GitLab

 src/xdisp.c (redisplay_window): Instead of moving point out of
 scroll margin, reject the force_start method, and try scrolling.

This undoes the previous change.  The check slows down the
interpreter, and is not needed to prevent a crash.  See
<http://lists.gnu.org/archive/html/emacs-devel/2013-07/msg00693.html>.
* doc/lispref/eval.texi (Special Forms): Mention 'lambda'.  Also, say that
non-well-formed expressions result in unspecified behavior, though
Emacs will not crash.

* configure.ac (etc, lisp): No need to create specially.
Configure already creates lisp when generating lisp/Makefile;
src/Makefile now creates etc when needed.

* src/Makefile.in ($(etc)/DOC, temacs$(EXEEXT)): Ensure etc/ exists.

* configure.ac (USE_NCURSES): New symbol.
* src/dispnew.c (init_display): Depend on USE_NCURSES, not GNU_LINUX,
to decide whether ncurses is being used.  Without this change,
GCC complains about tgetent not being declared, on a system
that has tinfo installed but ncurses not installed.

* data.c (Fsetq_default):
* eval.c (Fif, Fcond, Fprog1, Fsetq, Fquote, Ffunction, Fdefvar)
(Fdefconst, FletX, Flet, Fwhile, Fcatch, Funwind_protect)
(Fcondition_case):
Tune by taking advantage of the fact that ARGS is always a list
when a function is declared to have UNEVALLED args.

GCC 4.6 complains about -Wunused-local-typedefs, introduced in 4.7.

when including <sys/user.h>, to prevent Sparc/ARM machine/frame.h
from messing up Emacs's 'struct frame'.

Fixes: debbugs:14923

It was inadvertently omitted.  It's needed only if
HAVE_MENUS && ! (USE_X_TOOLKIT || USE_GTK).

frame have focus.

Fixes: debbugs:14895

* callproc.c (child_setup): When the child's exec fails, output
the program name, as that's more useful.  Use O_NONBLOCK to avoid
deadlock.
* process.c (create_process_1): Remove; no longer needed.
(create_process): Remove timer hack; no longer needed, now that
the child avoids deadlock.

* configure.ac (PTY_TTY_NAME_SPRINTF): Use PTY_NAME_SIZE,
not sizeof pty_name, since pty_name is now a pointer to the array.
* src/process.c (PTY_NAME_SIZE): New constant.
(pty_name): Remove static variable; it's now auto.
(allocate_pty): Define even if !HAVE_PTYS; that's simpler.
Take pty_name as an arg rather than using a static variable.
All callers changed.
(create_process): Recover pty_flag from process, not from volatile local.
(create_pty): Stay inside array even when pty allocation fails.
(Fmake_serial_process): Omit unnecessary initializaiton of pty_flag.

Mention that it's needed only for older GCCs.

(decode_coding_iso_2022): Check the single-shift area.  (Bug#8522)

* filelock.c (create_lock_file) [!O_CLOEXEC]: Use fcntl with FD_CLOEXEC.
(create_lock_file): Use write, not emacs_write.
* image.c (slurp_file, png_load_body):
* process.c (Fnetwork_interface_list, Fnetwork_interface_info)
(server_accept_connection):
Don't leak an fd on memory allocation failure.
* image.c (slurp_file): Add a cheap heuristic for growing files.
* xfaces.c (Fx_load_color_file): Block input around the fopen too,
as that's what the other routines do.  Maybe input need not be
blocked at all, but it's better to be consistent.
Avoid undefined behavior when strlen is zero.

(NSTATICS): Now a constant; doesn't need to be a macro.

 src/keyboard.c (kbd_buffer_get_event): Use Display_Info instead of
 unportable 'struct x_display_info'.
 (DISPLAY_LIST_INFO): Delete macro: not needed, since Display_Info
 is a portable type.

(procfs_ttyname): Don't use uninitialized storage if emacs_fopen
or fscanf fails.
(system_process_attributes): Prefer plain char to unsigned char
when either will do.  Clean up properly if interrupted or if
memory allocations fail.  Don't assume sscanf succeeds.  Remove
no-longer-needed workaround to stop GCC from whining.  Read
command-line once, instead of multiple times.  Check read status a
bit more carefully.

The code assumed that int is treated like ptrdiff_t in a vararg
function, which is not a portable assumption.  There was a similar
-- though these days less likely -- porting problem with various
assumptions that pointers of different types all smell the same as
far as vararg functions is conserved.  To make this problem less
likely in the future, redo the API to use varargs functions.
* alloc.c (make_save_value): Remove this vararg function.
All uses changed to ...
(make_save_int_int_int, make_save_obj_obj_obj_obj)
(make_save_ptr_int, make_save_funcptr_ptr_obj, make_save_memory):
New functions.
(make_save_ptr): Rename from make_save_pointer, for consistency with
the above.  Define only on platforms that need it.  All uses changed.

(DISPLAY_LIST_INFO): New macro.
(kbd_buffer_get_event): Do not access members that are not present
in X11.  Revert nadvertent change of "!=" to "=".

* lisp/frame.el (blink-cursor-timer-function, blink-cursor-suspend):
  Add check for W32.
* src/keyboard.c (kbd_buffer_get_event):
* src/w32term.c (x_focus_changed): Port FOCUS_(IN|OUT)_EVENT changes to W32.

(get_boot_time_1): Rework to avoid using emacs_open.
This doesn't actually fix a leak, but is better anyway.
(read_lock_data): Use read, not emacs_read.

* doc.c (get_doc_string): Fix memory leak when doc file absent.
(get_doc_string, Fsnarf_documentation):
Fix file descriptor leak on error.

* term.c (Fresume_tty) [!MSDOS]: Close fd if fdopen (fd) fails.
(init_tty) [!DOS_NT]: Likewise.  Also close fd if isatty (fd) fails.

Include <errno.h>.
(load_charset_map_from_file): Don't leak file descriptor on error.
Use plain record_xmalloc since the allocation is larger than
MAX_ALLOCA; that's simpler here.  Simplify test for exhaustion
of entries.
* eval.c (record_unwind_protect_nothing):
* fileio.c (fclose_unwind):
New functions.
* lread.c (load_unwind): Remove.  All uses replaced by fclose_unwind.
The replacement doesn't block input, but that no longer seems
necessary.

(Fload): Close some races that leaked fds or streams when 'load'
was interrupted.
(Fload, openp): Report error number of last nontrivial failure to open.
ENOENT counts as trivial.
* eval.c (do_nothing, clear_unwind_protect, set_unwind_protect_ptr):
New functions.
* fileio.c (close_file_unwind): No need to test whether FD is nonnegative,
now that the function is always called with a nonnegative arg.
* lisp.h (set_unwind_protect_ptr, set_unwind_protect_int): Remove.
All uses replaced with ...
(clear_unwind_protect, set_unwind_protect_ptr): New decls.

* callproc.c (Fcall_process):
* doc.c (Fsnarf_documentation):
* fileio.c (Frename_file, Fadd_name_to_file, Fmake_symbolic_link):
* process.c (set_socket_option):
Don't let a constructor trash errno.
* doc.c: Include <errno.h>.

* callproc.c (Fcall_process_region): Pass nil, not "/dev/null",
to Fcall_process when the input is empty.  This simplifies the
code a bit.  It makes no difference on POSIXish platforms but
apparently it fixes a bug on w32.

Fixes: debbugs:14885

* fileio.c (close_file_unwind): Don't close if FD is negative;
this can happen when unwinding a zapped file descriptor.
(Finsert_file_contents): Unwind-protect the fd before the point marker,
in case Emacs runs out of memory between the two unwind-protects.
Don't trash errno when closing FD.
Zap the FD in the specpdl when closing it, instead of deferring
the removal of the unwind-protect; this fixes a bug where a child
function unwinds the stack past us.

Fixes: debbugs:14839

This also lessens the need to write wrappers for callbacks,
and the need for make_save_pointer.
* alloca.c (free_save_value):
* atimer.c (run_all_atimers):
Now extern.
* alloc.c (safe_alloca_unwind):
* atimer.c (unwind_stop_other_atimers):
* keyboard.c (cancel_hourglass_unwind) [HAVE_WINDOW_SYSTEM]:
* menu.c (cleanup_popup_menu) [HAVE_NS]:
* minibuf.c (choose_minibuf_frame_1):
* process.c (make_serial_process_unwind):
* xdisp.h (pop_message_unwind):
* xselect.c (queue_selection_requests_unwind):
Remove no-longer-needed wrapper.  All uses replaced by the wrappee.
* alloca.c (record_xmalloc):
Prefer record_unwind_protect_ptr to record_unwind_protect with
make_save_pointer.
* alloca.c (Fgarbage_collect):
Prefer record_unwind_protect_void to passing a dummy.
* buffer.c (restore_buffer):
* window.c (restore_window_configuration):
* xfns.c, w32fns.c (do_unwind_create_frame)
New wrapper.  All record-unwind uses of wrappee changed.
* buffer.c (set_buffer_if_live):
* callproc.c (call_process_cleanup, delete_temp_file):
* coding.c (code_conversion_restore):
* dired.c (directory_files_internal_w32_unwind) [WINDOWSNT]:
* editfns.c (save_excursion_restore)
(subst_char_in_region_unwind, subst_char_in_region_unwind_1)
(save_restriction_restore):
* eval.c (restore_stack_limits, un_autoload):
* fns.c (require_unwind):
* keyboard.c (recursive_edit_unwind, tracking_off):
* lread.c (record_load_unwind, load_warn_old_style_backquotes):
* macros.c (pop_kbd_macro, restore_menu_items):
* nsfns.m (unwind_create_frame):
* print.c (print_unwind):
* process.c (start_process_unwind):
* search.c (unwind_set_match_data):
* window.c (select_window_norecord, select_frame_norecord):
* xdisp.c (unwind_with_echo_area_buffer, unwind_format_mode_line)
(fast_set_selected_frame):
* xfns.c, w32fns.c (unwind_create_tip_frame):
Return void, not a dummy Lisp_Object.  All uses changed.
* buffer.h (set_buffer_if_live): Move decl here from lisp.h.
* callproc.c (call_process_kill):
* fileio.c (restore_point_unwind, decide_coding_unwind)
(build_annotations_unwind):
* insdel.c (Fcombine_after_change_execute_1):
* keyboard.c (read_char_help_form_unwind):
* menu.c (unuse_menu_items):
* minibuf.c (run_exit_minibuf_hook, read_minibuf_unwind):
* sound.c (sound_cleanup):
* xdisp.c (unwind_redisplay):
* xfns.c (clean_up_dialog):
* xselect.c (x_selection_request_lisp_error, x_catch_errors_unwind):
Accept no args and return void, instead of accepting and returning
a dummy Lisp_Object.  All uses changed.
* cygw32.c (fchdir_unwind):
* fileio.c (close_file_unwind):
* keyboard.c (restore_kboard_configuration):
* lread.c (readevalllop_1):
* process.c (wait_reading_process_output_unwind):
Accept int and return void, rather than accepting an Emacs integer
and returning a dummy object.  In some cases this fixes an
unlikely bug when the corresponding int is outside Emacs integer
range.  All uses changed.
* dired.c (directory_files_internal_unwind):
* fileio.c (do_auto_save_unwind):
* gtkutil.c (pop_down_dialog):
* insdel.c (reset_var_on_error):
* lread.c (load_unwind):
* xfns.c (clean_up_file_dialog):
* xmenu.c, nsmenu.m (pop_down_menu):
* xmenu.c (cleanup_widget_value_tree):
* xselect.c (wait_for_property_change_unwind):
Accept pointer and return void, rather than accepting an Emacs
save value encapsulating the pointer and returning a dummy object.
All uses changed.
* editfns.c (Fformat): Update the saved pointer directly via
set_unwind_protect_ptr rather than indirectly via make_save_pointer.
* eval.c (specpdl_func): Remove.  All uses replaced by definiens.
(unwind_body): New function.
(record_unwind_protect): First arg is now a function returning void,
not a dummy Lisp_Object.
(record_unwind_protect_ptr, record_unwind_protect_int)
(record_unwind_protect_void): New functions.
(unbind_to): Support SPECPDL_UNWIND_PTR etc.
* fileio.c (struct auto_save_unwind): New type.
(do_auto_save_unwind): Use it.
(do_auto_save_unwind_1): Remove; subsumed by new do_auto_save_unwind.
* insdel.c (struct rvoe_arg): New type.
(reset_var_on_error): Use it.
* lisp.h (SPECPDL_UNWIND_PTR, SPECPDL_UNWIND_INT, SPECPDL_UNWIND_VOID):
New constants.
(specbinding_func): Remove; there are now several such functions.
(union specbinding): New members unwind_ptr, unwind_int, unwind_void.
(set_unwind_protect_ptr): New function.
* xselect.c: Remove unnecessary forward decls, to simplify maintenance.

* fileio.c (Fcopy_file, Finsert_file_contents, Fwrite_region):
Say "Read error" and "Write error", rather than "I/O error", or
"IO error reading", or "IO error writing", when a read or write
error occurs.
* process.c (Fmake_network_process, wait_reading_process_output)
(send_process, Fprocess_send_eof, wait_reading_process_output):
Capitalize diagnostics consistently.  Put "failed foo" at the
start of the diagnostic, so that we don't capitalize the
function name "foo".  Consistently say "failed" for such
diagnostics.
* sysdep.c, w32.c (serial_open): Now accepts Lisp string, not C string.
All callers changed.  This is so it can use report_file_error.
* sysdep.c (serial_open, serial_configure): Capitalize I/O
diagnostics consistently as above.

If the file name is neither null nor a pair, package it up as a
singleton list.  All callers changed, both to this function and to
report_file_error.  This fixes a bug where the memory allocator
invoked by list1 set errno so that the immediately following
report_file_error reported the wrong errno value.