Partly fixes bug#20241.
* lisp/emacs-lisp/bytecomp.el (byte-compile-setq): Issue a warning when a
`setq' form with an odd number of arguments is compiled.
Fixes bug#21983.
* lisp/emacs-lisp/cconv.el (cconv-convert): Don't silently insert a nil last
argument into a `setq' when there're an odd number of args. This enables the
byte compiler to issue a message in this case.
* src/lread.c (Fload): Call 'unbind_to' with 'Fmodule_load' as the
2nd arg, to avoid the "binding stack not balanced" error.
(syms_of_lread) <module-file-suffix>: New Lisp variable.
* lisp/files.el (module-file-suffix): Declare.
(load-file): Remove 'module-file-suffix' from
'completion-ignored-extensions', to allow completion on modules.
* etc/NEWS: Mention 'module-file-suffix'.
* lisp/simple.el (open-line): Remove INTERACTIVE argument.
* test/automated/simple-test.el (open-line-indent, open-line-hook):
Adjust accordingly.
This change finishes what my commit of Thu Nov 19 17:32:37 2015 -0600
(git commit c59353896) started. It turns out that having INTERACTIVE
cause `post-self-insert-hook' to run (via `newline') meant `open-line'
still had the electric indent behavior, as `post-self-insert-hook'
normally contains `electric-indent-post-self-insert-function' ever
since `electric-indent-mode' has been on by default. Tracing the code
change in `open-line' is mildly twisty, because Artur Malabarba's
earliest two commits of 24 Oct 2015 first removed the `interactive'
form entirely (git commit 6939896e2) and then restored it with the new
extra "p" already added (git commit bd4f04f86), such that there is no
single-commit diff in which one sees the second "p" appear. Thus this
change is effectively a reversion of parts of each of those commits.
This could close bug#21884, at least until further discussion.
* lisp/progmodes/etags.el (etags-tags-completion-table):
Adhere closer to the "implicit tag name" definition. Simplify
the regexp. Search for the explicit tag name first, and when
not found, search locally for the implicit one. (Bug#21934)
The problem was in:
* src/syntax.c (update_syntax_table_forward): Propertize even when truncated
which is hence not unreverted.
The rest is:
* src/syntax.h (UPDATE_SYNTAX_TABLE_FORWARD_FAST):
(UPDATE_SYNTAX_TABLE_FAST): Re-introduce.
All callers in regex.c changed back to the _FAST versions.
* test/automated/message-mode-tests.el: Tweak the test to rely on auto
propertization in backward-sexp.
Although the patch does fix Bug#21688 and prevents a core dump,
it also makes the message-mode-propertize test fail; see:
http://lists.gnu.org/archive/html/emacs-devel/2015-11/msg01667.html
Perhaps someone else can come up with a better fix some day.
* src/syntax.c (update_syntax_table_forward):
Propertize even when truncated.
* src/syntax.h (UPDATE_SYNTAX_TABLE_FORWARD_FAST):
(UPDATE_SYNTAX_TABLE_FAST): Remove.
All callers changed back to the non-_FAST versions.
This was motivated by the recent addition of module code,
which added some ENABLE_CHECKING-enabled checks that are
useful elsewhere too.
* src/alloc.c (compact_font_cache_entry):
* src/fns.c (sweep_weak_table):
* src/lread.c (oblookup):
Use gc_asize rather than doing it by hand.
* src/emacs-module.c (module_make_global_ref)
(module_free_global_ref, module_vec_size):
Omit assertions that lisp.h now checks.
* src/lisp.h (XFASTINT, ASIZE): In functional implementations,
check that the result is nonnegative. Use eassume, as this
info can help a bit when optimizing production code.
(XSYMBOL) [!USE_LSB_TAG]: Assert that argument is a symbol,
to be consistent with the USE_LSB_TAG case.
(gc_asize): New function, when ASIZE is needed in the gc.
(gc_aset): Use it.
(HASH_TABLE_P): Move definition up, so that it can be used ...
(XHASH_TABLE): ... here, to assert that the arg is a hash table.
* src/w32term.c (w32_initialize):
* src/w32console.c (initialize_w32_display):
* src/w32fns.c (globals_of_w32fns): Don't record the main thread
ID independently for each type of session (GUI, TTY, batch).
* src/w32term.c (w32_init_main_thread): New function, records the
main thread's thread ID.
* src/w32term.h: Add prototype for w32_init_main_thread.
* src/emacs.c (main) [WINDOWSNT]: Call w32_init_main_thread.
* src/emacs-module.c [WINDOWSNT]: Rename main_thread_id to
main_thread, for consistency with other threading libraries. All
users changed. Include w32term.h.
(check_main_thread) [WINDOWSNT]: Simplify the test: no need to
make sure the main thread is alive, as we hold a handle on it
opened by w32_init_main_thread.
(module_init) [WINDOWSNT]: Reuse the thread ID recorded by
w32_init_main_thread, instead of calling the requisite APIs once
more.
* src/window.c (grow_mini_window, shrink_mini_window): Set the
frame's 'window_sizes_changed' flag.
* src/xdisp.c (redisplay_internal): Call the hooks on
'window-size-change-functions' if the call to 'echo_area_display'
sets the frame's 'window_sizes_changed' flag.
(syms_of_xdisp) <window-size-change-functions>:
Update doc string to indicate the mini-window resizes trigger a
call to the hooks, and don't promise that will happen "before
redisplay". (Bug#19576, Bug#21333)
* doc/lispref/windows.texi (Window Hooks): Update the description
of 'window-size-change-functions'.
* src/fns.c (Frequire): Doc fix to include the dynamic module
support.
* src/lread.c (Fload, Vload_suffixes): Doc fixes to include the
dynamic module support.
(Fload): Treat the module suffix the same as '*.el' and '*.elc'
wrt the MUST-SUFFIX argument.
* etc/NEWS: Expand documentation of dynamically loaded modules.
* lisp/progmodes/xref.el (xref-backend-functions): Move the
default value into a separate `add-hook' call (bug#21964).
* lisp/progmodes/elisp-mode.el (emacs-lisp-mode):
Don't declare the xref-backend-functions variable.
It doesn't make any difference.
* src/emacs-module.c (module_free_global_ref): Add a FIXME
comment about error reporting. Fix a recently-introduced typo
that double-decremented the refcount.
* src/emacs-module.h (emacs_module_init): New decl.
Without it, GCC might complain about a module that defines
emacs_module_init without using it. This also checks the
API better.
* src/emacs-module.c (MODULE_SETJMP_1): Fix typo that I
introduced while reindenting the code earlier, and add a
comment explaining the unusual use of do-while here.
If a function on the hook `window-size-change-functions' doesn't
restore the current buffer, functions that save and restore the
current window configuration (like `y-or-no-p') could silently
change the current buffer. When `write-file' asked the user
confirmation to overwrite a file, `y-or-no-p' changed the current
buffer, and the wrong buffer was saved to the file.
* lisp/follow.el (follow-windows-start-end): Call `select-frame'
using the `norecord' parameter.
(follow-window-size-change): Restore current buffer. Call
`select-frame' using the `norecord' parameter. Cleanup.
* src/emacs-module.c (struct module_fun_env)
(module_make_function, module_funcall, Fmodule_call):
* src/emacs-module.h (struct emacs_runtime, struct emacs_env_25):
Use ptrdiff_t, not int, for arg counts.
* src/emacs-module.c (module_make_function): Don’t bother
checking arity against MOST_POSITIVE_FIXNUM, as that’s
unnecessary here. Make the checking clearer by negating it.
(module_make_function, Fmodule_call): No need to use xzalloc
since the storage doesn’t need to be cleared.
(module_funcall): Don’t use VLA, since C11 doesn’t guarantee support
for it, and many implementations are buggy with large VLAs anyway.
Use SAFE_ALLOCA_LISP instead.
(module_vec_set): Don’t crash if i < 0.
(module_vec_get): Don’t crash if i < MOST_NEGATIVE_FIXNUM.
(module_vec_set, module_vec_get): Do fixnum checks only when
i is out of array bounds, for efficiency in the usual case.
(Fmodule_load): Simplify fixnum range check.
(Fmodule_call): Simplify arity check. Use xnmalloc to detect
integer overflow in array allocation size.
* modules/mod-test/mod-test.c: Include stdlib.h, to avoid warnings
about missing prototype of malloc.
* modules/mod-test/Makefile (CFLAGS): Add -std=gnu99, to avoid
compiler warnings.
* src/dynlib.c [WINDOWSNT]: Include errno.h, lisp.h, and w32.h.
No need to include windows.h, as w32.h already does that.
<dynlib_last_err>: New static variable.
(dynlib_reset_last_error): New function.
(dynlib_open): Convert forward slashes to backslashes. Convert
file names from UTF-8 to either UTF-16 or the current ANSI
codepage, and call either LoadLibraryW or LoadLibraryA. If the
argument is NULL, return a handle to the main module, like
'dlopen' does. Record the error, if any, for use by dynlib_error.
(dynlib_sym): Check the handle for validity. Record the error, if
any, for use by dynlib_error.
(dynlib_error): Call w32_strerror to produce the error string, and
zero out the last error code, like dlerror does.
(dynlib_close): Check the handle for validity. Record the error,
if any, for use by dynlib_error. Don't call FreeLibrary with a
handle for the main module.
* src/w32.c (globals_of_w32): Call dynlib_reset_last_error.
* src/dynlib.c, src/emacs-module.c: Include <config.h> first.
* src/dynlib.h: Do not include config.h.
It’s every .c file’s responsibility to include config.h first.
* src/emacs-module.c: Include emacs-module.h immediately after
config.h, to test that emacs-module.h doesn’t depend on
include files other than config.h.
* src/lisp.h (PUSH_HANDLER): Remove.
All callers changed to use push_handler directly.
* src/eval.c (internal_condition_case)
(internal_condition_case_1, internal_condition_case_2)
(internal_condition_case_n):
Use same pattern as for other invokers of push_handler.
(push_handler, push_handler_nosignal): Use call-by-value
instead of call-by-reference. All uses changed.
(push_handler): Simplify by rewriting in terms of
push_handler_nosignal.
(push_handler_nosignal): Profile any newly allocated memory.
* lisp/version.el (emacs-repository--version-git-1): Do not assume
HEAD is at .git/HEAD, it can also be at .git/worktrees/<branch>/HEAD.
(emacs-repository-get-version): Grok linked worktrees when EXTERNAL
is nil too.
(occur-1, occur-engine): Use it.
* lisp/isearch.el (isearch-occur): Propertize regexp with
isearch-string and isearch-regexp-function-descr for
occur-regexp-descr to display the correct description
message in the header (bug#21176, bug#21180).
* lisp/simple.el (open-line): Remove electric indent code.
(electric-indent-just-newline): Don't declare.
* test/automated/simple-test.el (open-line-indent): Adjust test.
This partly reverts Artur Malabarba's change that added electric
indent sensitivity to `open-line' (Oct 24 22:26:27 2015 +0100, git
commit bd4f04f86), and adjusts a new test he added right afterwards
(Sat Oct 24 23:43:06 2015 +0100, git commit 207f235e3) accordingly.
However, the new INTERACTIVE argument to `open-line', which he also
added in the first commit, is not reverted here.
See the thread "Questioning the new behavior of `open-line'." on the
Emacs Devel mailing list, and in particular this message:
From: Artur Malabarba
Subject: Re: Questioning the new behavior of `open-line'.
To: Karl Fogel
Cc: David Kastrup, Pierpaolo Bernardi, emacs-devel
Date: Wed, 18 Nov 2015 21:03:58 +0000
Message-ID: \
<CAAdUY-KN06pvCMy5bt3+Buk3yeKjf6n9iB2FaSTTOPpCqPwyhA@mail.gmail.com>
https://lists.gnu.org/archive/html/emacs-devel/2015-11/msg01707.html
* src/emacs-module.c (Fmodule_load):
Simplify and avoid unnecessary initialization of priv member to 0.
* src/emacs-module.c: (module_vec_set, module_vec_get, module_vec_size)
Generally speaking, at the C level the Emacs source code prefers
signed types like ‘ptrdiff_t’ to unsigned types like ‘size_t’,
partly to avoid the usual signedness confusion when comparing values.
Change the module API to follow this convention.
Use ‘int’ for small values that can’t exceed INT_MAX.
* modules/mod-test/mod-test.c (Fmod_test_globref_make)
(Fmod_test_string_a_to_b, Fmod_test_vector_fill)
(Fmod_test_vector_eq):
* src/emacs-module.c (struct emacs_value_frame)
(module_make_global_ref, module_free_global_ref)
(module_copy_string_contents, module_make_string)
(module_vec_set, module_vec_get, module_vec_size):
* src/emacs-module.h (struct emacs_runtime, struct emacs_env_25):
* src/lread.c (suffix_p):
Prefer signed to unsigned integer types.
Remove ‘const’ qualifier from locals that were newly added.
We don’t normally bother declaring locals with ‘const’ even
though they are not modified, for the same reason we don’t
bother declaring them with ‘register’ even though their
addresses are not taken; the advantage in compile-time
checking isn’t worth the loss of readability.
* modules/mod-test/mod-test.c (Fmod_test_non_local_exit_funcall)
(Fmod_test_vector_fill, Fmod_test_vector_eq):
* src/emacs-module.c (MODULE_SETJMP_1)
(module_make_global_ref, module_free_global_ref)
(module_non_local_exit_get, module_make_function)
(module_extract_integer, module_extract_float)
(module_get_user_ptr, module_set_user_ptr)
(module_get_user_finalizer, module_set_user_finalizer)
(module_vec_get, Fmodule_call)
(module_non_local_exit_signal_1)
(module_non_local_exit_throw_1, lisp_to_value)
(finalize_storage, allocate_emacs_value, mark_modules)
(module_handle_signal, module_handle_throw)
(module_format_fun_env):
* src/eval.c (push_handler, push_handler_nosignal)
(init_handler):
* src/lread.c (suffix_p):
Omit unnecessary ‘const’.
* modules/mod-test/mod-test.c (sum, Fmod_test_sum):
* src/emacs-module.c (module_extract_integer)
(module_make_integer):
* src/emacs-module.h (struct emacs_env_25):
Prefer intmax_t to int64_t. This doesn’t change the generated
code on any of the machines Emacs currently ports to, but it’s
at least in theory more future-proof as C99 doesn’t guarantee
that int64_t exists.
* src/dynlib.c (dynlib_close): #ifdef out for now, as it’s not used.
* src/eval.c, src/lisp.h (lisp_eval_depth): Now static.
* src/module.c (Fmodule_load): Fix pointer signedness bug.
(Fmodule_call): Tell GCC that the default case is unreachable.
This is mostly indenting and spacing changes. Also, remove
some unnecessary static decls instead of bothering to reindent them.
* src/module.h (EMACS_EXTERN_C_BEGIN): Remove, and do this inline,
as most other Emacs files do for this sort of thing.
