* configure.ac (HAVE_STATEMENT_EXPRESSIONS): Remove.
For USE_STACK_LISP_OBJECTS, we always assume __GNUC__.
* lisp.h (union Aligned_Cons) [!GCALIGNED]: Define as such.
(SCOPED_CONS_INITIALIZER): New macro.
(scoped_cons) [USE_STACK_LISP_OBJECTS]: Use it.
(USE_LOCAL_ALLOCA): Remove.
(local_cons, local_list1, local_list2, local_list3, local_list4):
Remove. Stack overflow checking makes them too slow.
(make_local_vector): Likewise. Also we just don't have enough
users for it.
(enum LISP_STRING_OVERHEAD): Remove.
(local_string_init, local_vector_init): Remove prototypes.
(make_local_string, build_local_string): Redesign to target short
compile-time string constants, fall back to regular string allocation
where appropriate.
(lisp_string_size): New function.
(verify_ascii) [ENABLE_CHECKING]: Add prototype.
* alloc.c (local_string_init, local_vector_init): Remove.
(verify_ascii) [ENABLE_CHECKING]: New function.
* buffer.c, charset.c, chartab.c, data.c, editfns.c, emacs.c, fileio.c:
* fns.c, font.c, fontset.c, frame.c, keyboard.c, keymap.c, lread.c:
* menu.c, minibuf.c, process.c, textprop.c, xdisp.c, xfns.c, xfont.c:
* xselect.c, xterm.c: All related users changed.
* configure.ac (HAVE_STRUCT_ATTRIBUTE_ALIGNED)
(HAVE_STATEMENT_EXPRESSIONS): New configure-time checks.
* src/alloc.c (local_string_init, local_vector_init):
New functions, defined if USE_LOCAL_ALLOCATORS.
Mostly, these are moved here from lisp.h, as it's not
clear it's worth making them inline.
* src/lisp.h (USE_STACK_LISP_OBJECTS): Default to false.
(GCALIGNED): Depend on HAVE_STRUCT_ATTRIBUTE_ALIGNED and
USE_STACK_LISP_OBJECTS, not on a laundry list.
(local_string_init, local_vector_init): New decls.
(union Aligned_Cons): New type.
(scoped_cons): Use it. Give up on the char trick, as it's a too
much of a maintenance hassle; if someone wants this speedup
they'll just need to convince their compiler to align properly.
Conversely, use the speedup if struct Lisp_Cons happens to
be aligned even without a directive. Better yet, help it along
by using union Aligned_Cons rather than struct Lisp_Cons.
(pointer_valid_for_lisp_object): Remove. This check is not
necessary, since make_lisp_ptr is already doing it. All uses removed.
(local_vector_init, local_string_init): Move to alloc.c.
(build_local_vector): Remove this awkward macro, replacing with ...
(make_local_vector): New macro, which acts more like a function.
Use statement expressions and use __COUNTER__ to avoid macro
capture. Fall back on functions if these features are not supported.
(build_local_string, make_local_string): Likewise.
* lisp.h (build_local_string): Rename to ...
(make_local_string): ... this macro.
(build_local_string, scoped_list1, scoped_list3): New macros.
(toplevel) [USE_STACK_LISP_OBJECTS]: Define all new macros
and functions as such, use regular fallbacks otherwise.
* alloc.c (verify_alloca) [USE_STACK_LISP_OBJECTS]: Define
conditionally.
Respect MAX_ALLOCA and fall back to regular GC for large objects.
* character.h (parse_str_as_multibyte): Move prototype to ...
* lisp.h (parse_str_as_multibyte): ... here.
(struct Lisp_Cons): Add GCALIGNED attribute if supported.
(scoped_cons, scoped_list2, build_local_vector, build_local_string):
New macros.
(scoped_cons_init, pointer_valid_for_lisp_object, local_vector_init)
(local_string_init): New functions.
* alloc.c (verify_alloca) [ENABLE_CHECKING]: New function.
(init_alloc_once): Call it.
This follows up on the recent thread in emacs-devel on alloca; see:
http://lists.gnu.org/archive/html/emacs-devel/2014-09/msg00042.html
This patch also cleans up alloca-related glitches noted while
examining the code looking for unbounded alloca.
* alloc.c (listn):
* callproc.c (init_callproc):
Rewrite to avoid need for alloca.
* buffer.c (mouse_face_overlay_overlaps)
(report_overlay_modification):
* buffer.h (GET_OVERLAYS_AT):
* coding.c (make_subsidiaries):
* doc.c (Fsnarf_documentation):
* editfns.c (Fuser_full_name):
* fileio.c (Ffile_name_directory, Fexpand_file_name)
(search_embedded_absfilename, Fsubstitute_in_file_name):
* fns.c (Fmake_hash_table):
* font.c (font_vconcat_entity_vectors, font_update_drivers):
* fontset.c (fontset_pattern_regexp, Ffontset_info):
* frame.c (Fmake_terminal_frame, x_set_frame_parameters)
(xrdb_get_resource, x_get_resource_string):
* ftfont.c (ftfont_get_charset, ftfont_check_otf, ftfont_drive_otf):
* ftxfont.c (ftxfont_draw):
* image.c (xbm_load, xpm_load, jpeg_load_body):
* keyboard.c (echo_add_key, menu_bar_items, tool_bar_items):
* keymap.c (Fdescribe_buffer_bindings, describe_map):
* lread.c (openp):
* menu.c (digest_single_submenu, find_and_call_menu_selection)
(find_and_return_menu_selection):
* print.c (PRINTFINISH):
* process.c (Fformat_network_address):
* scroll.c (do_scrolling, do_direct_scrolling, scrolling_1):
* search.c (search_buffer, Fmatch_data, Fregexp_quote):
* sound.c (wav_play, au_play):
* syntax.c (skip_chars):
* term.c (tty_menu_activate, tty_menu_show):
* textprop.c (get_char_property_and_overlay):
* window.c (Fset_window_configuration):
* xdisp.c (safe__call, next_overlay_change, vmessage)
(compute_overhangs_and_x, draw_glyphs, note_mouse_highlight):
* xfaces.c (face_at_buffer_position):
* xmenu.c (x_menu_show):
Use SAFE_ALLOCA etc. instead of plain alloca, since the
allocation size isn't bounded.
* callint.c (Fcall_interactively): Redo memory_full check
so that it can be done at compile-time on some platforms.
* coding.c (MAX_LOOKUP_MAX): New constant.
(get_translation_table): Use it.
* callproc.c (call_process): Use SAFE_NALLOCA instead of
SAFE_ALLOCA, to catch integer overflows on size calculation.
(exec_failed) [!DOS_NT]: New function.
(child_setup) [!DOS_NT]: Use it.
* editfns.c (Ftranspose_regions):
Hoist USE_SAFE_ALLOC + SAFE_FREE out of 'if'.
* editfns.c (check_translation):
Allocate larger buffers on the heap.
* eval.c (internal_lisp_condition_case):
Check for MAX_ALLOCA overflow.
* fns.c (sort_vector): Use SAFE_ALLOCA_LISP rather than Fmake_vector.
(Fbase64_encode_region, Fbase64_decode_region):
Avoid unnecessary calls to SAFE_FREE before 'error'.
* buffer.c (mouse_face_overlay_overlaps):
* editfns.c (Fget_pos_property, check_translation):
* eval.c (Ffuncall):
* font.c (font_unparse_xlfd, font_find_for_lface):
* ftfont.c (ftfont_drive_otf):
* keyboard.c (echo_add_key, read_decoded_event_from_main_queue)
(menu_bar_items, tool_bar_items):
* sound.c (Fplay_sound_internal):
* xdisp.c (load_overlay_strings, dump_glyph_row):
Use an ordinary auto buffer rather than alloca, since the
allocation size is fixed and small.
* ftfont.c: Include <c-strcase.h>.
(matching_prefix): New function.
(get_adstyle_property): Use it, to avoid need for alloca.
* keyboard.c (echo_add_key):
* keymap.c (describe_map): Use ptrdiff_t, not int.
* keyboard.c (echo_add_key): Prefer sizeof to strlen.
* keymap.c (Fdescribe_buffer_bindings): Use SBYTES, not SCHARS,
when counting bytes.
* lisp.h (xlispstrdupa): Remove, replacing with ...
(SAFE_ALLOCA_STRING): ... new macro with different API.
This fixes a portability problem, namely, alloca result
passed to another function. All uses changed.
(SAFE_ALLOCA, SAFE_ALLOCA_LISP): Check for MAX_ALLOCA,
not MAX_ALLOCA - 1.
* regex.c (REGEX_USE_SAFE_ALLOCA, REGEX_SAFE_FREE)
(REGEX_ALLOCATE): New macros.
(REGEX_REALLOCATE, REGEX_ALLOCATE_STACK, REGEX_REALLOCATE_STACK)
(REGEX_FREE_STACK, FREE_VARIABLES, re_match_2_internal):
Use them.
* xdisp.c (message3): Use SAFE_ALLOCA_STRING rather than doing it
by hand.
(decode_mode_spec_coding): Store directly into buf rather than
into an alloca temporary and copying the temporary to the buf.
Fixes: debbugs:18410
* configure.ac (AC_CHECK_FUNCS): Check for qsort_r.
* src/fns.c (sort_vector, sort_vector_compare): New functions.
(sort_list): Likewise, refactored out of ...
(Fsort): ... adjusted user. Mention vectors in docstring.
(sort_vector_predicate) [!HAVE_QSORT_R]: New variable.
* src/alloc.c (make_save_int_obj): New function.
* src/lisp.h (enum Lisp_Save_Type): New member SAVE_TYPE_INT_OBJ.
(make_save_int_obj): Add prototype.
* test/automated/fns-tests.el (fns-tests-sort): New test.
dumping and the system malloc after dumping. (Bug#18222)
* configure.ac (HYBRID_MALLOC): New macro; define to use gmalloc
before dumping and the system malloc after dumping. Define on Cygwin.
* src/conf_post.h (malloc, realloc, calloc, free) [HYBRID_MALLOC]:
Define as macros, expanding to hybrid_malloc, etc.
(HYBRID_GET_CURRENT_DIR_NAME): New macro.
(get_current_dir_name) [HYBRID_GET_CURRENT_DIR_NAME]: Define as
macro.
* src/gmalloc.c: Set up the infrastructure for HYBRID_MALLOC, with a
full implementation on Cygwin. Remove Cygwin-specific code that
is no longer needed.
(malloc, realloc, calloc, free, aligned_alloc) [HYBRID_MALLOC]:
Redefine as macros expanding to gmalloc, grealloc, etc.
(DUMPED, ALLOCATED_BEFORE_DUMPING) [CYGWIN]: New macros.
(get_current_dir_name) [HYBRID_GET_CURRENT_DIR_NAME]: Undefine.
(USE_PTHREAD, posix_memalign) [HYBRID_MALLOC]: Don't define.
(hybrid_malloc, hybrid_calloc, hybrid_free, hybrid_realloc)
[HYBRID_MALLOC]:
(hybrid_get_current_dir_name) [HYBRID_GET_CURRENT_DIR_NAME]:
(hybrid_aligned_alloc) [HYBRID_MALLOC && (HAVE_ALIGNED_ALLOC ||
HAVE_POSIX_MEMALIGN)]: New functions.
* src/alloc.c (aligned_alloc) [HYBRID_MALLOC && (ALIGNED_ALLOC ||
HAVE_POSIX_MEMALIGN)]: Define as macro expanding to
hybrid_aligned_alloc; declare.
(USE_ALIGNED_ALLOC) [HYBRID_MALLOC && (ALIGNED_ALLOC ||
HAVE_POSIX_MEMALIGN)]: Define.
(refill_memory_reserve) [HYBRID_MALLOC]: Do nothing.
* src/sysdep.c (get_current_dir_name) [HYBRID_GET_CURRENT_DIR_NAME]:
Define as macro, expanding to gget_current_dir_name, and define
the latter.
* src/emacs.c (main) [HYBRID_MALLOC]: Don't call memory_warnings() or
malloc_enable_thread(). Don't initialize malloc.
* src/lisp.h (NONPOINTER_BITS) [CYGWIN]: Define (because GNU_MALLOC is
no longer defined on Cygwin).
(refill_memory_reserve) [HYBRID_MALLOC]: Don't declare.
* src/sheap.c (bss_sbrk_buffer_end): New variable.
* src/unexcw.c (__malloc_initialized): Remove variable.
* src/ralloc.c: Throughout, treat HYBRID_MALLOC the same as
SYSTEM_MALLOC.
* src/xdisp.c (decode_mode_spec) [HYBRID_MALLOC]: Don't check
Vmemory_full.
There is certainly nothing wrong with writing code like 'lo <= i
&& i <= hi', even if LO happens to a constant. There isn't even
anything wrong in general with writing 'a <= b' if A happens to
be a constant. At any rate stylistic changes shouldn't
be done like this without discussion.
src/dosfns.c (dos_memory_info): New function.
src/dosfns.h (dos_memory_info): Add prototype.
src/alloc.c (Fmemory_info) [MSDOS]: Call dos_memory_info.
src/vm-limit.c (get_lim_data) [MSDOS]: Call dos_memory_info, instead
of doing some of its job.
Linux 'sysinfo' function and 'struct sysinfo' type.
* src/alloc.c (Fmemory_info): New function.
* lisp/files.el (warn-maybe-out-of-memory): New function.
(find-file-noselect): Use it.
* lisp.h (struct Lisp_Sub_Char_Table): Use C integers for depth and
min_char slots. Adjust comment.
(enum char_table_specials): Rename from CHAR_TABLE_STANDARD_SLOTS.
Add SUB_CHAR_TABLE_OFFSET member.
(make_uninit_sub_char_table): New function.
* alloc.c (mark_char_table): Add extra argument to denote char table
subtype. Adjust to match new layout of sub char-table.
(mark_object): Always mark sub char-tables with mark_char_table.
* chartab.c (make_sub_char_table, copy_sub_char_table)
(sub_char_table_ref, sub_char_table_ref_and_range, sub_char_table_set)
(sub_char_table_set_range, optimize_sub_char_table, map_sub_char_table)
(map_sub_char_table_for_charset, uniprop_table_uncompress):
All related users changed.
* lread.c (read1): Adjust to match new layout of sub char-table.
Now the mark_object's stack is just 32 bytes on a 64-bit
system, which means extra 20% off the stack usage.
* alloc.c (mark_save_value): As before, refactored out from ...
(mark_object): ... adjusted user. Also add comment.
This change moves a few cold paths from mark_object to NO_INLINE
functions and adjusts symbol marking loop. According to GCC 4.8.2
-Wstack-usage, this reduces mark_object's stack usage from 80 to
48 bytes on a 64-bit system. For a long byte-force-recompile runs,
stack usage at the mark phase is reduced up to 28%. Surprisingly,
it also gains up to 3% in speed (with default '-O2 -g3' flags).
* alloc.c (mark_compiled, mark_localized_symbol): New functions,
refactored out from ...
(mark_object): ... adjusted user. Also mark symbols in a tight
inner loop.
(mark_face_cache): Add NO_INLINE.
This doesn't fix Bug#17598, but it does fix a regression since Emacs
built with xlc until recently, and perhaps it'll help us fix Bug#17598.
* configure.ac (GC_SETJMP_WORKS): Define for AIX, too.
Merge from gnulib, incorporating:
2014-05-30 ftoastr: work around compiler bug in IBM xlc 12.1
* lib/ftoastr.c: Update from gnulib.
* src/alloc.c (gdb_make_enums_visible): Remove FLOAT_TO_STRING_BUFSIZE.
* src/conf_post.h (FLEXIBLE_ARRAY_MEMBER) [__IBMC__]: Don't define to empty.
* src/lisp.h (FLOAT_TO_STRING_BUFSIZE): Make it a macro, instead of an enum,
to work around a compiler bug in IBM xlc 12.1.
relocatable data of non-pure Lisp string.
* alloc.c (maybe_lisp_pointer): New function, refactored out of ...
(mark_maybe_pointer): ... adjusted user.
(relocatable_string_data_p): New function.
* lisp.h (relocatable_string_data_p): Add prototype.
* xdisp.c (message_with_string): If ENABLE_CHECKING, make sure
the pointer to relocatable Lisp data is not used.
src/alloc.c (garbage_collect_1): New function, with all of the guts
of Fgarbage_collect.
(mark_stack): Accept an argument END and don't mark Lisp objects
on the stack beyond the address given by END. Calculation of END
was moved to Fgarbage_collect.
(Fgarbage_collect): Calculate the end address of the stack portion
that needs to be examined by mark_stack, and pass that address to
garbage_collect_1, which will pass it to mark_stack. See
http://lists.gnu.org/archive/html/emacs-devel/2014-05/msg00270.html
for more details about the underlying problems. In particular,
this avoids dumping Emacs with the large hash-table whose value is
held in purify-flag for most of the time loadup.el runs.
Likewise for xd_signature in dbusbind.c.
This is more consistent with the usual practice in Emacs, which is
that any non-nil value counts as true.
* doc/lispref/sequences.texi (Bool-Vectors): Coalesce discussion of how to
print them. bool-vector's args need not be t or nil.
* src/alloc.c (Fbool_vector): Don't require args to be t or nil.
* src/dbusbind.c (xd_signature): Likewise, for booleans.
* src/data.c, lisp.h (Qbooleanp):
* src/lisp.h (CHECK_BOOLEAN): Remove. All uses removed.
* alloc.c (struct cons_block, struct float_block): On 64-bit hosts,
bits_word is typically a tad more efficient for mark bits than
unsigned is, so use bits_word. All uses changed.
* lisp.h (BITS_PER_INT): Remove; no longer used.
This ports to GCC 4.9.0 with -fsanitize=undefined.
* alloc.c (bool_vector_fill, SETMARKBIT, UNSETMARKBIT):
* data.c (Fash):
* regex.c (extract_number):
* lisp.h (make_number, XINT):
Do not shift a 1 bit left into a sign bit.
* alloc.c (struct cons_block, struct float_block): Use unsigned,
not int, for gcmarkbits. All uses changed.
The DATA_SEG_BITS hack was needed for older 32 bit platforms.
As a result of this change, Emacs won't work on IRIX 6.5 with IRIX
cc, but that platform is so old that SGI itself stopped supporting
it in December 2013. If you still need Emacs on IRIX, please
either compile with GCC and port the undumping code, or run
'./configure --with-wide-int'.
* configure.ac (DATA_SEG_BITS): Remove. All uses removed.
* src/alloc.c (gdb_make_enums_visible): Update to match lisp.h.
* src/lisp.h (GCTYPEBITS): Move definition up, and switch to the
DEFINE_GDB_SYMBOL_START way to define it.
(NONPOINTER_BITS): New macro.
(EMACS_INT): Use it.
[!USE_LSB_TAG && !WIDE_EMACS_INT]: Fail, and suggest reporting
the problem and/or configuring --with-wide-int.
(USE_LSB_TAG): Simplify, based on above changes.
(gdb_DATA_SEG_BITS): Remove. All uses removed.
* src/vm-limit.c (exceeds_lisp_ptr): Remove. All uses removed.
This port requires IRIX cc, as I did not have time to get
undump working with the old GCC on the system I had access to,
but that's better than nothing.
* configure.ac (gl_GCC_VERSION_IFELSE): Remove unused macro
that wouldn't have worked anyway, with IRIX cc.
(emacs_cv_clang, emacs_cv_sanitize_address)
(ns_osx_have_104, ns_osx_have_105):
Don't assume '#error' makes the compiler fail,
as this doesn't work with IRIX cc.
(CFLAGS, LIBS): Don't let the GnuTLS results infect later 'configure'
checks. This runs afoul of an IRIX configuration where GnuTLS is
in an optional library that also contains getdelim, and causes
a later 'configure' to incorrectly think getdelim is supported.
* src/alloc.c (TAGGABLE_NULL): New constant,
for porting to hosts with nontrivial DATA_SEG_BITS settings.
(next_vector, set_next_vector): Use it.
* src/conf_post.h (INET6) [IRIX6_5]: Define.
(HAVE_GETADDRINFO) [IRIX6_5]: Undef.
* src/data.c (BITS_PER_ULL): Don't assume ULLONG_MAX is defined.
* src/lisp.h (lisp_h_XPNTR): Don't OR in bits that aren't masked out,
for consistency with how TAGGABLE_NULL is computed.
Fixes: debbugs:9684
(symbol_block_pinned): New var.
(Fmake_symbol): Initialize `pinned'.
(purecopy): New function, extracted from Fpurecopy. Mark symbols as
pinned and signal an error for un-purifiable objects.
(pure_cons): Use it.
(Fpurecopy): Use it, except for objects that can't be purified.
(mark_pinned_symbols): New function.
(Fgarbage_collect): Use it.
(gc_sweep): Remove hack made unnecessary.
* src/lisp.h (struct Lisp_Symbol): New bitfield `pinned'.
* alloc.c: Include execinfo.h if available.
(SUSPICIOUS_OBJECT_CHECKING): New macro; define unconditionally.
(suspicious_free_record): New structure.
(suspicious_objects,suspicious_object_index)
(suspicious_free_history, suspicious_free_history_index): New
variables.
(find_suspicious_object_in_range,detect_suspicious_free,Fsuspicious_object):
New functions.
(cleanup_vector): Call find_suspicious_object_in_range.
