* lisp/net/shr.el (shr-outline-search):
Don't check for the beginning of the line.
Suggested by Omar Antolín Camarena <omar@im.unam.mx>.
Confirmed by Rahguzar <rahguzar@mailbox.org>.
Quickly verify that the number of unchanged characters before and
after the region that changes are supposed to be limited to doesn't
change.
Suggested by Stefan Monnier.
* lisp/subr.el (combine-change-calls-1): Copy BEG into a marker, and
count the characters after the changed region. After the changes,
ensure the character counts still match.
* test/lisp/subr-tests.el (subr-test-combine-change-calls-error): New.
* lisp/auth-source.el (auth-source-backend-parse): Use `run-hook-wrapped'.
Suggested by Augusto Stoffel <arstoffel@gmail.com>.
Transform a warning into a debug message.
No reason not to. Tested with gfm-view-mode and markdown-ts-mode
* lisp/progmodes/eglot.el (eglot-client-capabilities): Annouce
markdown support for completion docs.
* lisp/textmodes/markdown-ts-mode.el
(markdown-ts--fontify-delimiter): When `markdown-ts-hide-markup'
is enabled in `markdown-ts-view-mode', also hide the whole line
containing a `fenced_code_block_delimiter' (including its
terminating newline) so Eldoc/LSP markdown snippets render
without stray blank lines around the code block. Scoped to
view-mode and to fenced delimiters on purpose: the same handler
is shared by inline delimiters (emphasis, code span, link
brackets) where munching surrounding whitespace would collapse
word separators, and tuning rendering for
hide-markup-while-editing is not a goal.
The grammar reports leading spaces as part of the atx_heading "marker"
and we cannot use the length of the marker as a result. Instead, count
the number of consecutive # after any blanks to determine its "level."
* lisp/textmodes/markdown-ts-mode.el
(markdown-ts--fontify-atx-heading): Count the octothorpes rather
than using the length of the marker node's text.
Do not assume every command run in
'markdown-ts--run-command-in-code-block' produces output that needs to
be merged from the temp/work buffer into the source buffer. One example
is 'xref-find-definitions', the temp buffer of which is unrelated to the
source buffer.
* lisp/textmodes/markdown-ts-mode.el
(markdown-ts-code-block-commands): Add 'complete-symbol'.
(markdown-ts-code-block-ignore-output-commands): New defvar.
(markdown-ts--run-command-in-code-block): Ignore command output
when necessary.
This avoids an alloca in x_set_mouse_color.
* src/xfns.c (x_set_mouse_color): Use local array rather than
alloca, since the string is small.
* src/xterm.c (X_ERROR_MESSAGE_SIZE): Move defn from here ...
* src/xterm.h: ... to here, and make it an enum not a macro.
* src/sysdep.c (serial_open): On failure, simply return -1
and set errno; do not call report_file_error, as the
caller is supposed to do that if needed.
Problem discovered on Fedora 44 x86-64 when using GCC 16.1.1 with
-fsanitize=address, with test/src/process-tests.el tests that use
process-tests--with-raised-rlimit. This function overrides the
default of 1024 for the maximum number of open files, which causes
undefined behavior (subscript errors) in src/process.c.
* src/process.c (inrange_fd, inrange_pipe): New functions.
(allocate_pty, create_process, create_pty, Fmake_pipe_process)
(Fmake_serial_process, connect_network_socket)
(network_interface_info, server_accept_connection)
(Fprocess_send_eof, child_signal_init):
Check that all newly allocated file descriptors are less than
FD_SETSIZE; close them and fail otherwise.
(create_pty, Fmake_pipe_process, Fmake_serial_process)
(connect_network_socket, server_accept_connection)
(child_signal_init): Remove no-longer-needed comparisons
to FD_SETSIZE, now that inrange_fd and inrange_pipe
do the checking for us.
* src/regex-emacs.c (forall_firstchar):
Avoid undefined behavior in the 2nd eassert when !bufp && !pend.
This pacifies GCC 16.1.1 20260501 (Red Hat 16.1.1-1) x86-64
when Emacs is configured with --enable-gcc-warnings.
Problem discovered with GCC 16.1.1 -fsanitize=undefined.
* src/fns.c (maybe_resize_hash_table): Avoid undefined
behavior when h->key_and_value or h->hash are null pointers,
in which case we call memcpy (destination, NULL, 0)
which has undefined behavior in C89 through C23.
* src/emacs-module.c (module_bignum_count_max):
Now of type ptrdiff_t, instead of likely being of type size_t.
(module_extract_big_integer): Omit now-unnecessary prefix +, a
now-unnecessary eassert against PTRDIFF_MAX, and and an
unnecessary cast to ptrdiff_t.
* src/comp.c (emit_static_object): Avoid an malloc/free of
a 1 KiB buffer; just put it on the stack. Use strnlen+mempcpy
instead of strncpy as there is no need to zero-fill buff.
Use int for values that must fit in int since we are
passing them to gcc_jit_context_new_rvalue_from_int.
Problem discovered with GCC 16.1.1 -fsanitize=undefined.
* src/data.c (Fash): Don’t left-shift a negative number;
behavior is undefined (ISO C23 § 6.5.8 ¶ 4).
When fetching images asynchronously, keep track of the
destination region and refrain from inserting the image if
that region has been modified in the mean time.
* lisp/net/shr.el (shr--image-fetched, shr--async-put-image): New functions.
(shr-insert-image, shr-zoom-image, shr-image-displayer, shr-tag-img):
Use them.
* lisp/mail/rmailmm.el (rmail-mime-render-html-shr): Add FIXME.
* lisp/erc/erc-stamp.el (erc-insert-timestamp-right): Explicitly add
face to right-hand stamp's `display' margin string. Commit d24b10ca
"Introduce 'margin' face for window margin background" did away with the
"fall-through" behavior in which a margin spec's string that lacks its
own face property inherited whatever the associated `display' char
happened to have. (Bug#80693)
;; The old language in "(elisp) Display Margins" said "if the
;; string to be displayed in the margin doesn't specify a face, its
;; face is determined using the same rules and priorities as it is
;; for strings displayed in the text area (see Displaying Faces),"
;; and that "if this results in undesirable 'leaking' of faces into
;; the margin, make sure the string has an explicit face specified
;; for it." It seems ERC and likely more than a few other packages
;; never came to heed this guidance and have erroneously depended
;; on such behavior for years if not decades.
;;
;; * test/lisp/erc/resources/fill/snapshots/merge-01-start.eld:
;; * test/lisp/erc/resources/fill/snapshots/merge-02-right.eld:
;; * test/lisp/erc/resources/fill/snapshots/merge-wrap-01.eld:
;; * test/lisp/erc/resources/fill/snapshots/merge-wrap-indicator-pre-01.eld:
;; * test/lisp/erc/resources/fill/snapshots/monospace-01-start.eld:
;; * test/lisp/erc/resources/fill/snapshots/monospace-02-right.eld:
;; * test/lisp/erc/resources/fill/snapshots/monospace-03-left.eld:
;; * test/lisp/erc/resources/fill/snapshots/monospace-04-reset.eld:
;; * test/lisp/erc/resources/fill/snapshots/spacing-01-mono.eld: Update
;; snapshot.
I ran into the issue described in the comment with the current
code in project-find-file-in and project-find-dir, when using
'C-x p p' to switch between projects.
* lisp/vc/vc-hooks.el (vc-refresh-state): When calling into the
backend, override any let-bindings of default-directory.
* lisp/net/newst-treeview.el
(newsticker--treeview-get-selected-item): If an item is already
selected, use it. (Bug#80972)
Copyright-paperwork-exempt: yes
* test/src/process-tests.el (process-tests/broken-pipe): Use
CONNECTION-TYPE.
(process-tests/broken-pipe/pipe-all)
(process-tests/broken-pipe/pipe-stdin): Skip via 'skip-when'.
(process-tests/broken-pipe/pty)
(process-tests/broken-pipe/pty-stdin): Remove these invalid tests; EPIPE
from a PTY doesn't make sense.
This ensures that if the child process closed its stdin and Emacs tries
to write to it, the process can still do any remaining work and exit
normally. In practice, this can occur with commands like "head(1)"
(bug#79079).
* src/fileio.c (file_for_stream): New function, extracted from...
(Fset_binary_mode): ... here.
(Ffile__close_stream): New function.
* src/process.c (send_process): When encountering EPIPE, only close the
fd for the pipe to the child process's stdin.
* lisp/eshell/esh-io.el (eshell-output-object-to-target): Don't check
for process liveness anymore.
* test/src/process-tests.el (process-tests/broken-pipe): New function.
(process-tests/broken-pipe/pipe, process-tests/broken-pipe/pty)
(process-tests/broken-pipe/pipe-stdin)
(process-tests/broken-pipe/pty-stdin): New tests.
* etc/NEWS: Announce this change.
* test/src/process-tests.el (start-process-should-not-modify-arguments):
Clean up test process.
(process-test--check-pipe-process): New macro...
(process-test-make-pipe-process-no-buffer): ... call it.
* src/lisp.h (SAFE_ALLOCA_LISP_EXTRA): Remove. It is no longer
used, it makes life more difficult in the feature/igc3 branch,
and having it around tempted me to start using it again.
* src/editfns.c (styled_format): Don’t call SAFE_ALLOCA_LISP_EXTRA,
as this makes life more difficult in the feature/igc3 branch.
Also, allocate another byte for the format string trailing '\0',
so that we don’t rely in the str2num trick with trailing '\1'.
Problems reported by Pip Cet (Bug#81057#32).
Also, check alloca size more exactly. Ameliorate the extra
conditional branches by doing all the internal size calculations
before a conditional branch on overflow.
All tests now use 'jsonrpc--with-python-fixture' with a Python3
subprocess instead of the in-Emacs TCP server. Changed the "harakiri"
method to be a request instead of a notification for to reduce chance of
"Sentinel hasn't run" warning.
The two in-Emacs-RPC-specific error tests ('errors-with--32601' and
'signals-an--32603-JSONRPC-error') are dropped with the fixture itself,
as the error paths they exercise are internal to the Emacs Lisp
dispatcher and have no direct Python equivalent. They will have to be
re-done later on in other form.
* test/lisp/jsonrpc-resources/server-emacsrpc.py: New file.
* test/lisp/jsonrpc-resources/server-anxious-nested.py: Use new
harakiri.
* test/lisp/jsonrpc-resources/server-emacsrpc.py: Use new harakiri.
* test/lisp/jsonrpc-resources/server-harakiri.py: Use new harakiri.
* test/lisp/jsonrpc-resources/server-remote-during-sync-1.py: Use new
harakiri.
* test/lisp/jsonrpc-resources/server-remote-during-sync-2.py: Use new
harakiri.
* test/lisp/jsonrpc-resources/server-remote-error.py: Use new harakiri.
* test/lisp/jsonrpc-resources/common.py (harakiri): New definition.
* test/lisp/jsonrpc-tests.el
(jsonrpc--with-python-fixture): Rework, move up.
(jsonrpc-connection-ready-p): Move up.
(jsonrpc--call-with-emacsrpc-fixture)
(jsonrpc--with-emacsrpc-fixture)
(errors-with--32601)
(signals-an--32603-JSONRPC-error): Remove.
(returns-3, times-out, doesnt-time-out, stretching-it-but-works)
(deferred-action-toolate, deferred-action-intime)
(deferred-action-complex-tests): Migrate to Python fixture.
(scontrol-remote-during-sync-1, scontrol-remote-during-sync-2)
(scontrol-anxious-nested, scontrol-remote-error)
(shutdown-clean-after-notification): Tweak.
The fringe indicator uses a custom lightning-bolt bitmap, an alternative
to the margin indicator on GUI frames. It is non-interactive, however.
* lisp/progmodes/eglot.el (eglot--fringe-action): New fringe bitmap.
(eglot-code-action-indications): Add 'left-fringe' to default value
and to docstring. Update incompatibility note.
(eglot-code-action-suggestion): Handle 'left-fringe' indication.
Problem reported by Helmut Eller (Bug#81057).
On x86-64 this patch shrinks USEFUL_PRECISION_MAX
from 16382 to 1074, SPRINTF_BUFSIZE from 21318 to 1386,
and sizeof initial_buffer from 22318 to 2386.
This also fixes a problem reported privately by Pip Cet:
exactly formatting the smallest positive IEEE double 2**-1074
with %f needs %.1074f and 1074 is DBL_MANT_DIG - DBL_MIN_EXP,
not merely 1 - DBL_MIN_EXP.
* src/editfns.c (USEFUL_PRECISION_MAX, SPRINTF_BUFSIZE):
Base these on double, not long double, since long double
is not the worst case (it is used only for converted u?intmax_t).
(USEFUL_PRECISION_MAX): Add DBL_MANT_DIG - 1 as per Pip Cet.
* src/editfns.c (styled_format): Streamline allocation of
auxiliary tables, by allocating them all in one go
rather than via separate alloca / mallocs.
