Without this fix, (substitute-object-in-subtree #0=(#0# 'a) 'a)
would dump core, since the C code would recurse indefinitely through
the infinite structure. This patch adds an argument to the function,
and renames it to lread--substitute-object-in-subtree as the function
is not general-purpose and should not be relied on by outside code.
See Bug#23660.
* src/intervals.c (traverse_intervals_noorder): ARG is now void *,
not Lisp_Object, so that callers need not cons unnecessarily.
All callers changed. Also, remove related #if-0 code that was
“temporary” in the early 1990s and has not been compilable for
some time.
* src/lread.c (struct subst): New type, for substitution closure data.
(seen_list): Remove this static var, as this info is now part of
struct subst. All uses removed.
(Flread__substitute_object_in_subtree): Rename from
Fsubstitute_object_in_subtree, and give it a 3rd arg so that it
doesn’t dump core when called from the top level with an
already-cyclic structure. All callers changed.
(SUBSTITUTE): Remove. All callers expanded and then simplified.
(substitute_object_recurse): Take a single argument SUBST rather
than a pair OBJECT and PLACEHOLDER, so that its address can be
passed around as part of a closure; this avoids the need for an
AUTO_CONS call. All callers changed. If the COMPLETED component
is t, treat every subobject as potentially circular.
(substitute_in_interval): Take a struct subst * rather than a
Lisp_Object, for the closure data. All callers changed.
* test/src/lread-tests.el (lread-lread--substitute-object-in-subtree):
New test, to check that the core dump does not reoccur.
When configured with --without-ns, HAVE_NS is not defined on macOS,
thus 'memory-limit' calls the deprecated sbrk(2) function. Avoid that
by using the pre-defined __APPLE__ preprocessor macro.
* src/alloc.c (Fmemory_limit): Never use sbrk(2) on macOS.
Instead of a lambda, create a new type containing all data required to
call the function, and support it in the evaluator. Because this type
now also needs to store the function documentation, it is too big for
Lisp_Misc; use a pseudovector instead. That also has the nice benefit
that we don't have to add special support to the garbage collector.
Since the new type is user-visible, give it a predicate.
Now we can easily support 'help-function-args' and 'func-arity'; add
unit tests for these.
* src/lisp.h (allocate_module_function, MODULE_FUNCTIONP)
(XMODULE_FUNCTION): New pseudovector type 'module function'.
* src/eval.c (FUNCTIONP): Also treat module functions as functions.
(funcall_lambda, Ffuncall, eval_sub): Add support for calling module
functions.
(Ffunc_arity): Add support for detecting the arity of module
functions.
* src/emacs-module.c (module_make_function): Adapt to new structure.
Return module function object directly instead of wrapping it in a
lambda; remove FIXME.
(funcall_module): New function to call module functions. Replaces
`internal--module-call' and is called directly from eval.c.
(syms_of_module): Remove internal helper function, which is no longer
needed.
(module_function_arity): New helper function.
* src/data.c (Ftype_of): Adapt to new implementation.
(Fmodule_function_p, syms_of_data): New user-visible function. Now
that module functions are first-class objects, they deserve a
predicate. Define it even if not compiled with --enable-modules so
that Lisp code doesn't have to check for the function's existence.
* src/doc.c (Fdocumentation): Support module functions.
* src/print.c (print_object): Adapt to new implementation.
* src/alloc.c (mark_object): Specialized garbage collector support is
no longer needed.
* lisp/help.el (help-function-arglist): Support module functions.
While there, simplify the arity calculation by using `func-arity',
which does the right thing for all kinds of functions.
* test/data/emacs-module/mod-test.c: Amend docstring so we can test
the argument list.
* test/src/emacs-module-tests.el (mod-test-sum-docstring): Adapt to
new docstring.
(mod-test-non-local-exit-signal-test): Because `internal--module-call'
is gone, the backtrace has changed and no longer leaks the
implementation.
(module--func-arity): New test for `func-arity'.
(module--help-function-arglist): New test for `help-function-arglist'.
This resolves a couple of FIXMEs in emacs-module.c.
* src/lisp.h (MODULE_FUNCTIONP, XMODULE_FUNCTION): New functions.
* src/alloc.c (make_module_function): New function.
(mark_object): GC support.
* src/data.c (Ftype_of, syms_of_data): Handle module function type.
* src/print.c (print_object): Print support for new type.
* src/emacs-module.c (module_make_function, Finternal_module_call):
Use new module function type, remove FIXMEs.
(module_format_fun_env): Adapt and give it external linkage.
* test/src/emacs-module-tests.el (module-function-object): Add unit
test.
* src/alloc.c (GC_CHECK_MARKED_OBJECTS): Define to 1 by default of
ENABLE_CHECKING is defined.
(mark_object): Test for GC_CHECK_MARKED_OBJECTS being non-zero,
instead of being defined.
Since copy-sequence seems to be needed anyway for records, have it
work on records, and remove copy-record as being superfluous.
* doc/lispref/records.texi (Records, Record Functions):
* lisp/emacs-lisp/cl-macs.el (cl-defstruct):
* lisp/emacs-lisp/eieio.el (make-instance, clone):
* test/src/alloc-tests.el (record-3):
Use copy-sequence, not copy-record, to copy records.
* doc/lispref/sequences.texi (Sequence Functions)
(Array Functions): Document that aref and copy-sequence
work on records.
* etc/NEWS: Omit copy-record.
* src/alloc.c (Fcopy_record): Remove.
* src/data.c (Faref): Document that arg can be a record.
* src/fns.c (Fcopy_sequence): Copy records, too.
* objects.texi (Record Type): improve description of what
`type-of' returns for records.
(Type Descriptors): new section.
* elisp.texi: reference it.
* records.texi (Records): reference it. Document behaviour when type
slot is a record.
* alloc.c (Fmake_record, Frecord): mention type desciptors.
* doc/lispref/records.texi (Records): Mention size limit.
* etc/NEWS: Mention records.
* src/alloc.c (allocate_pseudovector, allocate_record):
Prefer 'PSEUDOVECTOR_SIZE_MASK' to its definiens.
(allocate_record): Check arg range here, not in callers, as this
simplifies the code. Use allocate_vectorlike instead of
allocate_vector, to avoid duplicate runtime tests.
(Fmake_record, record): Don't mention PSEUDOVECTOR_SIZE_BITS in
the doc string, as it is not visible to the user.
(Fmake_record, record, Fcopy_record):
Prefer make_lisp_ptr to XSETVECTOR.
(record): Broaden memcpy to copy the type, too.
* src/alloc.c (allocate_record): New function.
(Fmake_record, Frecord, Fcopy_record): New functions.
(syms_of_alloc): defsubr them.
(purecopy): Work with records.
* src/data.c (Ftype_of): Return slot 0 for record objects, or type
name if record's type holds class.
(Frecordp): New function.
(syms_of_data): defsubr it. Define `Qrecordp'.
(Faref, Faset): Work with records.
* src/fns.c (Flength): Work with records.
* src/lisp.h (prec_type): Add PVEC_RECORD.
(RECORDP, CHECK_RECORD, CHECK_RECORD_TYPE): New functions.
* src/lread.c (read1): Add syntax for records.
* src/print.c (PRINT_CIRCLE_CANDIDATE_P): Add RECORDP.
(print_object): Add syntax for records.
* test/lisp/emacs-lisp/cl-print-tests.el (cl-print-tests-2):
New test.
* test/src/alloc-tests.el (record-1, record-2, record-3):
New tests.
* doc/lispref/elisp.texi, doc/lispref/objects.texi,
doc/lispref/records.texi: Add documentation for records.
* src/lisp.h (PSEUDOVECTOR_TYPE): New function, extracted from mark_object.
(PSEUDOVECTOR_TYPEP): Change type of `code'.
* src/alloc.c (sweep_vectors): Remove out-of-date assertion.
(mark_object): Use PSEUDOVECTOR_TYPE.
* src/data.c (Ftype_of): Use switch on pvec type.
* src/print.c (print_object): Use switch on pvec type.
* lisp/emacs-lisp/cl-generic.el (cl--generic-typeof-types):
Add recently added types.
* src/alloc.c (purecopy_hash_table):
* src/fns.c (maybe_resize_hash_table, Fmake_hash_table):
(Fhash_table_rehash_size):
* src/lisp.h (struct Lisp_Hash_Table.rehash_size):
The rehash_size member of struct Lisp_Hash_Table is now a
float, not a Lisp_Object.
* src/alloc.c (purecopy_hash_table): Assign members in order.
* src/fns.c (make_hash_table): Use EMACS_INT for size and
float for rehash_size, instead of Lisp_Object for both.
All callers changed.
* src/lisp.h (DEFAULT_REHASH_SIZE): Now float, not double,
and 1 smaller.
* src/print.c (print_object): Simplify by calling
Fhash_table_rehash_size and Fhash_table_rehash_threshold.
Avoid unnecessary NILP.
* src/alloc.c (purecopy_hash_table): Assign, don’t purecopy.
* src/fns.c (set_hash_next_slot, set_hash_index_slot): Hash index
arg is now ptrdiff_t index (or -1 if empty), not Lisp_Object
integer (or Qnil if empty). All callers changed.
(larger_vecalloc): New static function.
(larger_vector): Use it.
(HASH_NEXT, HASH_INDEX): Move here from lisp.h. Return ptrdiff_t
index (or -1) not Lisp_Object integer (or Qnil). All callers changed.
* src/fns.c (make_hash_table, maybe_resize_hash_table, hash_lookup)
(hash_put, hash_remove_from_table, hash_clear, sweep_weak_table):
* src/profiler.c (evict_lower_half, record_backtrace):
-1, not nil, is now the convention for end of collision list.
* src/fns.c (maybe_resize_hash_table): Avoid double-initialization
of the free list. Reallocate H->next last, in case other
reallocations exhaust memory.
* src/lisp.h (struct Lisp_Hash_Table): ‘next_free’ is now
ptrdiff_t, not Lisp_Object. Adjust commentary for ‘next’ and
‘index’, which no longer contain nil.
(HASH_NEXT, HASH_INDEX): Move to src/fns.c.
* src/lisp.h (struct Lisp_Hash_Table): Change type of
`rehash_threshold' and `pure' fields and move them after `count'.
* src/fns.c (make_hash_table): Change type of `rehash_threshold' and `pure'.
(Fmake_hash_table, Fhash_table_rehash_threshold):
* src/category.c (hash_get_category_set):
* src/xterm.c (syms_of_xterm):
* src/profiler.c (make_log):
* src/print.c (print_object):
* src/alloc.c (purecopy_hash_table, purecopy): Adjust accordingly.
alloc.c had bitrotted a bit, and used an undefined symbol
stack_base when Emacs was built on Solaris sparc, leading to
compilation failures. Also, code related to __builtin_unwind_init
was unnecessarily duplicated. Fix the bitrot and remove some
duplication.
* src/alloc.c: Remove uses of GC_SAVE_REGISTERS_ON_STACK, since it
is never defined.
(test_setjmp) [!HAVE___BUILTIN_UNWIND_INIT && GC_SETJMP_WORKS]:
Define a no-op dummy, to simplify use.
(test_setjmp) [!GC_SETJMP_WORKS]: Test setjmp_tested_p here rather
than in the caller, to simplify use.
(stacktop_sentry): New type.
(__builtin_unwind_init) [!HAVE___BUILTIN_UNWIND_INIT]: New macro.
(SET_STACK_TOP_ADDRESS): New macro, containing code that was duplicated.
(flush_stack_call_func, Fgarbage_collect): Use it.
(init_alloc): Omit unnecessary initialization.
After dumping, Emacs need not re-test setjmp.
This fixes some infinite loops that cannot be quitted out of,
e.g., (defun foo () (nth most-positive-fixnum '#1=(1 . #1#)))
when byte-compiled and when run under X. See:
http://lists.gnu.org/archive/html/emacs-devel/2017-01/msg00577.html
This also attempts to keep the performance improvements I recently
added, as much as possible under the constraint that the infloops
must be caught. In some cases this fixes infloop bugs recently
introduced when I removed immediate_quit.
* src/alloc.c (Fmake_list):
Use rarely_quit, not maybe_quit, for speed in the usual case.
* src/bytecode.c (exec_byte_code):
* src/editfns.c (Fcompare_buffer_substrings):
* src/fns.c (Fnthcdr):
* src/syntax.c (scan_words, skip_chars, skip_syntaxes)
(Fbackward_prefix_chars):
Use rarely_quit so that users can C-g out of long loops.
* src/callproc.c (call_process_cleanup, call_process):
* src/fileio.c (read_non_regular, Finsert_file_contents):
* src/indent.c (compute_motion):
* src/syntax.c (scan_words, Fforward_comment):
Remove now-unnecessary maybe_quit calls.
* src/callproc.c (call_process):
* src/doc.c (get_doc_string, Fsnarf_documentation):
* src/fileio.c (Fcopy_file, read_non_regular, Finsert_file_contents):
* src/lread.c (safe_to_load_version):
* src/sysdep.c (system_process_attributes) [GNU_LINUX]:
Use emacs_read_quit instead of emacs_read in places where
C-g handling is safe.
* src/eval.c (maybe_quit): Move comment here from lisp.h.
* src/fileio.c (Fcopy_file, e_write):
Use emacs_write_quit instead of emacs_write_sig in places where
C-g handling is safe.
* src/filelock.c (create_lock_file): Use emacs_write, not
plain write, as emacs_write no longer has a problem.
(read_lock_data): Use emacs_read, not read, as emacs_read
no longer has a problem.
* src/fns.c (rarely_quit): Move to lisp.h and rename to
incr_rarely_quit. All uses changed..
* src/fns.c (Fmemq, Fmemql, Fassq, Frassq, Fplist_put, Fplist_member):
* src/indent.c (compute_motion):
* src/syntax.c (find_defun_start, back_comment, forw_comment)
(Fforward_comment, scan_lists, scan_sexps_forward):
Use incr_rarely_quit so that users can C-g out of long loops.
* src/fns.c (Fnconc): Move incr_rarely_quit call to within
inner loop, so that it catches C-g there too.
* src/keyboard.c (tty_read_avail_input): Remove commented-out
and now-obsolete code dealing with interrupts.
* src/lisp.h (rarely_quit, incr_rarely_quit): New functions,
the latter moved here from fns.c and renamed from rarely_quit.
(emacs_read_quit, emacs_write_quit): New decls.
* src/search.c (find_newline, search_buffer, find_newline1):
Add maybe_quit to catch C-g.
* src/sysdep.c (get_child_status): Always invoke maybe_quit
if interruptible, so that the caller need not bother.
(emacs_nointr_read, emacs_read_quit, emacs_write_quit):
New functions.
(emacs_read): Rewrite in terms of emacs_nointr_read.
Do not handle C-g or signals; that is now for emacs_read_quit.
(emacs_full_write): Replace PROCESS_SIGNALS two-way arg
with INTERRUPTIBLE three-way arg. All uses changed.
* src/alloc.c
(purecopy_hash_table) New function, makes a copy of the given hash
table in pure storage.
Add new struct `pinned_object' and `pinned_objects' linked list for
pinning objects.
(Fpurecopy) Allow purifying hash tables
(purecopy) Pin hash tables that are either weak or not declared with
`:purecopy t`, use purecopy_hash_table otherwise.
(marked_pinned_objects) New function, marks all objects in pinned_objects.
(garbage_collect_1) Use it. Mark all pinned objects before sweeping.
* src/lisp.h Add new field `pure' to struct `Lisp_Hash_Table'.
* src/fns.c: Add `purecopy' parameter to hash tables.
(Fmake_hash_table): Check for a `:purecopy PURECOPY' argument, pass it
to make_hash_table.
(make_hash_table): Add `pure' parameter, set h->pure to it.
(Fclrhash, Fremhash, Fputhash): Enforce that the table is impure with
CHECK_IMPURE.
* src/lread.c: (read1) Parse for `purecopy' parameter while reading
hash tables.
* src/print.c: (print_object) add the `purecopy' parameter while
printing hash tables.
* src/category.c, src/emacs-module.c, src/image.c, src/profiler.c,
src/xterm.c: Use new (make_hash_table).
There’s no longer need to have QUIT stand for a slug of C statements.
Use the more-obvious function-call syntax instead.
Also, use true and false when setting immediate_quit.
These changes should not affect the generated machine code.
* src/lisp.h (QUIT): Remove. All uses replaced by maybe_quit.
This avoids the confusion of using two different phrases "main thread"
and "primary thread" internally to mean the same thing. See:
http://lists.gnu.org/archive/html/emacs-devel/2016-12/msg01142.html
* src/thread.c (main_thread): Rename from primary_thread,
since the new name no longer clashes with main_thread_id
and Emacs internals normally call this the "main thread".
(init_main_thread): Rename from init_primary_thread.
(main_thread_p): Rename from primary_thread_p.
All uses changed.
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.
* src/thread.c (unmark_threads): Remove. All uses removed.
* src/thread.h (struct thread_state): Remove m_byte_stack_list member.
All uses removed. m_stack_bottom is now the first non-Lisp field.
* src/alloc.c (mark_object) [GC_CHECK_MARKED_OBJECTS]: Don't abort
for thread objects. They are marked via the all_threads list, and
therefore don't need to be inserted into the red-black tree, so
mem_find will never find them. Reported by Daniel Colascione
<dancol@dancol.org> in
http://lists.gnu.org/archive/html/emacs-devel/2016-12/msg00817.html.
Backport from master. Sun C 5.14 supports C11 but not GCC
extensions, and so refuses to compile Emacs without this patch.
* src/alloc.c (lmalloc, lrealloc): Don't use INT_ADD_WRAPV on
size_t, as in general this macro is restricted to signed types.
This simplifies the code a bit, and makes the structs more
shareable and less likely to become corrupt.
* src/alloc.c (cleanup_vector):
* src/font.c (valid_font_driver, font_prepare_cache)
(font_finish_cache, font_get_cache, font_clear_cache)
(register_font_driver, font_update_drivers):
* src/font.h (struct font, struct font_driver_list)
(valid_font_driver):
struct font_drivers are now const.
* src/font.c, src/ftcrfont.c, src/ftfont.c, src/nsfont.m, src/xfont.c:
Omit no-longer-necessary decls.
* src/ftcrfont.c (syms_of_ftcrfont):
* src/ftxfont.c (syms_of_ftxfont):
* src/xftfont.c (syms_of_xftfont):
Omit no-longer-necessary initialization code.
* src/ftcrfont.c (ftcrfont_driver):
* src/ftfont.c (ftfont_driver):
* src/ftxfont.c (ftxfont_driver):
* src/macfont.m (macfont_driver):
* src/nsfont.m (nsfont_driver):
* src/xfont.c (xfont_driver):
* src/xftfont.c (xftfont_driver):
Use C99-style initializer for ease of maintenance, and make it const.
* src/ftcrfont.c, src/ftxfont.c, src/xftfont.c:
Refer to functions like ftfont_text_extents directly.
* src/ftfont.c (ftfont_get_cache, ftfont_list, ftfont_list_family)
(ftfont_has_char, ftfont_encode_char, ftfont_text_extents)
(ftfont_get_bitmap, ftfont_anchor_point, ftfont_otf_capability)
(ftfont_variation_glyphs, ftfont_filter_properties)
(ftfont_combining_capability):
* src/xfont.c (xfont_get_cache):
Now extern, so that other modules’ struct font_drivers can use
them directly.
* src/macfont.m (macfont_descriptor_entity):
* src/nsfont.m (nsfont_open):
Use constant directly; this is clearer.
Clean up some of the bitrot affecting the CANNOT_DUMP code. This
lets the build succeed again, and fixes the testing framework so
that most test cases now pass. About twenty test cases still
fail, though, and we still have Bug#24974.
* configure.ac (CANNOT_DUMP): Now empty if CANNOT_DUMP.
(SYSTEM_MALLOC): Now true if CANNOT_DUMP. There should no longer
be any point to messing with a private memory allocator unless
Emacs is dumping.
* src/alloc.c (alloc_unexec_pre, alloc_unexec_post, check_pure_size):
* src/image.c (reset_image_types):
* src/lastfile.c (my_endbss, _my_endbss, my_endbss_static):
Do not define if CANNOT_DUMP.
* src/emacs.c (might_dump) [CANNOT_DUMP]: Now always false and local.
(daemon_pipe) [!WINDOWSNT]: Now static.
* test/Makefile.in (mostlyclean): Remove *.tmp files.
(make-test-deps.mk): Elide CANNOT_DUMP chatter.
Problem reported by Ashish Shukla (Bug#24892). I tested
this on FreeBSD 11 x86-64 with HAVE_SBRK manually undefined.
* configure.ac (system_malloc): Set to 'yes' if there is no sbrk.
(sbrk): Check whether it exists.
* src/alloc.c (my_heap_start) [!GNU_LINUX]:
Do not define, since this function is now used only on GNU/Linux,
and sbrk might not exist on other platforms.
(malloc_initialize_hook) [!GNU_LINUX]:
Do not call my_heap_start, since its side effect will never be used.
(Fmemory_limit) [!HAVE_SBRK]: Do not call sbrk.
* src/unexelf.c (unexec) [!HAVE_SBRK]: Assume that nothing like
sbrk exists.
2deb20c Port last_marked change to full-program optimizers
27eb668 Adapt GDB scripts to '--enable-check-lisp-object-type' builds
dac64e3 Avoid optimizing out the last_marked[] array
4f406e9 CC Mode manual: remove reference to former Emacs variable las...
44e402e Allow to disable compaction of font caches
4ff4b66 Allow selection of font for symbols as in Emacs 24.x
c03d44b ; Fix last commit
d4be4f3 ; Fix indexing in lispref manual
ed399f2 ; Minor improvement in documentation of generators
197a6bc Fix horizontal scrolling during Isearch
3566644 Fix infloop in redisplay due to truncated lines and invisible...
# Conflicts:
# etc/NEWS
* src/font.c (syms_of_font) <inhibit-compacting-font-caches>: New
boolean variable.
* src/alloc.c (compact_font_caches): Use it to bypass the call to
compact_font_cache_entry. (Bug#24634) (Bug#24565)
* etc/NEWS: Mention the new variable.
This follows up on recent problems with the fact that config.h
includes stdlib.h etc.; some files need to include stdlib.h later.
config.h generally should limit itself to includes that are
universally safe; outside of MS-Windows, only stdbool.h makes
the cut among the files currently included. So, move the
other includes to just the files that need them (Bug#24506).
* configure.ac (config_opsysfile): Remove, as this generic hook
is no longer needed.
* lib-src/etags.c, src/unexmacosx.c, src/w32.c, src/w32notify.c:
* src/w32proc.c (_GNU_SOURCE):
Remove, as it’s OK for config.h to do this now.
* src/conf_post.h: Include <ms-w32.h>, instead of the generic
config_opsysfile, for simplicity as this old way of configuring is
now done only for the MS-Windows port. Do not include <ms-w32.h>
if DEFER_MS_W32_H, for the benefit of the few files that want its
effects later. Do not include <alloca.h>, <string.h>, or
<stdlib.h>. Other files modified to include these headers as
needed, or to not include headers that are no longer needed.
* src/lisp.h: Include <alloca.h> and <string.h> here, since
some of the inline functions need them.
* src/regex.c: Include <alloca.h> if not emacs. (If emacs,
we can rely on SAFE_ALLOCA.) There is no longer any need to
worry about HAVE_ALLOCA_H.
* src/unexmacosx.c: Rely on config.h not including stdlib.h.
* src/w32.c, src/w32notify.c, src/w32proc.c (DEFER_MS_W32_H):
Define before including <config.h> first, and include <ms-w32.h>
after the troublesome headers.
This removes a workaround where Emacs did not use flexible
array members when compiled with IBM XL C. Instead, avoid
the problem by making the aliasing issues more obvious to
this compiler.
* admin/merge-gnulib: Don’t remove m4/flexmember.m4.
* m4/flexmember.m4: Copy from gnulib.
* configure.ac (AC_C_FLEXIBLE_ARRAY_MEMBER): Remove workaround.
* src/alloc.c (allocate_string_data): Rephrase to avoid aliasing
problem that would otherwise mess up code generated for flexible
array members by IBM XL C for AIX, V12.1.
* src/conf_post.h (FLEXIBLE_ARRAY_MEMBER): Remove; now done
by gnulib code.
* src/lisp.h:
* src/print.c (Fprin1_to_string):
* src/eval.c (signal_or_quit):
* src/alloc.c (garbage_collect_1): Remove declarations, setting,
and testing the value of abort_on_gc. It is no longer needed, and
using it causes rare aborts in GC for no good reason. (Bug#23912)
These changes are needed to conform to the C standard's rule for
allocating structs containing flexible array members. C11 says
that malloc (offsetof (struct s, m) + n) does not suffice to
allocate a struct with an n-byte tail; instead, malloc’s arg
should be rounded up to the nearest multiple of alignof (struct s).
Although this is arguably a defect in C11, gcc -O2 + valgrind
sometimes complains when this rule is violated, and when debugging
it’s better to keep valgrind happy.
For details please see the thread containing the message at:
https://gcc.gnu.org/ml/gcc-patches/2016-09/msg00416.html
* lib-src/ebrowse.c, src/alloc.c, src/image.c, src/process.c:
Include flexmember.h.
* lib-src/ebrowse.c (add_sym, add_member, make_namespace)
(register_namespace_alias):
* src/alloc.c (SDATA_SIZE, allocate_string_data):
* src/image.c (xpm_cache_color, imagemagick_create_cache):
* src/process.c (Fmake_network_process):
Use FLEXSIZEOF instead of offsetof and addition.
* src/alloc.c (SDATA_SIZE, vector_alignment):
Use FLEXALIGNOF instead of sizeof (ptrdiff_t).
* src/lisp.h (ALIGNOF_STRUCT_LISP_VECTOR):
Remove, as alloc.c can now calculate this on its own.
These changes remove some assumptions about heap allocator
alignment that may not be true on unusual platforms.
* src/alloc.c (POWER_OF_2): New macro.
(ROUNDUP): Use it.
(BLOCK_ALIGN): Verify that it is a power of 2.
(aligned_alloc): Check that alignment passed to posix_memalign
satisfies POSIX restrictions.
(lisp_align_malloc): Check that size passed to aligned_alloc
satisfies C11 restrictions.
(MALLOC_IS_GC_ALIGNED): Check that GCALIGNMENT is 8, since the
code has not been verified to work with other GCALIGNMENT values
and the ice is thin here. On GNU/Linux, malloc can return a value
that is a multiple of 8 but not 16, even though __alignof__
(max_align_t) is 16. See:
https://gcc.gnu.org/ml/gcc-patches/2016-08/msg01902.html
(lmalloc) [USE_ALIGNED_ALLOC]: Use aligned_alloc only if size
is a multiple of alignment, since C11 says the behavior is
undefined otherwise.
(lmalloc, lrealloc): Don't use INT_ADD_WRAPV on size_t, as in
general this macro is restricted to signed types. Remove
assertion that the result is a multiple of GCALIGNMENT, as that
need not be true.
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/dispextern.h (FACE_FROM_ID_OR_NULL): Renamed from
FACE_OPT_FROM_ID; all callers changed.
* src/xdisp.c (extend_face_to_end_of_line): Call FACE_FROM_ID, not
FACE_FROM_ID_OR_NULL, as the resulting face is immediately
dereferenced.
(fill_gstring_glyph_string): Call FACE_FROM_ID, not
FACE_FROM_ID_OR_NULL, as the resulting face will be dereferenced
when the glyph string is drawn.
(BUILD_COMPOSITE_GLYPH_STRING): Call FACE_FROM_ID, not
FACE_FROM_ID_OR_NULL, as the resulting face will be dereferenced
in fill_composite_glyph_string.
(calc_line_height_property): Call FACE_FROM_ID_OR_NULL rather that
FACE_FROM_ID, since the function and its caller can cope with that
situation. Conflate 3 tests of missing face or font into just
one.
* src/xfaces.c (Fx_list_fonts, Fface_font, lookup_face): Call
FACE_FROM_ID_OR_NULL rather that FACE_FROM_ID, since these
functions can cope with that situation.
(lookup_derived_face): Don't call FACE_FROM_ID if the result will
not be used.
* src/w32console.c (w32_face_attributes): Remove redundant 'eassert'.
