* src/ccl.c (ccl_driver): Defend against signed integer
overflow (Bug#42660). Perhaps some of this is unnecessary,
but it is safe and ccl.c is low-priority these days.
* src/pdumper.c (pdumper_find_object_type_impl): When checking
last_mark_bits, require the offset to be less than
discardable_start, not cold_start. This fixes a typo introduced in
2020-08-14T21:33:21Z!eggert@cs.ucla.edu (Bug#42832).
* lisp/simple.el (read--expression): No longer bind
read--expression-try-read here.
* lisp/simple.el (read-expression-map): Bind
read--expression-try-read here (bug#42893).
This new specialized command was recently added in [1: 4a6dd13fa4].
It reestablishes the bindings every time `read--expression' is
invoked, which is wrong because it makes it impossible for users
to remove these bindings.
1: 4a6dd13fa4
Change 'M-:' to not error out on incomplete expressions.
In Elisp mode, ElDoc will try the variable's docstring first, then the
function signature. If the former doesn't have one, don't hide the
latter.
* lisp/progmodes/elisp-mode.el (elisp-eldoc-var-docstring): If no
symbol docstring, don't declare any.
* lisp/textmodes/flyspell.el
(flyspell-correct-on-mouse-3): New option to bind
'flyspell-correct-word' to 'mouse-3'.
(flyspell--set-correct-on-mouse-3): New function to update option.
(flyspell-mode): Update 'flyspell-mouse-map' if above option is
set.
* doc/emacs/fixit.texi (Spelling): Mention the new option.
* etc/NEWS: Announce the new option.
* lisp/subr.el (user-original-login-name): Mark as obsolete. This
XEmacs compat alias was obsoleted even by XEmacs, so there is no point
in keeping it around.
* lisp/eshell/esh-mode.el: (eshell-output-filter): Match
current-buffer behavior of comint-output-filter (bug#42870).
This change (a) sets the current buffer to the process-buffer when
invoking preoutput filter functions and (b) only invokes them when the
process-buffer is live. Otherwise, the preoutput filter functions be
invoked in whatever buffer happens to be focused, breaking hooks that
read buffer-local variables.
(eudc-bob-generic-keymap, eudc-bob-image-keymap)
(eudc-bob-sound-keymap, eudc-bob-url-keymap, eudc-bob-mail-keymap):
Move initialization into declaration. Use RET rather than `return`.
(eudc-jump-to-event): Delete; use `mouse-set-point` instead.
(eudc-bob-save-object): Rewrite using `write-region`.
(eudc-bob-popup-menu): Use `popup-menu`.
Problem reported by Pip Cet in:
https://lists.gnu.org/r/emacs-devel/2020-08/msg00444.html
* src/fns.c (Fdelete): Fix correctness bug via a simpler (though more
memory-intensive) approach. It’s probably not worth optimizing
the memory usage yere.
* test/src/fns-tests.el (test-vector-delete): Add test for the bug.
* src/fns.c (Fdelete): When deleting from a vector, call Fequal
only once per vector element. This is faster when Fequal is slow,
and avoids the need to preinitialize the vector result. Finish
when the result is exhausted, not when the input is exhausted;
the two are equivalent but the former may be faster.
* test/src/fns-tests.el (test-vector-delete): New test.
Fvector is less error-prone than make_uninit_vector, as it
avoids the possibility of a GC crash due to an uninitialized
vector. So prefer Fvector to make_uninit_vector when this is
easy (and when there's no significant performance difference).
Inspired by a suggestion by Pip Cet in:
https://lists.gnu.org/r/emacs-devel/2020-08/msg00313.html
* src/ccl.c (Fregister_ccl_program):
* src/ccl.c (Fregister_ccl_program):
* src/charset.c (Fdefine_charset_internal):
* src/font.c (Fquery_font, Ffont_info, syms_of_font):
* src/fontset.c (font_def_new, Fset_fontset_font):
* src/ftfont.c (ftfont_shape_by_flt):
* src/hbfont.c (hbfont_shape):
* src/macfont.m (macfont_shape):
* src/search.c (Fnewline_cache_check):
* src/xfaces.c (Fx_family_fonts):
* src/xfns.c (Fx_window_property_attributes):
Prefer Fvector to make_uninit_vector when either is easy.
* src/fontset.c (font_def_new): Now a function with one less
arg instead of a do-while macro, and renamed from FONT_DEF_NEW.
All uses changed.
Avoid problems if GC occurs while initializing a vector.
Problem with Fdelete reported by Pip Cet in:
https://lists.gnu.org/r/emacs-devel/2020-08/msg00313.html
I looked for similar problems elsewhere and found quite a few.
* src/coding.c (make_subsidiaries):
* src/composite.c (syms_of_composite):
* src/font.c (build_style_table, Ffont_get_glyphs):
* src/nsselect.m (clean_local_selection_data):
* src/nsxwidget.m (js_to_lisp):
* src/syntax.c (init_syntax_once):
* src/window.c (Fcurrent_window_configuration):
* src/xselect.c (selection_data_to_lisp_data)
(clean_local_selection_data):
Use make_nil_vector instead of make_uninit_vector.
* src/fns.c (Fdelete):
* src/xwidget.c (webkit_js_to_lisp):
Use allocate_nil_vector instead of allocate_vector.
* src/search.c (Fnewline_cache_check):
Use make_vector instead of make_uninit_vector.
* src/xdisp.c (get_next_display_element): When
nobreak-char-display is t, display NBSP and non-ASCII hyphens as
themselves, not as their ASCII counterparts, just with the
nobreak-space/nobreak-hyphen face. (Bug#42811)
* lisp/ffap.el (ffap-file-name-with-spaces): New variable (bug#8439).
(ffap-search-backward-file-end, ffap-search-forward-file-end)
(ffap-dir-separator-near-point): New functions.
(ffap-string-at-point): Use the variable and the new functions to
guess at files containing strings.
* lisp/erc/erc-match.el (erc-current-nick-highlight-type,
erc-pal-highlight-type, erc-fool-highlight-type,
erc-keyword-highlight-type, erc-dangerous-host-highlight-type): Add
`message' type for highlighting the entire message but not the
sender's nick.
(erc-match-message): Check for the new `message' highlight type and
propertize the message (not including the nick) accordingly.
* etc/NEWS: Announce the addition of the `message' highlight type.
Problem reported by Lars Ingebrigtsen, and problem diagnosis
and most of this patch by Pip Cet (Bug#42832).
* src/pdumper.c (dump_bitsets_init): Rename from dump_bitset_init.
All callers changed. Initialize two bitsets with a single malloc
call.
(struct pdumper_loaded_dump_private): New member last_mark_bits.
(pdumper_find_object_type_impl): Return PDUMPER_NO_OBJECT if
the last_mark_bits’ bit is clear.
(pdumper_set_marked_impl): Assert that the last_mark_bits’
bit is set.
(pdumper_clear_marks_impl): Save mark_bits into
last_mark_bits before clearing mark_bits.
Co-authored-by: Pip Cet <pipcet@gmail.com>
* configure.ac (OPTION_DEFAULT_IFAVAILABLE): New macro. Use it to
define the --with-json option. Add with_json and HAVE_JSON to the
'MISSING' checks (bug#39953).