with a function argument (replacing
now unnecessary syntax argument).
With this patchset, size of Emacs binary on x86_64 machine is reduced by
around 60 kB:
new-sizes:-rwx------ 3 mpn eng 30254720 Jul 27 23:31 src/emacs
old-sizes:-rwx------ 3 mpn eng 30314828 Jul 27 23:29 src/emacs
* src/regex.h (re_pattern_buffer): Don’t define syntax field #ifdef emacs.
(re_compile_pattern): Replace syntax with posix_backtracking argument.
* src/regex.c (print_compiled_pattern): Don’t print syntax #ifdef emacs.
(regex_compile): #ifdef emacs, replace syntax argument with
posix_backtracking which is now used instead of testing for
RE_NO_POSIX_BACKTRACKING syntax.
(re_match_2_internal): Don’t access bufp->syntax #ifndef emacs.
(re_compile_pattern): Replace syntax with posix_backtracking argument.
* src/search.c (compile_pattern_1): Pass boolean posix_backtracking
instead of syntax to re_compile_pattern.
Instead of using a global variable for storing regex syntax, pass it
to re_compile_pattern. This is only enabled when compiling Emacs (i.e.
‘#ifdef emacs’).
* src/regex.h (re_set_syntax): Declare only #ifndef emacs.
(re_compile_pattern): Now takes syntax argument #ifdef emacs.
* src/regex.c (re_syntax_options): Define only #ifndef emacs.
(re_compile_pattern): Use the new syntax argument #ifdef emacs.
* src/search.c (compile_pattern_1): Don’t use re_set_syntax and
instead pass syntax to re_compile_pattern directly.
There is no way to specify before_dot and after_dot opcodes in a regex
so code handling those ends up being dead. Remove it.
* src/regex.c (print_partial_compiled_pattern, regex_compile,
analyze_first, re_match_2_internal): Remove handling and references to
before_dot and after_dot opcodes.
re_wctype function is used in three separate places and in all of
those places almost exact code extracting the name from [:name:]
surrounds it. Furthermore, re_wctype requires a NUL-terminated
string, so the name of the character class is copied to a temporary
buffer.
The code duplication and unnecessary memory copying can be avoided by
pushing the responsibility of parsing the whole [:name:] sequence to
the function.
Furthermore, since now the function has access to the length of the
character class name (since it’s doing the parsing), it can take
advantage of that information in skipping some string comparisons and
using a constant-length memcmp instead of strcmp which needs to take
care of NUL bytes.
* src/regex.c (re_wctype): Delete function. Replace it with:
(re_wctype_parse): New function which parses a whole [:name:] string
and returns a RECC_* constant or -1 if the string is not of [:name:]
format.
(regex_compile): Use re_wctype_parse.
* src/syntax.c (skip_chars): Use re_wctype_parse.
* src/w32fns.c (Fw32_frame_geometry): Fix calculation of tool
bar width (Bug#24107). Reindent.
* src/xfns.c (frame_geometry): Fix calculation of tool bar
width on non-GTK builds.
* src/coding.c (coding_restore_undo_list): New function.
(decode_coding_gap, decode_coding_object): Temporarily set the
buffer's undo list to t, to avoid warnings when decoding changes
the buffer many times, e.g. due to post-read-conversion. Use
coding_restore_undo_list to make sure the undo list is restored.
(Bug#23813)
* src/regex.c (mutually_exclusive_p): Fix how whether character is
unibyte is tested when calling execute_charset function. This bug has
been introduced by [6dc6b00: Fix ‘[[:cc:]]*literal’ regex failing to
match ‘literal’] which dropped a call to IS_REAL_ASCII (c) macro.
Reinstitute it.
* src/indent.c (compute_motion): Don't turn on and don't use the
width cache unless the buffer's width-table is non-nil. This
avoids segfaults because code that uses the width cache assumes
the width-table exists. (Bug#24064)
The regex engine tries to optimise Kleene star by avoiding backtracking
when it can detect that star’s operand cannot match what follows it in
the pattern.
For example, when ‘[[:alpha:]]*1’ tries to match a ‘foo’, the engine
will test the longest match for ‘[[:alpha:]]*’, namely ’foo’ which is
the entire string. Literal digit one still present in the pattern will
however not match the remaining empty string.
Normally, backtracking would be performed trying a shorter match for the
character class (namely ‘fo’ leaving ‘o’ in the string), but since the
engine knows whatever would be put back into the string cannot possibly
match literal digit one so no backtracking will be attempted.
In the regexes of the form ‘[[:CC:]]*X’, the optimisation can be applied
if the character class CC does not match character X. In the above
example, this holds because digit one is not in alpha character class.
This test is performed by mutually_exclusive_p function but it did not
check class bits of a charset opcode. This resulted in an assumption
that character classes do not match multibyte characters. For example,
it would incorrectly conclude that [[:alpha:]] doesn’t match ‘ż’.
This, in turn, led to the aforementioned Kleene star optimisation being
incorrectly applied in patterns such as ‘[[:graph:]]*☠’ (which should
match ‘☠’ but doesn’t as can be tested by executing
(string-match-p "[[:graph:]]*☠" "☠")
which should return 0 but instead yields nil.
This issue affects any class witch matches multibyte characters, i.e.
if ‘[[:cc:]]’ matches a multibyte character X then ‘[[:cc:]]*X’ will
fail to match ‘X’.
* src/regex.c (executing_charset): A new function for executing the
charset and charset_not opcodes. It performs check on the character
taking into consideration existing bitmap, range table and class bits.
It also advances the pointer in the regex bytecode past the parsed
opcode.
(CHARSET_LOOKUP_RANGE_TABLE_RAW, CHARSET_LOOKUP_RANGE_TABLE): Removed.
Code now included in executing_charset.
(mutually_exclusive_p, re_match_2_internal): Changed to take advantage
of executing_charset function.
* test/src/regex-tests.el: New file with tests for the character class
matching.
Although for decades ‘signal’ has been documented to not return,
a corner case in the Lisp debugger causes ‘signal’ to return.
Remove the corner case and adjust Emacs internals accordingly.
An alternative would be to document the corner case, but this
would complicate the Lisp API unnecessarily. (Bug#24047)
* src/eval.c (signal_or_quit): New function, with most of the
old contents of Fsignal.
(quit): New function, which uses signal_or_quit and which
might return. All keyboard-based callers of Fsignal (Qquit,
Qnil) changed to use this new function instead.
(Fsignal): Use signal_or_quit. Now _Noreturn. All callers
changed.
(xsignal): Move to lisp.h.
* src/lisp.h (xsignal): Now an inline function, as it's now
just an alias for Fsignal.
It's important to adjust the match data in between calling
before-change-functions and after-change-functions, so that buffer
change hooks will always see match-data consistent with buffer content.
(Bug #23917)
* src/insdel.c (replace_range): Add new parameter ADJUST_MATCH_DATA, if
true call update_search_regs. Update all callers (except
Freplace_match) to pass 0 for the new parameter.
* src/search.c (update_search_regs): New function, extracted from
Freplace_match.
(Freplace_match): Remove match data adjustment code, pass 1 for
ADJUST_MATCH_DATA to replace_range instead.
Backport from master (Bug#24033).
Inspired by a suggestion by Florian Weimer in:
https://sourceware.org/ml/libc-alpha/2016-07/msg00425.html
* src/emacs.c (main) [__PPC64__]:
Special case for __PPC64__, which needs ASLR disabled in
dumped Emacs too.
The new approach is less selective, but also less intrusive on C code.
* src/bytecode.c, src/image.c, src/keyboard.c: Ignore -Wclobbered.
* src/conf_post.h (NONVOLATILE): Remove. All uses removed.
* src/callproc.c (child_setup): Use emacs_exec_file
so that ASLR is enabled in the child process.
* src/emacs.c: Move some personality details into sys/sysdep.c.
Do not include <sys/personality.h>.
(main): Disable ASLR earlier, so that we don’t chdir twice.
* src/lisp.h (disable_address_randomization): New decl.
* src/sysdep.c (disable_address_randomization)
[HAVE_PERSONALITY_ADDR_NO_RANDOMIZE]: Move personality details
here from emacs.c.
(emacs_exec_file): New function.
