* src/frame.c (delete_frame): Remove a local with UNINIT that is
no longer needed now that GCC bug#85563 is fixed, as we don’t
need to worry about porting --enable-gcc-warnings to older GCC.
This change reverts this part of my commit
“Port --enable-gcc-warnings to GCC 8”
8c3215e7a4
dated Sat Apr 28 16:49:24 2018 -0700.
This approach ensures we don't use an SSDATA pointer after GC, and
that no Lisp callback code can modify the format string while we're
working on it.
* src/editfns.c (styled_format): Operate on a copy of the format
string rather than the original. Ensure final NUL byte is copied.
* src/w32console.c (w32con_clear_end_of_line): Set the space
glyphs' frame to NULL.
(w32con_write_glyphs): Handle face_id_frame == NULL, when called
from 'w32con_clear_end_of_line'.
(tty_draw_row_with_mouse_face): Adjust to changes in term.c.
* src/dispnew.c (gui_update_window_end): Don't consider mouse face
overwritten.
* src/xdisp.c (show_mouse_face): Accept an additional argument;
redefine the mouse cursor only if that argument is 'true'. All
callers changed. (Bug#75931)
* src/eval.c (internal_lisp_condition_case): Omit an unnecessary
‘volatile’ and an unnecessary pointer-to-volatile local var.
Perhaps these were needed in previous versions of Emacs, or to
pacify older versions of GCC when using --enable-gcc-warnings,
but they are not needed to pacify current GCC.
The "#$" syntax is recognized by Fread, which substitutes
Vload_file_name in its place. If Vload_file_name is bound
appropriately, no other value can produce an object EQ to the one
produced by "#$".
We use this to check the data vector for entries that we know should
have been initialized: if the value is still equal to what we bound
Vload_file_name to when it was read, it wasn't initialized, and we
abort.
* lisp/emacs-lisp/comp.el (comp--#$): New defvar.
(comp--finalize-container): Use it.
* src/comp.c (ABI_VERSION): Bump.
(emit_static_object): Ensure 'comp--#$' prints as "#$".
(load_static_obj): Ensure '#$' reads as Vcomp__hashdollar.
(check_comp_unit_relocs): Adjust assertion.
(syms_of_comp): Define 'comp--#$'.
* src/pdumper.c (dump_do_dump_relocation): Adjust assertion.
* src/term.c (handle_one_term_event): Don’t access possibly
uninitialized storage if frame is nil. This fixes an issue introduced
in commit 5eae7f5227 dated 2025-01-26
14:43:51 -0800. Issue caught by --enable-gcc-warnings, which enables
-Wanalyzer-use-of-uninitialized-value with gcc (GCC) 14.2.1 20250110
(Red Hat 14.2.1-7).
* src/treesit.c (resolve_language_symbol): Move forward.
(treesit_ensure_query_compiled): Resolve language remapping
here.
(Ftreesit_parser_list): Don't resolve language remaping here,
because there's no need: parsers now carries the remapped
language.
(Ftreesit_query_compile): Don't resolve language remapping here,
because we resolve language remapping when actually compiling
the query. Also we want the query to carry the unmapped
language.
* src/dispnew.c (write_matrix): Don't set the cursor here. Parameter to
set the cursor or not removed.
(combine_updates_for_frame): Call tty_set_cursor here, depending on
whether or not the selected frame is related to the updated frame.
(update_frame_with_menu): Adapt to change of write_matrix.
* src/dispnew.c (WINDOW_TO_FRAME_VPOS, WINDOW_TO_FRAME_VPOS): Macros
removed.
(window_to_frame_vpos, window_to_frame_vpos): Always define, and use
them instead of the removed macros.
* src/frame.c (make_frame): Handle 'expose-hidden-buffer'.
(expose-hidden-buffer): New variable to handle hidden buffers.
* src/buffer.c (Fother_buffer): Mention that it does not return
a hidden buffer.
* lisp/frame.el (make-frame): In doc-string describe handling
of hidden buffers.
* doc/lispref/frames.texi (Creating Frames): Explain handling
of hidden buffers in description of 'make-frame'.
* src/xdisp.c (vmessage): Avoid undefined behavior if
FRAME_MESSAGE_BUF_SIZE (f) is zero, or if doprnt generates output
containing only encoding errors. Although it’s not clear whether
either is possible, it is better to be safe. Also, clarify via a
new local message_bufsize.
* src/xdisp.c (vmessage): Increase buffer size to fit an extra
multibyte character. On buffer overflow, drop the last multibyte
character to determine an accurate byte length.
With ‘configure --enable-gcc-warnings’,
gcc (GCC) 14.2.1 20250110 (Red Hat 14.2.1-7) on x86-64
complained, as it apparently did not want to assume that enum
variables have one of the listed values.
Work around the problem by using tables rather than
switch statements. This gives up a little static
checking but I don’t see any other easy way to pacify
GCC without disabling the warnings entirely.
* src/pdumper.c (dump_fwd_int, dump_fwd_bool, dump_fwd_obj)
(dump_fwd_buffer_obj, dump_fwd_kboard_obj):
Last arg is now void const * so that these functions’ addresses
can all be put into the same array.
(dump_fwd, dump_anonymous_allocate_posix, dump_map_file_posix):
Use an array rather than a switch.
(mem_prot_posix_table): New static constant table.
* src/pgtkterm.c (pgtk_enumerate_devices): Prefer
make_formatted_string to snprintf + build_string + eassert, as
it’s simpler and won’t crash Emacs if the eassert fails.
* src/comp.c (format_string): Prefer strcpy to doing things by hand
in a place where strcpy is easier to read, generates
more-efficient code, and similar parts of Emacs do strcpy.
* src/pgtkterm.c: Ignore -Wanalyzer-null-dereference.
I don’t have time to look into all the diagnostics,
but the ones I checked seemed to be false alarms with
gcc -std=gnu23 on x86-64, and we are ignoring this
diagnostic in other modules. This is GCC 14.2.1
20250110 (Red Hat 14.2.1-7).
* src/window.c (last_mouse_window): New global var. All static
instances removed, and all their uses replaced with this global
var. This fixes a very unlikely bug where last_mouse_window was
GC’ed and a new window created in its place. It also fixes several
places that assumed NIL_IS_ZERO without static_asserting it.
(init_window_once): Initialize the new var.
* src/pgtkselect.c (pgtk_size_for_format): Remove.
(pgtk_get_window_property): Third arg is now void **, not unsigned
char **. Omit last arg. All callers changed. Simplify
allocation to pacify GCC 14.2.1 complaint about malloc size
mismatching size of array: there’s no need for an extra byte when
converting from ldata to idata.
(selection_data_to_lisp_data): Third arg is now void const *,
not unsigned char const *. All callers changed.
Omit unnecessary casts. Make some locals more local.
