This ports to POSIXish platforms like macOS that lack SOCK_CLOEXEC.
Fix suggested by Eli Zaretskii in:
https://lists.gnu.org/r/emacs-devel/2018-12/msg00055.html
* lib-src/emacsclient.c (set_local_socket):
Don’t use SOCK_CLOEXEC; that’s cloexec_socket’s job.
Recent changes in sysdep.c and emacsclient unnecessarily
removed useful code from DOS_NT builds. This changeset
reinstates that code.
* nt/inc/ms-w32.h (tcdrain): Redirect to _commit.
(fdatasync): No need to redirect anymore.
* lib-src/emacsclient.c (flush_stdout): Don't avoid calling
tcdrain on DOS_NT platforms.
* src/sysdep.c (reset_sys_modes): Don't ifdef away the call to
tcdrain on DOS_NT platforms.
fdatasync is for storage devices, not ttys.
* admin/merge-gnulib (GNULIB_MODULES): Remove fdatasync.
* lib/fdatasync.c, m4/fdatasync.m4: Remove.
* lib-src/Makefile.in (LIB_FDATASYNC):
* src/Makefile.in (LIB_FDATASYNC):
Remove. All uses removed.
* lib-src/emacsclient.c [!DOS_NT]:
Include <termios.h>, for tcdrain.
* lib-src/emacsclient.c (flush_stdout):
* src/sysdep.c (reset_sys_modes): On ttys, use tcdrain instead
of fdatasync (except don’t use either function if DOS_NT).
* lib/gnulib.mk.in, m4/gnulib-comp.m4: Regenerate.
* lib-src/emacsclient.c (socket_status): New arg UID.
All uses changed.
(set_local_socket): Don’t create the unbound socket unless the
initial sanity checks on the socket file succeed; this
simplifies cleaning it up. Check socket ownership again
after connecting, to fix a race (Bug#33366).
This reverts commit c418c85617.
This reinstates the original fix, as it had nothing to do
with the behavior reported in bug#33571, which seems to be
the expected behavior.
* lib-src/emacsclient.c: Disable -Wformat-truncation=2,
to avoid false alarms about the new snprintf calls.
(local_sockname): New function.
(set_local_socket): Use it. Prefer XDG_RUNTIME_DIR (if set)
for location of socket directory. Avoid unnecessary memory
allocation by using snprintf to destination.
* lisp/server.el (server-socket-dir): Prefer XDG_RUNTIME_DIR if set.
* src/keyboard.c (KBD_BUFFER_SIZE): Now a constant, not a macro.
(kbd_fetch_ptr, kbd_store_ptr): These now always point somewhere
into kbd_buffer, instead of sometimes pointing just past the
end which led to serious bugs (Bug#33547). All uses changed.
(kbd_store_ptr): No longer volatile. This variable has not been
accessed by a signal handler for some time, it seems.
(next_kbd_event, prev_kbd_event): New functions.
(kbd_buffer_nr_stored, process_special_events): Simplify.
* lisp/autorevert.el (auto-revert-notify-add-watch):
Assert that a key in `auto-revert-notify-watch-descriptor-hash-list'
is a valid file notification descriptor. (Bug#33556)
The NS port uses indexes into a color table to specify the colors of
faces, whereas frames use RGBA pixel values. In
extend_face_to_end_of_line the two needed to be compared to ensure
that the backgrounds of certain faces are not extended to the edge of
the window, which was failing because of this difference, thus causing
a visual difference with other platforms. Convert from index to RGBA
when doing such comparisons.
* src/dispextern.h (FACE_COLOR_TO_PIXEL) [HAVE_NS]: New macro. Call
ns_color_index_to_rgba under NS only.
* src/nsgui.h: Add prototype for ns_color_index_to_rgba.
* src/nsterm.m (ns_color_index_to_rgba): New function. Converts a
color_table entry to corresponding RGBA pixel value.
* src/xdisp.c (extend_face_to_end_of_line): Call FACE_COLOR_TO_PIXEL
on face background color when comparing with frame color.
* lisp/gnus/gnus-sum.el: Remove explicit definition of
`gnus-summary-mode-hook', this is now created automatically.
* lisp/gnus/nnir.el (nnir-open-server): Attach `nnir-mode' to the
`gnus-summary-prepared-hook', instead of
`gnus-summary-mode-hook'. The latter no longer has access to the
buffer-local value of `gnus-newsgroup-name', which `nnir-mode'
needs.
74a3a79 (origin/emacs-26) Fix a typo in a doc string
911766d Minor markup fix in frames.texi
19ed1e9 * lisp/net/trampver.el (customize-package-emacs-version-alist...
d7132ad * lisp/mh-e/mh-e.el (customize-package-emacs-version-alist): ...
5f39260 * lisp/emacs-lisp/map-ynp.el (map-y-or-n-p): Pass format to m...
a291f62 Don't call xwidget functions until GTK has been initialized
f0531b8 Improve documentation of Ediff wordwise commands
2925ce5 Support Hunspell 1.7.0 in ispell.el
03bb7a8 Avoid clearing echo-area message by auto-save-visited-file-name
In particular, warning messages from compile_defun now contain the source
buffer name and line and column numbers. Typing CR on such a warning now
moves to the pertinent place in the source buffer.
This fixes bug #33475
* lisp/emacs-lisp/bytecomp.el (top-level): Require compile.elc?.
(emacs-lisp-compilation-file-name-or-buffer)
(emacs-lisp-compilation-parse-errors-filename-function): New
variables/constants.
(emacs-lisp-compilation-mode): New mode derived from compilation-mode.
(byte-compile-log-file): Check byte-compile-current-file for being a string,
not merely non-nil. Change wording in message from "buffer" to "in buffer".
Go into emacs-lisp-compilation-mode rather than the plain compilation-mode.
(compile-defun): Bind byte-compile-current-file to current-buffer, not nil.
* lisp/progmodes/compilation-mode
(compilation-parse-errors-filename-function): Amend comments to specify that
this function may return a buffer, and that it need not save the match data.
(Several places): Amend comments to allow for the use of a buffer rather than
a file name.
(compilation-next-error-function): If the "file name" in file struct is
actually a buffer, use it rather than compilation-find-file's result.
(compilation-get-file-structure): save-match-data around the call to
compilation-parse-errors-filename-function. Only call
command-line-normalize-file-name when `filename' is a string.
