* msdos/sedlibmk.inp (GL_GNULIB_STDC_[^ =] &c.): Rename from
GL_STDC_... and define to 1 before other variables prefixed
GL_GNULIB_ are defined to 0.
(NEXT_AS_FIRST_DIRECTIVE_STDBIT_H)
(NEXT_AS_FIRST_DIRECTIVE_STDCOUNTOF_H, STDCOUNTOF_H): Define.
(GL_HAVE_STDBIT_H_CONDITION): Replace
GL_GENERATE_STDBIT_H_CONDITION.
(CXX_HAVE_STDCOUNTOF_H, GL_GNULIB_STREQ, GL_GNULIB_MEMEQ)
(GL_GNULIB_STRNUL, GL_GENERATE_STDCOUNTOF_H_CONDITION):
Substitute "1".
* src/conf_post.h (_REGEX_AVOID_UCHAR_H) [MSDOS]: Define to 1.
* src/term.c (init_tty) [MSDOS]: Set TN_max_colors to 16,
which was deleted inadvertently with recent MS Windows
modifications.
* config.bat: Rename stdcountof.in.h appropriately.
* src/lisp.h (lisp_h_make_fixnum_wrap) [USE_LSB_TAG]: Use compound
literal, not cast. This pacifies -Wuseless-cast when building on
Fedora 44 x86-64 with ‘./configure --enable-gcc-warnings
CFLAGS='-O2 -DDEFINE_KEY_OPS_AS_MACROS'’.
Avoid useless casts on GCC 14+ platforms on non MS-Windows, by
using a cast only if HAVE_NTGUI. That cast should not provoke
-Wuseless-cast regardless of how MS-Windows defines HDC, since it
casts to an Emacs-defined type. This change lets the code compile
again on Fedora 44 with --enable-gcc-warnings.
* src/dispextern.h (PIX_CONTAINER_TO_CONTEXT): New macro.
It defaults to the identity macro.
(Emacs_Pix_Context): Default to Emacs_Pix_Container.
* src/image.c (pbm_load, png_load_body, jpeg_load_body)
(tiff_load, gif_load, svg_load_image): Use the new macro
instead of a cast.
* src/conf_post.h: Add leading underscores to 'gl_consolesafe_*'
functions, to follow Gnulib's renaming.
* src/image.c (pbm_load, png_load_body, jpeg_load_body)
(tiff_load): Revert previous change that removed "useless" casts,
as removing them triggers warnings from MinGW GCC 9.2.
C2y will standardize countof as the macro that Emacs uses the name
ARRAYELTS for. Switch to the standard name, which is supported
by GCC 16+, by Clang 21, and by the Gnulib stdcountof-h module already
in use for compilers that do not support countof.
Also, use countof in a few places where we missed using ARRAYELTS.
* admin/coccinelle/arrayelts.cocci: Suggest countof, not ARRAYELTS.
* admin/merge-gnulib (GNULIB_MODULES): Add stdcountof-h,
as it is now a direct rather than an indirect dependency.
* exec/trace.c, src/lisp.h, src/sfnt.c: Include <stdcountof.h>.
(ARRAYELTS): Remove. All uses replaced by countof.
* lib-src/ebrowse.c, lib-src/etags.c, lib-src/make-docfile.c:
* lib-src/seccomp-filter.c, lwlib/lwlib-Xaw.c:
Prefer <stdcountof.h> and countof to doing things by hand.
In addition to the automatic changes, also do the following,
needed due to recent Gnulib changes.
* admin/merge-gnulib (AVOIDED_MODULES): Add btoc32,
c32_apply_type_test, c32_get_type_test, c32isalnum, c32rtomb,
c32tolower, c32toupper, localeinfo, mbrtoc32-regular.
Remove btowc, iswctype, mbrtowc, wcrtomb, wctype, wctype-h.
Also remove iswblank, iswdigit, iswxdigit, locale-h, raise,
stdarg-h, some of which perhaps could have been removed earlier.
* configure.ac (_REGEX_AVOID_UCHAR_H): New macro.
GCC 16’s -Wuseless-cast warning can be useful.
Fix the useless casts it identifies, and also fix false positives
by using compound literals, which are safer anyway than casts.
* src/composite.c (composition_adjust_point)
(Ffind_composition_internal):
* lwlib/xlwmenu.c (xlwMenuResources, xlwMenuClassRec)
(resource_widget_value, XlwMenuDestroy, Select):
* src/alloc.c (process_mark_stack):
* src/data.c (Faref):
* src/emacs-module.c (module_extract_big_integer):
* src/fileio.c (Finsert_file_contents):
* src/frame.h (FRAME_MESSAGE_BUF_SIZE):
* src/gtkutil.c (xg_tool_item_stale_p, update_frame_tool_bar):
* src/image.c (pbm_load, png_load_body, jpeg_load_body)
(tiff_load, gif_load):
* src/pdumper.c (ptrdiff_t_to_dump_off, dump_queue_dequeue)
(field_relpos, dump_field_emacs_ptr)
(dump_object_start_pseudovector, pdumper_remember_scalar_impl)
(pdumper_load, syms_of_pdumper):
* src/regex-emacs.c (BUF_PUSH, BUF_PUSH_2, POINTER_TO_OFFSET):
* src/xdisp.c (remember_mouse_glyph, pint2str):
* src/xterm.c (cvt_string_to_pixel, handle_one_xevent):
Omit useless casts. Perhaps they were formerly needed,
but they should not be needed now.
* src/alloc.c (Fmemory_info):
* src/category.c (Fdefine_category, Fmodify_category_entry):
* src/data.c (Fash):
* src/dispextern.h (GLYPH_CODE_P):
* src/emacs.c (load_seccomp):
* src/fns.c (Flocale_info, maybe_resize_hash_table):
* src/indent.c (check_display_width):
* src/json.c (symset_size):
* src/lisp.h (XUNTAG, BOOL_VECTOR_LENGTH_MAX, obarray_size)
(hash_table_index_size):
* src/lread.c (make_obarray, grow_obarray, Fobarray_clear):
* src/menu.c (digest_single_submenu, x_popup_menu_1):
* src/term.c (init_tty):
* src/widget.c (update_wm_hints):
* src/xdisp.c (truncate_echo_area):
* src/xfns.c (x_set_border_pixel):
* src/xfont.c (xfont_match, xfont_open):
* src/xmenu.c (set_frame_menubar):
* test/src/emacs-module-resources/mod-test.c (emacs_module_init):
Use compound literal instead of a cast that is useless in
some platforms but not others.
* src/dispextern.h, src/haikugui.h, src/w32gui.h:
(WINDOW_HANDLE_UINTPTR): New macro.
* src/frame.c (gui_report_frame_params):
* src/xterm.c (x_try_cr_xlib_drawable):
Use it.
* src/lisp.h (XUNTAG): And tag with UINTPTR_MAX to pacify
gcc warning about a constant out of range.
(hash_idx_t): Make it int_least32_t, as it need not be exactly 32 bits.
(PRIdHASH_IDX): New macro.
* src/pdumper.c (dump_queue_dequeue): Use it.
* src/profiler.c (setup_cpu_timer): Make a local EMACS_INT
rather than int, to avoid need for casting later.
* src/syntax.c (uninitialized_interval):
Use 1u rather than 1 so the cast is always useful.
A compound literal wouldn’t do here, as this macro
needs to be an integer constant expression.
* src/xfns.c (XICCallback, XICProc): Remove macros.
(Xxic_preedit_start_callback): Use a cleaner way to specify it,
avoiding the need for type macros, and for a cast
if HAVE_XICCALLBACK_CALLBACK.
* src/xterm.c (handle_one_xevent): 2nd arg is now XEvent *
on all platforms, as there is no need to diverge, and
diverging meant we needed lots of unnecessary casts.
`visual-wrap--content-prefix' previously returned a column count
computed as
(max (string-width prefix)
(ceiling (string-pixel-width prefix) avg-space-width))
with two problems:
* `string-width' ignores `buffer-invisibility-spec', so an invisible
prefix (hidden ATX markers under `markdown-ts-hide-markup', for
example) still reserved its character count on line 1 via a
`min-width' display property, shifting the visible heading right.
* With variable-pitch fonts, rounding the prefix width up to whole
columns added visible padding whenever the natural width did not
fall on an exact column boundary.
Return the prefix's natural pixel width via `string-pixel-width'
instead, which accounts for any display transformation applied to
the prefix (invisibility, `display' replacements, text scaling,
proportional fonts). Drop the `min-width' property from
`visual-wrap--apply-to-line' so line 1 renders at its natural width.
Switch the continuation `wrap-prefix' to a mixed-unit `:align-to'
sum form:
(space :align-to (+ (PIX) (EXTRA-INDENT . width)))
where PIX is the prefix's pixel width and EXTRA-INDENT is
`visual-wrap-extra-indent' in canonical character widths. The
display engine resolves each term per the active frame and sums
them, so no Lisp-level unit conversion is needed.
Since `min-width' is no longer installed, the accumulation cycle
that commit 81a5beb8af (bug#73882) worked around cannot recur.
Drop the `min-width' strip from `visual-wrap--content-prefix' and
the `min-width' removal from `visual-wrap--remove-properties'.
Keep `min-width' in `visual-wrap--safe-display-specs' so that
lines where other modes install it are not skipped.
* lisp/visual-wrap.el (visual-wrap--content-prefix): Return pixel
width instead of column count; drop the `min-width' strip.
(visual-wrap--apply-to-line): Drop `min-width' on line 1; use
mixed-unit `:align-to' sum form for the continuation wrap-prefix.
(visual-wrap--adjust-prefix): Handle only string prefixes; the
numeric (pixel) case is now handled inline in `--apply-to-line'
via the mixed-unit `:align-to' sum form.
(visual-wrap--remove-properties): Drop `min-width' removal.
(visual-wrap--safe-display-specs): Add note about `min-width'.
* test/lisp/visual-wrap-tests.el: Update expected `wrap-prefix'
values to the new sum form.
(visual-wrap-tests/invisible-prefix): New test motivated by bug#81039.
(visual-wrap-tests/negative-extra-indent): New test; verify that a
large negative `visual-wrap-extra-indent' produces a valid
wrap-prefix (the display engine clamps the stretch to zero).
* test/manual/visual-wrap-test.el: New file. Manual test suite
for visual-eyeball verification of prefix alignment behavior.
Reported-by: Andrea Alberti <a.alberti82@gmail.com>
Co-authored-by: Stefan Monnier <monnier@iro.umontreal.ca>
Building code with code is tricky. E.g. the code in gnus-draft-setup
suffered from a security issue because it forgot to quote
some arguments (see commit 142b1e0d4c).
* lisp/gnus/gnus-salt.el (gnus-pick-setup-message):
* lisp/gnus/gnus-msg.el (gnus-inews-add-send-actions):
* lisp/gnus/gnus-draft.el (gnus-draft-setup): Prefer passing functions
to message-add-action over passing it ELisp expressions.
* lisp/gnus/message.el (message-do-actions): Drop errors but not silently.
These were reported by happy-barney on GitHub
https://github.com/HaraldJoerg/cperl-mode/issues
* lisp/progmodes/cperl-mode.el (cperl-init-faces): Don't mistake
$method as a method declaration.
Move matcher for "use require" higher to prevent "require" being
fontified as keyword.
* test/lisp/progmodes/cperl-mode-resources/sub-names.pl: Add a
test case for $method
* test/lisp/progmodes/cperl-mode-tests.el
(cperl-test-fontify-declarations): Add a test case for a module
name looking like a keyword
(cperl-test-fontify-sub-names): Verify that $method does not
declare a method
* src/cmds.c (internal_self_insert): If the autofill function changed
the newline character we inserted, don't attempt to restore point.
* test/src/cmds-tests.el (self-insert-nonascii-autofill): New.
* src/casefiddle.c (do_casify_natnum): Extend range a little, to cover
character events with all modifier flags set.
* test/src/casefiddle-tests.el (casefiddle-allflags): New test.
This would (rarely) result in composition properties being shared
across the concatenation of two copies of a string.
* src/editfns.c (styled_format): Include the first argument in the
range.
* test/src/editfns-tests.el (editfns-tests--format-composition-property):
New.
* src/treesit.c (treesit_check_position): Return the validated
value.
(Ftreesit_node_first_child_for_pos):
(Ftreesit_node_descendant_for_range):
(Ftreesit_query_capture):
(Ftreesit__linecol_at): Fold calls to fix_position into
treesit_check_position in treesit.c.
This pacifies GCC 16.1.1 x86-64 -Warray-bounds when compiling with
-fsanitize=address. It’s also cleaner on more-typical platforms.
* src/termhooks.h (EVENT_INIT): Define via a compound literal
rather than via a memset plus an assignment. This evaluates the
argument lvalue only once, and is more likely to catch type errors.
* src/sfnt.c (memory_full_up) [TEST]: New static function.
(xmalloc, xcalloc, xrealloc): Use it instead of aborting.
(eassert) [TEST]: Remove; no longer needed.
(xaddmalloc, xaddrealloc): New static convenience functions.
(sfnt_read_cmap_format_12, sfnt_read_loca_table_short)
(sfnt_read_loca_table_long, sfnt_read_glyf_table)
(sfnt_read_simple_glyph, sfnt_read_compound_glyph)
(sfnt_read_glyph, sfnt_build_append, sfnt_build_outline_edges)
(sfnt_raster_glyph_outline, sfnt_build_outline_fedges)
(sfnt_raster_glyph_outline_exact, sfnt_read_hmtx_table)
(sfnt_read_name_table, sfnt_read_meta_table)
(sfnt_read_ttc_header, sfnt_read_fpgm_table)
(sfnt_read_prep_table, sfnt_create_uvs_context)
(sfnt_read_gvar_table, sfnt_read_packed_deltas)
(sfnt_vary_simple_glyph, sfnt_vary_compound_glyph):
Use the new functions to test for size overflow more reliably.
Use a cleaner way to decide whether a pointer addresses
the heap not the stack and thus needs freeing.
Fix a few more unlikely overflows.
(sfnt_read_cmap_format_12): The recently-added eassert is no
longer needed, so remove it.
(sfnt_read_simple_glyph): Initialize glyph->simple early
to simplify later code, as is done in similar functions.
Complicate size test to avoid potential unsigned overflow.
Even the earliest Android had plain ‘long long’,
so use that instead of doing it by hand.
* src/sfnt.c (struct sfnt_large_integer, sfnt_multiply_divide_1)
(sfnt_multiply_divide_2, sfnt_large_integer_add)
(sfnt_multiply_divide_round) [!INT64_MAX]: Remove.
(sfnt_multiply_divide, sfnt_multiply_divide_rounded)
(sfnt_mul_fixed, sfnt_mul_fixed_round, sfnt_div_fixed)
(sfnt_div_f26dot6, sfnt_mul_f26dot6, sfnt_mul_f26dot6_round)
(sfnt_mul_f2dot14, sfnt_dot_fix_14): Simplify by using long long.
* src/xselect.c (selection_data_for_offset):
Offset is size_t, not long, since that’s what caller passes.
(selection_data_size): Truncate large selection sizes to a value
that is more likely to work without involving undefined behavior.
Do not exceed X_ULONG_MAX which is all X can handle, or
PTRDIFF_MAX which can confuse underlying code.
(x_start_selection_transfer): Invoke selection_data_size just once.
* src/alloc.c (xcalloc): New function.
* src/dispnew.c (save_current_matrix):
* src/fns.c (Finternal__hash_table_histogram):
* src/nsfns.m (Fns_display_monitor_attributes_list):
* src/pgtkfns.c (Fpgtk_display_monitor_attributes_list):
* src/pgtkselect.c (pgtk_own_selection):
* src/profiler.c (make_log):
* src/sfnt.c (sfnt_poly_edges_exact):
* src/xfns.c (x_get_monitor_attributes_xinerama)
(x_get_monitor_attributes_xrandr, Fx_display_monitor_attributes_list):
Use it instead of multiplying by hand, conceivably with overflow.
* src/profiler.c (make_log):
Check for overflow in internal size calculations.
Use xnmalloc instead of multiply + xmalloc.
* src/sfnt.c (xzalloc) [TEST]: Remove, replacing with ...
(xicalloc) [TEST]: ... this new function. All callers changed.
(eassert) [TEST]: New macro.
* src/treesit.c (treesit_calloc_wrapper): Remove, replacing its
use with xcalloc.
* exec/exec.c (format_pid): Simplify. No need for a local array.
(exec_0): Shrink local buffer.
If names are too long, fail instead of silently truncating them.
Be cautious in case symlink is zero length (shouldn’t be possible
in Android, but it’s easy to be safe).
* src/xterm.c (x_dnd_begin_drag_and_drop): The n_ask_actions arg
is an int, not a size_t, as XChangeProperty supports only int and
our caller passes an int.
