* admin/unidata/unidata-gen.el (unidata-check): Do not test special
casing mapping of characters since that mapping is not constructed from
the unidata.txt file.
Also, check for integer decoder and cons char earlier so that less
unnecessary processing is performed.
For larger input files with lots of shared data structures, an
association list is too slow.
* src/lread.c (read_objects_map, read_objects_completed): New
variables, replacing read_objects.
(readevalloop): Initialize them with hash tables before starting a
top-level read, if they're not already empty hash tables, and reset
them to Qnil afterwards if something was added to the hash tables.
(read_internal_start): Likewise.
(read1): Store first the placeholder and later the newly read object
into read_objects_map under the specified object number. If the new
object can contain a reference to itself, store it in
read_objects_completed.
(substitute_objects_recurse): Check read_objects_completed instead of
read_objects for the known possibly-recursive objects.
(syms_of_lread): Update initializations.
* configure.ac: Check for getc_unlocked.
* src/charset.c (read_hex, load_charset_map_from_file): Use
getc_unlocked instead of getc.
(getc_unlocked) [!HAVE_GETC_UNLOCKED]: Fall back to getc.
* src/lread.c (readbyte_from_file, Fget_file_char, read1,
getc_unlocked): Likewise.
* src/lread.c (read1): After reading an object using the "#n=" syntax,
if the read object is a cons cell, instead of recursively substituting
the placeholder with the new object, mutate the placeholder cons cell
itself to have the correct car and cdr values.
Values that don't contain other values cannot be circular, so checking
for circular objects is a waste of cycles.
* src/lread.c (substitute_object_recurse): If the subtree being
examined is a symbol, number, or property-less string, just return
it.
* src/bidi.c (MAX_BPA_STACK): Now a constant, not a macro.
Shrink it to allow for the two struct bidi_it objects in
the same frame.
(PUSH_BPA_STACK): Avoid integer overflow with enormous bidi cache.
(bidi_find_bracket_pairs): Use compile-time check instead of runtime.
Without this patch, ./temacs would dump core sometimes on Fedora
25 x86-64. The problem was that the hybrid allocator assumed that
all pointers into bss_sbrk_buffer are allocated via gmalloc. This
assumption is not true on Fedora, because the standard memory
allocator calls gdefault_morecore, which means its blocks are
interleaved with our blocks. Usually the code happened to work,
because our data structures agreed with the glibc data structures,
but this was merely luck due to a shared pedigree, and as glibc
mutates our luck has run out.
* src/gmalloc.c (ALLOCATED_BEFORE_DUMPING) [HYBRID_MALLOC]:
Remove; no longer needed.
(BLOCK): Use unsigned division, as that does the right thing near zero.
(register_heapinfo, __malloc_internal_nolock, __free_internal_nolock)
(_realloc_internal_nolock):
Big blocks now have type -1, not 0, as 0 now means the block is
not ours.
(morecore_nolock): Omit now-unnecessary casts to size_t.
(allocated_via_gmalloc) [HYBRID_MALLOC]: New function.
(hybrid_free, hybrid_realloc) [HYBRID_MALLOC]: Use it, to
avoid calling the wrong free or realloc function in some cases.
* lisp/gnus/gnus-art.el (article-date-ut):
Work for unfolded multi-line Date header.
(article-transform-date):
Refactor; add header name if it is missing in user-defined date line.
(article-date-user): Fix name of date type.
* lisp/minibuffer.el (completion-pcm--filename-try-filter)
(completion-pcm--all-completions): Use nreverse to undo the reversing
caused by using push in the loop.
Problem reported by Andrés Musetti (Bug#27248).
* src/widget.c (emacsFrameClassRec): Do not initialize superclass here.
(emacsFrameClass): Now a function (which initializes the
superclass) instead of a variable. All uses changed.
* autogen.sh (get_version): Simplify and make more reliable
by using expr rather than echo | sed. Check exit status of program.
Run program in subshell in case it cannot be executed.
(check_version): Check exit status of command rather than its output.
Check return status of get_version.
* lisp/gnus/mm-view.el (mm-display-inline-fontify):
Bind enable-local-variables to nil globally instead of making it
buffer-local; remove let-bind of local-enable-local-variables.
cf. <b4mtw3bbfp5.fsf@jpl.org> in the emacs-devel list.
* autogen.sh (get_version): Check return status of "--version".
(check_version): Try to distinguish between a missing tool
and a broken one. (Bug#27288)
This (and my previous patch) incorporate:
2017-06-17 diffseq: port to GCC 7 with --enable-gcc-warnings
2017-06-15 gettext-h: Update comment
* lib/diffseq.h: Copy from gnulib.
This was part of the original design of the module
API (https://lists.gnu.org/archive/html/emacs-devel/2015-02/msg00960.html),
but I didn't take it into account when adding the should_quit
function.
Instead of duplicating the environment fields or using the C
preprocessor, use configure to build emacs-module.h.
* configure.ac: Expand emacs-module.h template.
Add a new command 'replace-buffer-contents' that uses the Myers diff
algorithm to non-destructively replace the accessible portion of the
current buffer. The Myers algorithm is implemented in Gnulib.
* src/editfns.c (Freplace_buffer_contents): New command.
(set_bit, bit_is_set, buffer_chars_equal): New helper functions.
(syms_of_editfns): Define new command.
* test/src/editfns-tests.el (replace-buffer-contents-1)
(replace-buffer-contents-2): New unit tests.
* src/buffer.h (BUF_FETCH_CHAR_AS_MULTIBYTE): New helper macro.
* admin/merge-gnulib (GNULIB_MODULES): Add diffseq.h and minmax.h.
* lisp/textmodes/css-mode.el (css--complete-property-value): Be less
eager by looking for a colon after the property which values are being
completed for.
* test/lisp/textmodes/css-mode-tests.el (css-test-complete-property):
Add a test case ensuring that properties that are prefixes of other
properties don't hinder further completion.
