* src/process.c (create_process):
(create_pty):
(Fmake_pipe_process):
(Fmake_serial_process): Don't add the read_fd if the filter is t
(which means that we don't accept output from the filter).
This addresses a FIXME comment in lisp/emacs-lisp/comp.el,
relating to the number of subsidiary processes used by
comp-run-async-workers in native compilation.
* admin/merge-gnulib (GNULIB_MODULES): Add nproc.
* doc/lispref/processes.texi (Process Information), etc/NEWS:
Document num-processors.
* lib/gnulib.mk.in, m4/gnulib-comp.m4: Regenerate.
* lib/nproc.c, lib/nproc.h, m4/nproc.m4:
New files, copied from Gnulib by admin/merge-gnulib.
* lisp/emacs-lisp/comp.el (w32-get-nproc): Remove decl.
(comp-effective-async-max-jobs): Use num-processors.
* src/process.c: Include nproc.h.
(Fnum_processors): New function.
(syms_of_process): Define ‘all’, ‘current’, ‘num-processors’.
* src/w32proc.c (Fw32_get_nproc): Add FIXME comment.
* test/src/process-tests.el (process-num-processors): New test.
If a TLS handshake is attempted before the completion of an
asynchronous TCP connection has been ascertained, our local state will
not be set up correctly for further progress and the sentinel "open"
event will never be sent. This can occur if sufficient time passes
after the initiation of an async TCP connection so that by the time
`wait_reading_process_output` is called, the connection has already
been established on the TCP level.
This somewhat timing-sensitive bug has plagued HTTPS connections on
some platforms, notably macOS, for a long time (bug#49449).
* src/process.c (wait_reading_process_output): Gate the TLS handshake
by the NON_BLOCKING_CONNECT_FD flag. The flag will be cleared as soon
as the TCP socket is found to be writable.
* test/src/process-tests.el (process-async-https-with-delay):
New test.
* src/callproc.c (get_current_directory): Rename from
'encode_current_directory' and add boolean ENCODE flag.
(Fcall_process): Interpret INFILE relative to the working directory
from which PROGRAM is run, not 'default-directory'.
(call_process): Use 'get_current_directory'.
* src/process.c (Fmake_process): Use 'get_current_directory'.
* src/process.h (get_current_directory): Rename decl from
'encode_current_directory'.
* src/sysdep.c (sys_subshell): Use 'get_current_directory' (bug#49283).
* src/process.c (wait_reading_process_output): When looping through
fds, continue from where we left off.
(syms_of_process): Vprocess_prioritize_lower_fds: New variable
(bug#48118).
* src/process.c (add_non_keyboard_read_fd): Make this a public function.
(add_process_read_fd): Fold old, static add_non_keyboard_read_fd guts
into here.
* src/xsmfns.c (ice_conn_watch_CB): Call add_non_keyboard_read_fd
(bug#43834).
While no deadlocks caused by the blocking write end have been reported
yet, marking both ends nonblocking is consistent and also recommended
in the GNU/Linux manpage of 'select'.
* src/process.c (child_signal_init): Mark write end of self-pipe as
nonblocking.
Otherwise Emacs might hang when trying to read the pipe twice in a
row. This is consistent with the other file descriptors we pass to
'pselect'.
* src/process.c (child_signal_init): Make read end of pipe
non-blocking.
When we're trying to notify a process about a status change, we need
to ignore the SIGCHLD pipe temporarily, otherwise the code would
likely not run into the timeout case that's necessary for a status
change to happen.
* src/process.c (wait_reading_process_output): Ignore the SIGCHLD pipe
when notifying a process about a status change.
* test/src/process-tests.el (process-tests/sentinel-called)
(process-tests/sentinel-with-multiple-processes): New unit tests.
If we receive and handle a SIGCHLD signal for a process while waiting
for that process, 'pselect' might never return. Instead, we have to
explicitly 'pselect' that the process status has changed. We do this
by writing to a pipe in the SIGCHLD handler and having
'wait_reading_process_output' select on it.
* src/process.c (child_signal_init): New helper function to create a
pipe for SIGCHLD notifications.
(child_signal_read, child_signal_notify): New helper functions to
read from/write to the child signal pipe.
(create_process): Initialize the child signal pipe on first use.
(handle_child_signal): Notify waiters that a process status has
changed.
(wait_reading_process_output): Make sure that we also catch
SIGCHLD/process status changes.
* test/src/process-tests.el
(process-tests/fd-setsize-no-crash/make-process): Remove workaround,
which is no longer needed.
We first need to register the received process ID so that
'handle_child_signal' checks it. Otherwise we might never call
'waitpid' for these processes, risking deadlock.
* src/callproc.c (call_process):
* src/process.c (create_process): Don't unblock SIGCHLD before
registering the process ID to wait for.
* src/callproc.c (emacs_spawn): Accept a signal set from the caller.
* configure.ac: Instead of AC_CHECK_HEADER, use AC_COMPILE_IFELSE
with X11/Intrinsic.h when checking for X11/extensions/Xrender.h.
This suppresses a bogus "report a bug to bug-gnu-emacs" diagnostic
from 'configure' in Solaris 10.
(SETUP_SLAVE_PTY): Adjust to recent renaming of forkin to
std_in in callproc.c. Needed on Solaris and Unixware.
* lib-src/Makefile.in (LIB_GETRANDOM, LIBS_ETAGS): New vars,
needed because on Solaris 10 the Gnulib tempname module now needs
the -lrt library for clock_gettime. Throw in the LIB_GETRANDOM
stuff too while we’re at it; from getrandom.m4 it seems to be
needed for MingW.
(LIBS_MOVE, etags_libs): Use them.
* src/callproc.c [SETUP_SLAVE_PTY]: Include sys/stream.h
and sys/stropts.h, for SETUP_SLAVE_PTY’s definiens.
* src/process.c [NEED_BSDTTY]: Don’t include bsdtty.h; hasn’t been
needed in years.
[USG5_4]: Don’t include sys/stream.h or sys/stropts.h; these
directives havbe been moved to callproc.c because the only use of
SETUP_SLAVE_PTY is there now.
74a77ef299 Improve documentation of 'network-lookup-address-info'
c6d5555646 Display messages sent using ERC's /say
c156723769 Fix Rmail summary display when From: header is malformed
* src/process.c (Fnetwork_lookup_address_info):
* doc/lispref/processes.texi (Misc Network): Document the error
message emitted by 'network-lookup-address-info' when it fails.
This works even if another thread or process resets the resource limit
for open file descriptors, e.g., using 'prlimit' on GNU/Linux.
* src/process.c (create_process, create_pty, Fmake_pipe_process)
(Fmake_serial_process, connect_network_socket)
(server_accept_connection): Limit file descriptors to FD_SETSIZE.
* test/src/process-tests.el (process-tests--with-raised-rlimit): New
helper macro.
(process-tests--fd-setsize-test): Rename from
'process-tests--with-many-pipes'. Increase resource limit during test
if possible.
(process-tests/fd-setsize-no-crash/make-process)
(process-tests/fd-setsize-no-crash/make-pipe-process)
(process-tests/fd-setsize-no-crash/make-network-process)
(process-tests--new-pty): Rename callers.
Previously this was only checked in a few places. Now assert that
file descriptors are within the expected range whenever we'd otherwise
introduce undefined behavior.
* src/process.c (add_read_fd, add_process_read_fd, delete_read_fd)
(recompute_max_desc, delete_write_fd, compute_input_wait_mask)
(compute_non_process_wait_mask, compute_non_keyboard_wait_mask)
(compute_write_mask, clear_waiting_thread_info)
(update_processes_for_thread_death, Fset_process_thread)
(create_process, create_pty, Fmake_pipe_process)
(Fprocess_datagram_address, Fset_process_datagram_address)
(Fmake_serial_process, finish_after_tls_connection)
(connect_network_socket, deactivate_process)
(server_accept_connection, wait_reading_process_output)
(read_process_output, read_and_dispose_of_process_output)
(send_process, Fcontinue_process, Fprocess_send_eof)
(Fprocess_filter_multibyte_p, keyboard_bit_set)
(add_timer_wait_descriptor, setup_process_coding_systems): Add
assertions to document and check that file descriptors are within the
expected range when used as file descriptor set elements or array
subscripts.
* src/w32.h (set_process_dir):
* src/w32proc.c (set_process_dir): Change the argument to 'const
char *'.
* src/lisp.h (make_environment_block):
* src/callproc.c (make_environment_block): Now returns 'char **'.
(exec_failed) [DOS_NT]: Remove unused function.
* src/callproc.c (child_setup): NEW_ARGV and ENV are now 'char **'.
Making them 'const' breaks the MinGW build and is not needed for
other platforms.
* src/callproc.c (emacs_spawn): ARGV and ENVP arguments are now
'char *', for the same reason.
* src/process.c (create_process): Adapt to above changes.
Getting the vfork + execve combination right isn't easy, and the code
was partially duplicated between callproc.c and process.c. Centralize
the spawn operation in a single function that deals with the nasty
details. Going forward, we should be able to use posix_spawn from
either libc or Gnulib (or CreateProcessW on Windows) in the non-pty
case.
* src/callproc.c (emacs_spawn): New function to start an asynchronous
subprocess. Merge code from 'call_process' and 'create_process' into
this function.
(call_process): Use new 'emacs_spawn' function.
(child_setup): Make static, since there are no users outside this
compilation unit left.
(CHILD_SETUP_TYPE): Move from header file, since there are no users
outside this compilation unit left.
* src/process.c (create_process): Use new 'emacs_spawn' function.
This avoids the impression that 'child_setup' could do anything
Lisp-related.
* src/callproc.c (child_setup): Pass C pointer to current directory
name.
(call_process): Adapt callers.
* src/process.c (create_process): Adapt callers.
While 'child_setup' carefully avoids calls to async-signal-unsafe
functions like 'malloc', it seems simpler and less brittle to use
normal allocation outside the critical section between 'fork' and
'exec'.
* src/callproc.c (make_environment_block): New function to create the
environment block for subprocesses. Code largely extracted from
'child_setup' and adapted to use 'xmalloc' instead of 'alloca'.
(child_setup): Remove environment block allocation in favor of
passing the environment block as command-line argument.
(call_process): Adapt to new calling convention.
* src/process.c (create_process): Adapt to new calling convention.
Give get-buffer-create an optional argument to inhibit buffer hooks
in internal or temporary buffers for efficiency (bug#34765).
* etc/NEWS: Announce new parameter of get-buffer-create and
generate-new-buffer, and that with-temp-buffer and with-temp-file
now inhibit buffer hooks.
* doc/lispref/buffers.texi (Buffer Names): Fix typo.
(Creating Buffers): Document new parameter of get-buffer-create and
generate-new-buffer.
(Buffer List, Killing Buffers): Document when buffer hooks are
inhibited.
(Current Buffer):
* doc/lispref/files.texi (Writing to Files): Document that
with-temp-buffer and with-temp-file inhibit buffer hooks.
* doc/lispref/internals.texi (Buffer Internals): Document
inhibit_buffer_hooks flag. Remove stale comment.
* doc/misc/gnus-faq.texi (FAQ 5-8):
* lisp/simple.el (shell-command-on-region): Fix indentation.
* lisp/files.el (kill-buffer-hook): Document when hook is inhibited.
(create-file-buffer):
* lisp/gnus/gnus-uu.el (gnus-uu-unshar-article):
* lisp/international/mule.el (load-with-code-conversion):
* lisp/mh-e/mh-xface.el (mh-x-image-url-fetch-image):
* lisp/net/imap.el (imap-open):
* lisp/net/mailcap.el (mailcap-maybe-eval):
* lisp/progmodes/flymake-proc.el
(flymake-proc--read-file-to-temp-buffer)
(flymake-proc--copy-buffer-to-temp-buffer): Simplify.
* lisp/subr.el (generate-new-buffer): Forward new optional argument
to inhibit buffer hooks to get-buffer-create.
(with-temp-file, with-temp-buffer, with-output-to-string):
* lisp/json.el (json-encode-string): Inhibit buffer hooks in buffer
used.
* src/buffer.c (run_buffer_list_update_hook): New helper function.
(Fget_buffer_create): Use it. Add optional argument to set
inhibit_buffer_hooks flag instead of comparing the buffer name to
Vcode_conversion_workbuf_name. All callers changed.
(Fmake_indirect_buffer, Frename_buffer, Fbury_buffer_internal)
(record_buffer): Use run_buffer_list_update_hook.
(Fkill_buffer): Document when buffer hooks are inhibited. Use
run_buffer_list_update_hook.
(init_buffer_once): Inhibit buffer hooks in Vprin1_to_string_buffer.
(Vkill_buffer_query_functions, Vbuffer_list_update_hook): Document
when hooks are inhibited.
* src/buffer.h (struct buffer): Update inhibit_buffer_hooks
commentary.
* src/coding.h (Vcode_conversion_workbuf_name):
* src/coding.c (Vcode_conversion_workbuf_name): Make static again
since it is no longer needed in src/buffer.c.
(code_conversion_restore, code_conversion_save, syms_of_coding):
Prefer boolean over integer constants.
* src/fileio.c (Finsert_file_contents): Inhibit buffer hooks in
" *code-converting-work*" buffer.
* src/window.c (Fselect_window): Fix grammar. Mention
window-selection-change-functions alongside buffer-list-update-hook.
* test/src/buffer-tests.el: Fix requires.
(buffer-tests-inhibit-buffer-hooks): New test.
A long time ago, `read_key_sequence` used to read the keymaps at the
start, so if something happened between this start and the moment
the user actually hits a key, `read_key_sequence` could end up using
the wrong keymaps. To work around this problem, the code used
`record_asynch_buffer_change` to try and trigger `read_key_sequence`
to re-read the keymaps in some known cases.
Several years ago, `read_key_sequence` was changed so as to read the keymaps
only once the user hits a key, making this machinery now redundant
(and also harmful apparently in bug#5803 because it introduces
"spurious" events).
So we here remove `record_asynch_buffer_change` and the
`BUFFER_SWITCH_EVENT` and `Qbuffer_switch` pseudo-events it generated.
* src/termhooks.h (enum event_kind): Delete `BUFFER_SWITCH_EVENT`.
* src/keyboard.c: (record_asynch_buffer_change): Delete function.
(syms_of_keyboard): Delete `Qbuffer_switch`.
(force_auto_save_soon, readable_events)
(kbd_buffer_store_buffered_event, kbd_buffer_get_event)
(make_lispy_event):
* src/xterm.c (handle_one_xevent):
* src/w32term.c (w32_read_socket):
* src/process.c (wait_reading_process_output)
(read_and_dispose_of_process_output, exec_sentinel): Simplify accordingly.
* src/process.c (init_process_emacs): force glib's g_unix_signal
handler into lib_child_handler where it should belong.
Copyright-paperwork-exempt: yes
* src/process.c (update_process_mark): Make into its own function.
(Fmake_process, Fmake_pipe_process, Fmake_serial_process)
(connect_network_socket): Use it.
* src/process.c (wait_reading_process_output): Before the select,
check every interesting gnutls stream for available data in the
buffer. If some of them hit, and either there is no wait_proc or
the wait_proc is one of the gnutls streams with new data, set the
select timeout to 0 after the select, and merge the gnutls buffer
status into the select returns (bug#40665). This fixes a problem
where TLS connections would sometimes hang.
* src/bignum.c (check_integer_range, check_uinteger_max)
(check_int_nonnegative): New functions.
* src/frame.c (check_frame_pixels): New function.
(Fset_frame_height, Fset_frame_width, Fset_frame_size): Use it.
* src/lisp.h (CHECK_RANGED_INTEGER, CHECK_TYPE_RANGED_INTEGER):
Remove these macros. Unless otherwise specified, all callers
replaced by calls to check_integer_range, check_uinteger_range,
check_int_nonnegative.
* src/frame.c (gui_set_right_divider_width)
(gui_set_bottom_divider_width):
* src/nsfns.m (ns_set_internal_border_width):
* src/xfns.c (x_set_internal_border_width):
Using check_int_nonnegative means these functions no longer
incorrectly reject negative bignums; they treat them as 0,
just like negative fixnums.
The handling of :coding nil was different between
make-{network,serial}-process and make-{pipe}process. Now they all
handle :coding nil as if :coding had not been specified.
* process.c (Fmake_serial_process)
(set_network_socket_coding_system): Use plist-get to check if
:coding has been specified instead of plist-member, to ensure that
":coding nil" does not override coding-system-for-{read,write}.
* network-stream-tests.el (check-network-process-coding-system-bind)
(check-network-process-coding-system-no-override)
(check-network-process-coding-system-override): New tests.
* etc/NEWS: Describe change in make-network-process and
make-serial-process :coding behavior.
A common complaint about the module API is that modules can't
communicate asynchronously with Emacs. While it isn't possible to
call arbitrary Emacs functions asynchronously, writing to a pipe
should always be fine and is a pretty low-hanging fruit.
This patch implements a function that adapts an existing pipe
process. That way, users can use familiar tools like process filters
or 'accept-process-output'.
* src/module-env-28.h: Add 'open_channel' module function.
* src/emacs-module.c (module_open_channel): Provide definition for
'open_channel'.
(initialize_environment): Use it.
* src/process.c (open_channel_for_module): New helper function.
(syms_of_process): Define necessary symbol.
* test/src/emacs-module-tests.el (module/async-pipe): New unit test.
* test/data/emacs-module/mod-test.c (signal_system_error): New helper
function.
(signal_errno): Use it.
(write_to_pipe): New function running in the background.
(Fmod_test_async_pipe): New test module function.
(emacs_module_init): Export it.
* doc/lispref/internals.texi (Module Misc): Document new module
function.
* doc/lispref/processes.texi (Asynchronous Processes): New anchor
for pipe processes.
* etc/NEWS: Document 'open_channel' function.
