Enable GC_REMEMBER_LAST_MARKED by default (it was disabled in Emacs 29)
to make it easier to debug difficult-to-reproduce GC problems
encountered by users. This increases GC costs by about 5 %, which can
be avoided by turning the mark trace buffer back off using the new
--disable-gc-mark-trace option.
See discussion at
https://lists.gnu.org/archive/html/emacs-devel/2024-09/msg00240.html
* configure.ac (--disable-gc-mark-trace): New config option.
* etc/NEWS: Mention it.
* src/alloc.c: Enable it by default and avoid a compiler warning.
This fixes the problem whereby attaching GDB to a running Emacs
on MS-Windows would slow down keyboard input, because the
low-level keyboard hook installed by Emacs at startup was still
installed, but with Emacs stopped, the hook code couldn't run,
and therefore the OS would time-out waiting for the hook to
return. Now when GDB is attached to Emacs, it will remove the
hook right away.
* src/.gdbinit: Call 'remove_w32_kbdhook' if the keyboard hook is
already installed.
* src/alloc.c (defined_WINDOWSNT): New enum.
(gdb_make_enums_visible): Add 'defined_WINDOWSNT'.
The purpose of these changes is to improve the code handling the
display of buffers in windows, switching to previous and next
buffers in windows and restoring a previous state after quitting
or killing buffers. In particular it does:
- Add a new window parameter 'quit-restore-prev' so a window can
keep its initial 'quit-restore' parameter and undoing a sequence
of quit window operations becomes more reliable (Bug#59862).
- Optionally have 'kill-buffer' call 'quit-restore-window' for
all windows showing the argument buffer (Bug#59862).
- Add a new hook so it's possible to avoid that a window gets
deleted implicitly by functions like 'kill-buffer' (Bug#71386).
- Add a new option to make 'quit-restore-window' delete windows
more aggressively (Bug#59862).
- Immediately remove killed buffers from all windows' previous
and next buffers. For windows that are already dead, use a weak
hash table to be used by 'kill-buffer'. This avoids any special
handling of such windows by the garbage collector.
- Immediately remove 'quit-restore' and 'quit-restore-prev'
window parameters that reference killed buffers. These
parameters have no more use once their buffers got killed.
- Make sure that internal windows do not have any previous and
next buffers. This fixes a silly memory leak.
- Make sure that after set_window_buffer and some wset_buffer
calls the buffer now shown in the window does not appear in the
lists of that window's previous and next buffers. The old
behavior could make functions investigating these lists
erroneously believe that there still existed some other buffer
to switch to.
* src/alloc.c (mark_discard_killed_buffers): Remove function.
(mark_window): No more filter previous and next buffer lists.
* src/window.h (struct window): Move up prev_buffers and
next-buffers in structure; they are now treated by the collector
as usual.
* src/window.c (window_discard_buffer_from_alist)
(window_discard_buffer_from_list)
(window_discard_buffer_from_window)
(window_discard_buffer_from_dead_windows)
(Fwindow_discard_buffer): New functions.
(set_window_buffer): Discard BUFFER from WINDOW's previous and
next buffers.
(make_parent_window): Make sure internal windows have no previous
and next buffers.
(make_window): Don't initialize window's previous and next
buffers, they are handled by allocate_window now.
(Fdelete_window_internal): Add WINDOW to window_dead_windows_table.
(Fset_window_configuration): Remove resurrected window from
window_dead_windows_table. Make sure buffers set by wset_buffer
calls are not recorded in window's previous and next buffers.
(delete_all_child_windows): Add deleted windows to
window_dead_windows_table.
(window_dead_windows_table): New weak hash table to record dead
windows that are stored in saved window configurations.
* src/buffer.c (Fkill_buffer): Call new function
'window_discard_buffer_from_dead_windows'.
* lisp/window.el (window-deletable-functions): New hook.
(window-deletable-p): Update doc-string. Run
'window-deletable-functions' (Bug#71386).
(unrecord-window-buffer): New argument ALL. Move body to
'window-discard-buffer-from-window' so that if ALL is non-nil,
WINDOW's 'quit-restore' and 'quit-restore-prev' parameters get
removed too.
(switch-to-prev-buffer): Don't care about killed buffers here;
'replace-buffer-in-windows' should have done that already. Use
'unrecord-window-buffer'.
(switch-to-next-buffer): Don't care about killed buffers here;
'replace-buffer-in-windows' should do that now.
(kill-buffer-quit-windows): New option.
(delete-windows-on): Update doc-string. Handle new option
'kill-buffer-quit-windows'. Update 'unrecord-window-buffer'
calls.
(replace-buffer-in-windows): Update doc-string. Handle new
option 'kill-buffer-quit-windows' (Bug#59862). Update call to
'unrecord-window-buffer'.
(quit-restore-window-no-switch): New option.
(quit-restore-window): Update doc-string. Handle additional
values of BURY-OR-KILL so to not kill a buffer about to be
killed by the caller. Handle 'quit-restore-prev' parameter
(Bug#59862). Handle new option 'quit-restore-window-no-switch'
(Bug#59862).
(quit-windows-on): Update doc-string. Call 'quit-window-hook'
and call 'quit-restore-window' directly so that the buffer does
not get buried or killed by the latter. Update
'unrecord-window-buffer' call.
(display-buffer-record-window): Update doc-string. Handle new
`quit-restore-prev' parameter (Bug#59862).
(switch-to-buffer): Call 'display-buffer-record-window' so a
latter 'quit-restore-window' can use its parameters.
* doc/lispref/windows.texi (Deleting Windows): Describe implicit
deletion of windows and new hook 'window-deletable-functions'.
(Buffers and Windows): Update description of
'replace-buffer-in-windows'. Describe new option
'kill-buffer-quit-windows'.
(Quitting Windows): Describe 'quit-restore-prev' parameter and
new option 'quit-restore-window-no-switch'. Update description
of 'quit-restore-window'.
(Window Parameters): Mention 'quit-restore-prev' parameter.
* etc/NEWS: Add entries for 'window-deletable-functions',
'kill-buffer-quit-windows', 'quit-restore-window-no-switch'.
mention new parameter 'quit-restore-prev' and new argument
values for 'quit-restore-window'.
* src/alloc.c (make_clear_bool_vector): It’s now the caller’s
responsibility to make sure the bool vector length is in range.
Add an eassert to double-check this. This lets some locals be
ptrdiff_t not EMACS_INT.
(Fmake_bool_vector, Fbool_vector):
Check that bool vector lengths are in range.
* src/lisp.h (BOOL_VECTOR_LENGTH_MAX): New macro.
(bool_vector_words, bool_vector_bytes): Avoid undefined
behavior if size == EMACS_INT_MAX - (BITS_PER_BITS_WORD - 1).
This is mostly theoretical but it’s easy to do it right.
* src/lread.c (read_bool_vector): Use EMACS_INT, not just ptrdiff_t.
Check that length doesn’t exceed BOOL_VECTOR_LENGTH_MAX.
This fixes an unlikely integer overflow where the calculated size
went negative.
Although loading uninitialized works from memory and then ignoring
the result works fine on conventional architectures, it
technically has undefined behavior in C, so redo bool_vector
allocation so that the code never does that. This can improve
performance when allocating large vectors of nil, since calloc can
clear the memory lazily.
* src/alloc.c (make_clear_bool_vector): New function,
a generalization of make_uninit_bool_vector.
(make_uninit_bool_vector): Use it.
(Fmake_bool_vector): If !INIT, rely on make_clear_bool_vector.
* src/alloc.c (Fbool_vector):
* src/fns.c (Freverse): Don’t access uninitialized bool_vector words.
* src/alloc.c (mark_frame): Mark this frame's image cache, if it
exist.
(mark_terminals): Cease marking T->image_cache.
* src/androidfns.c (unwind_create_frame, Fx_create_frame)
(android_create_tip_frame):
* src/haikufns.c (unwind_create_frame, haiku_create_frame)
(haiku_create_tip_frame):
* src/nsfns.m (unwind_create_frame):
* src/pgtkfns.c (unwind_create_frame, Fx_create_frame)
(x_create_tip_frame):
* src/xfns.c (unwind_create_frame, Fx_create_frame)
(x_create_tip_frame):
* src/w32fns.c (unwind_create_frame, Fx_create_frame)
(w32_create_tip_frame): Remove adjustments of the frame image
cache's reference count rendered redundant by the assignment of
image caches to individual frames rather than terminals.
* src/dispextern.h (struct image_cache) <scaling_col_width>: New
field.
* src/frame.c (gui_set_font): In lieu of clearing F's image
cache unconditionally, establish whether the column width as
considered by compute_image_size has changed, and if so, adjust
or reassign the frame's image cache.
(make_frame): Clear F->image_cache.
* src/frame.h (struct frame) <image_cache>: New field.
(FRAME_IMAGE_CACHE): Return F->image_cache.
* src/image.c (make_image_cache): Clear C->scaling_col_width.
(cache_image): Adjust to new means of assigning image caches to
frames.
* src/termhooks.h (struct terminal) <image_cache>: Delete field.
* src/xfaces.c (init_frame_faces): Do image cache assignment
with all new frames.
This is a new warning diagnostic in GCC 14.
* lib-src/etags.c (mercury_heuristics_ratio):
* src/pgtkselect.c, src/xselect.c (selection_request_stack):
* src/xselect.c (outstanding_transfers):
* src/xterm.c (pending_selection_requests)
(x_dnd_waiting_for_motif_finish_display):
Now static.
* lib-src/make-docfile.c (close_emacs_globals):
Arrange for lispsym to be declared with extern first,
when compiling lread.c.
* src/alloc.c (gdb_make_enums_visible) [__GNUC__]:
* src/emacs.c (RCS_Id):
* src/keyboard.c (stop_character):
* src/print.c (print_output_debug_flag):
Now declared with extern first.
* src/lisp.h (DEFINE_GDB_SYMBOL_BEGIN) [MAIN_PROGRAM]:
Arrange for ID to be declared extern first.
* src/lisp.h (garbage_collection_inhibited):
* src/xterm.h (x_frame_parm_handlers):
Declare here, so that its interface is properly checked. Other decls
removed.
* lisp/cus-edit.el (Custom-mode): Enable text conversion, now
that fields are correctly treated.
* src/alloc.c (mark_frame): Mark f->conversion.field.
* src/androidterm.c (android_update_selection): Adjust
conversion region and selection position by the field start and
end.
* src/editfns.c (find_field): Export function.
* src/frame.c (make_frame): Clear f->conversion.field.
* src/frame.h (struct text_conversion_state) <field>: New field.
* src/lisp.h (find_fields, reset_frame_conversion): Export
functions.
* src/minibuf.c (Fread_from_minibuffer): Reset frame conversion
if Voverriding_text_conversion_style is set.
* src/textconv.c (textconv_query): Narrow to field.
(reset_frame_conversion): New function.
(reset_frame_state): Clear conversion field.
(really_delete_surrounding_text): Narrow to field.
(locate_and_save_position_in_field): New function.
(really_request_point_update, really_set_point_and_mark)
(complete_edit_check, handle_pending_conversion_events_1)
(handle_pending_conversion_events, get_conversion_field)
(set_composing_region, textconv_set_point_and_mark, replace_text)
(get_extracted_text, get_surrounding_text, report_point_change):
Compute, narrow to and offset by the currently active field
whenever point is updated or a command is received.
(syms_of_textconv): Revise doc strings.
* src/textconv.h (get_conversion_field): Export function.
In preparation for the use of `PVEC_COMPILED` objects for
interpreted functions, rename them to use a more neutral name.
* src/lisp.h (enum pvec_type): Rename `PVEC_COMPILED` to `PVEC_CLOSURE`.
(enum Lisp_Compiled): Use `CLOSURE_` prefix i.s.o `COMPILED_`.
Also use `CODE` rather than `BYTECODE`.
(CLOSUREP): Rename from `COMPILEDP`.
(enum Lisp_Closure): Rename from `Lisp_Compiled`.
* src/alloc.c, src/bytecode.c, src/comp.c, src/data.c, src/eval.c,
* src/fns.c, src/lisp.h, src/lread.c, src/pdumper.c, src/print.c,
* src/profiler.c: Rename all uses accordingly.
* src/.gdbinit (xclosure): Rename from `xcompiled`.
(xcompiled): New obsolete alias.
(xpr): Adjust accordingly. Also adjust to new PVEC_CLOSURE tag name.
This reverts commit de6b1e1efb.
While it did simplify code, there aren't much in the way of technical
benefits the change at this time, and there were protest against the
unwarranted style change.
The new opaque type replaces the previous use of vectors for obarrays.
`obarray-make` now returns objects of this type. Functions that take
obarrays continue to accept vectors for compatibility, now just using
their first slot to store an actual obarray object.
obarray-size and obarray-default-size now obsolete.
* lisp/obarray.el (obarray-default-size, obarray-size):
Declare obsolete.
(obarray-make, obarrayp, obarray-clear): Remove from here.
* src/fns.c (reduce_emacs_uint_to_hash_hash): Remove from here.
* src/lisp.h (struct Lisp_Obarray, OBARRAYP, XOBARRAY, CHECK_OBARRAY)
(make_lisp_obarray, obarray_size, check_obarray)
(obarray_iter_t, make_obarray_iter, obarray_iter_at_end)
(obarray_iter_step, obarray_iter_symbol, DOOBARRAY, knuth_hash): New.
(reduce_emacs_uint_to_hash_hash): Moved here.
* src/lread.c (check_obarray): Renamed and reworked as...
(checked_obarray_slow): ...this.
(intern_sym, Funintern, oblookup, map_obarray)
(Finternal__obarray_buckets): Adapt to new type.
(obarray_index, allocate_obarray, make_obarray, grow_obarray)
(obarray_default_bits, Fobarray_make, Fobarrayp, Fobarray_clear): New.
* etc/emacs_lldb.py (Lisp_Object):
* lisp/emacs-lisp/cl-macs.el (`(,type . ,pred)):
* lisp/emacs-lisp/cl-preloaded.el (cl--typeof-types):
* lisp/emacs-lisp/comp-common.el (comp-known-type-specifiers):
* lisp/emacs-lisp/comp.el (comp-known-predicates):
* src/alloc.c (cleanup_vector, process_mark_stack):
* src/data.c (Ftype_of, syms_of_data):
* src/minibuf.c (Ftry_completion, Fall_completions, Ftest_completion):
* src/pdumper.c (dump_obarray_buckets, dump_obarray, dump_vectorlike):
* src/print.c (print_vectorlike_unreadable):
* test/lisp/abbrev-tests.el (abbrev-make-abbrev-table-test):
* test/lisp/obarray-tests.el (obarrayp-test)
(obarrayp-unchecked-content-test, obarray-make-default-test)
(obarray-make-with-size-test):
Adapt to new type.
This makes both lookups and rehashing cheaper. The index vector size
is now always a power of 2. The first table size is reduced to
6 (from 8), because index vectors would become excessively big
otherwise.
* src/lisp.h (struct Lisp_Hash_Table): Replace index_size with
index_bits. All references adapted.
(hash_table_index_size): New accessor; use it where applicable.
* src/fns.c (hash_index_size): Replace with...
(compute_hash_index_bits): ...this new function, returning the log2 of the
index size. All callers adapted.
(hash_index_index): Knuth multiplicative hashing instead of remainder.
(maybe_resize_hash_table): Reduce first table size from 8 to 6.
* src/lisp.h (TAG_PTR_INITIALLY): Rename from TAG_PTR, since calls
can be used only as initializers, and the convention elsewhere in
lisp.c is to give these macros names ending in ‘_INITIALLY’.
This should help avoid confusion such as we recently experienced
in make_lisp_symbol_internal. All uses changed.
`struct composition` kept an index into the internal `key_and_value` array
of hash tables, which only worked because of details of how
hash-tables are handled. Replace it with a reference to the
key stored at that location in the hash-table, which saves us an
indirection while at it.
* src/composite.h (struct composition): Replace `hash_index` with
the actual `key`.
(COMPOSITION_KEY): Simplify accordingly.
(mark_composite): Declare.
* src/composite.c (get_composition_id): Adjust accordingly.
(mark_composite): New function.
* src/charset.c (mark_charset): Uncomment.
* src/lisp.h (mark_charset): Declare.
* src/alloc.c (garbage_collect): Call `mark_charset` and `mark_composite`.
* src/pdumper.c (hash_table_contents): Remove invalid comment, since
compositions aren't dumped.
This removes hacks from code that had to be careful not to use
Qunbound as a hash table key, at the cost of a minor hack in
the GC marker.
* src/lisp.h (INVALID_LISP_VALUE, HASH_UNUSED_ENTRY_KEY):
Define as a null-pointer float.
* src/alloc.c (process_mark_stack): Add hack to ignore that value.
* src/pdumper.c (dump_object_needs_dumping_p)
(pdumper_init_symbol_unbound, pdumper_load):
* src/print.c (PRINT_CIRCLE_CANDIDATE_P): Remove hacks for Qunbound.
Be more systematic about putting space before paren in calls,
and in avoiding unnecessary parentheses in macros.
This was partly inspired by my wading through gcc -E output
while debugging something else, and seeing too many parens.
This patch does not change the generated .o files on my platform.
This saves several words in the hash table object at the cost of an
indirection at runtime. This seems to be a gain in overall
performance.
FIXME: We cache hash test objects in a rather clumsy way. A better
solution is sought.
* src/lisp.h (struct Lisp_Hash_Table): Use a pointer to the test
struct. All references adapted.
* src/alloc.c (garbage_collect):
* src/fns.c (struct hash_table_user_test, hash_table_user_tests)
(mark_fns, get_hash_table_user_test): New state for caching test
structs, and functions managing it.
Using xmalloc for allocating these arrays is much cheaper than using
Lisp vectors since they are no longer marked or swept by the GC, and
deallocated much sooner. This makes GC faster and less frequent, and
improves temporal locality.
Zero-sized tables use NULL for their (0-length) vectors except the
index vector which has size 1 and uses a shared constant static vector
since it cannot be modified anyway. This makes creation and
destruction of zero-sized hash tables very fast; they consume no
memory outside the base object.
* src/lisp.h (struct Lisp_Hash_Table): Retype the index, next, hash
and key_and_value vectors from Lisp_Object to appropriately typed
arrays (although hash values are still stored as Lisp fixnums). Add
explicit table_size and index_size members. All users updated.
* src/alloc.c (gcstat): Add total_hash_table_bytes.
(hash_table_allocated_bytes): New.
(cleanup_vector): Free hash table vectors when sweeping
the object.
(hash_table_alloc_bytes, hash_table_free_bytes): New.
(sweep_vectors): Update gcstat.total_hash_table_bytes.
(total_bytes_of_live_objects): Use it.
(purecopy_hash_table): Adapt allocation of hash table vectors.
(process_mark_stack): No more Lisp slots in the struct to trace.
* src/fns.c (empty_hash_index_vector): New.
(allocate_hash_table): Allocate without automatically GCed slots.
(alloc_larger_vector): Remove.
(make_hash_table, copy_hash_table, maybe_resize_hash_table):
Adapt vector allocation and initialisation.
* src/pdumper.c (hash_table_freeze, hash_table_thaw, dump_hash_table)
(dump_hash_table_contents):
Adapt dumping and loading to field changes.
This avoids any extra allocation for such vectors, including empty
tables read by the Lisp reader, and provides extra safety essentially
for free.
* src/fns.c (make_hash_table): Allow tables to be 0-sized. The index
will always have at least one entry, to avoid extra look-up costs.
* src/alloc.c (process_mark_stack): Don't mark pure objects,
because empty vectors are pure.
This takes less space (saves an entire word) and is more type-safe.
No change in behaviour.
* src/lisp.h (hash_table_weakness_t): New.
(struct Lisp_Hash_Table): Replace Lisp object `weak` with enum
`weakness`.
* src/fns.c
(keep_entry_p, hash_table_weakness_symbol): New.
(make_hash_table): Retype argument. All callers updated.
(sweep_weak_table, Fmake_hash_table, Fhash_table_weakness):
* src/alloc.c (purecopy_hash_table, purecopy, process_mark_stack):
* src/pdumper.c (dump_hash_table):
* src/print.c (print_object): Use retyped field.
The profiler stored data being collected in Lisp hash tables but
relied heavily on their exact internal representation, which made it
difficult and error-prone to change the hash table implementation.
In particular, the profiler has special run-time requirements that are
not easily met using standard Lisp data structures: accesses and
updates are made from async signal handlers in almost any messy
context you can think of and are therefore very constrained in what
they can do.
The new profiler tables are designed specifically for their purpose
and are more efficient and, by not being coupled to Lisp hash tables,
easier to keep safe.
The old profiler morphed internal hash tables to ones usable from Lisp
and thereby made them impossible to use internally; now export_log
just makes new hash table objects for Lisp. The Lisp part of the
profiler remains entirely unchanged.
* src/alloc.c (garbage_collect): Mark profiler tables.
* src/eval.c (get_backtrace): Fill an array of Lisp values instead of
a Lisp vector.
* src/profiler.c (log_t): No longer a Lisp hash table but a custom
data structure: a fully associative fixed-sized cache that maps
fixed-size arrays of Lisp objects to counts.
(make_log): Build new struct.
(mark_log, free_log, get_log_count, set_log_count, get_key_vector)
(log_hash_index, remove_log_entry, trace_equal, trace_hash)
(make_profiler_log, free_profiler_log, mark_profiler): New.
(cmpfn_profiler, hashtest_profiler, hashfn_profiler)
(syms_of_profiler_for_pdumper): Remove.
(approximate_median, evict_lower_half, record_backtrace, export_log)
(Fprofiler_cpu_log, Fprofiler_memory_log, syms_of_profiler):
Adapt to the new data structure.
This code was introduced in 2014 to catch a GC bug that, according to
Paul Eggert in 2019, "seems to have been fixed" (see 2b552f3489
2019-08-21 "Don’t debug fset by default"). It has been marked
obsolete since that time, and no one has mentioned it on our mailing
lists since. Let's just get rid of it.
* src/alloc.c
(SUSPICIOUS_OBJECT_CHECKING) [ENABLE_CHECKING]: Don't define.
(suspicious_free_record, suspicious_objects, suspicious_object_index)
(suspicious_free_history, suspicious_free_history_index)
(note_suspicious_free) [SUSPICIOUS_OBJECT_CHECKING]: Delete.
(find_suspicious_object_in_range)
(detect_suspicious_free): Delete functions.
(cleanup_vector)
(allocate_vectorlike): Don't call above deleted functions.
(Fsuspicious_object): Delete DEFUN.
(syms_of_alloc) <Ssuspicious_object>: Delete defsubr.
Instead of scanning vector_free_lists from the appropriate size until
we find a nonempty bucket, start at the last bucket where we last put
something in. This may favour splitting larger vectors than necessary
but in general saves a lot of time in the allocation of small vectors.
Original patch by Ihor Radchenko.
* src/alloc.c (last_inserted_vector_free_idx): New variable.
(setup_on_free_list): Set it.
(allocate_vector_from_block): Use it.
(sweep_vectors): Reset it.
The latter half of vector_free_lists was never used in any meaningful
way but it did require traversal during allocation and GC. Reduce it
to sizes we actually allocate, with a bucket for bigger ones.
* src/alloc.c (VECTOR_MAX_FREE_LIST_INDEX): Rename to...
(VECTOR_FREE_LIST_ARRAY_SIZE): ... this and adjust its value.
(vector_free_lists): Use new, smaller size.
(setup_on_free_list, allocate_vector_from_block):
Adapt to new vector_free_lists size.
(pseudovector_nbytes): New function extracted from...
(vectorlike_nbytes): ...here.
* src/alloc.c: The size of a vector block is bound by the number of
words, not bytes, represented by the pseudovector header RESTSIZE
field, because that limits how big a PVEC_FREE object can be.
* src/alloc.c (NEAR_STACK_TOP): Hoist from here ...
* src/thread.h: ... to here.
* src/print.c (print_object): Use NEAR_STACK_TOP instead of raw
buffer address. This is more natural, and pacifies GCC 13.
