GitLab

* vc/vc.el and all backends: API simplification; could-register
is no longer a public method.  (vc-cvs.el still has a private
implementation.)

* vc/vc.el, and all backends: API cleanup; the backend diff method
takes an explicit async flag.  This eliminates a particularly ugly
global.

The main change is to only use the `category' text-property only when
available.  For that many calls are changed to use c-get-char-property,
c-next-single-property-change, c-sc-scan-lists,
c-sc-parse-partial-sexp, c-unmark-<->-as-paren.

* lisp/progmodes/cc-align.el (c-lineup-respect-col-0): New function.

* lisp/progmodes/cc-cmds.el (c-forward-into-nomenclature)
(c-backward-into-nomenclature): Use cc-subword if subword-mode is
not available.
(c-beginning-of-defun, c-end-of-defun, c-mark-function)
(c-indent-line-or-region): Use c-region-is-active-p.

* lisp/progmodes/cc-defs.el (c-version): Bump up to 5.33.
(c-use-category): New const.
(c-next-single-property-change): New macro.
(c-region-is-active-p): Prefer region-active-p when available.
(c-search-backward-char-property): Fix old min/max typo; probably
a copy/paste error.
(c-mark-<-as-paren, c-mark->-as-paren, c-unmark-<->-as-paren):
Turn them into macros that obey c-use-category.
(c-sc-scan-lists-no-category+1+1, c-sc-scan-lists-no-category+1-1)
(c-sc-scan-lists-no-category-1+1, c-sc-scan-lists-no-category-1-1)
(c-sc-scan-lists, c-sc-parse-partial-sexp)
(c-looking-at-non-alphnumspace): New macros.
(c-sc-parse-partial-sexp-no-category): New function.
(c-emacs-features): Add `category-properties' element.

* lisp/progmodes/cc-engine.el (c-invalidate-state-cache, c-parse-state):
Handle the case where categories are not available.
(c-record-parse-state-state, c-replay-parse-state-state):
Handle marker values.
(c-before-change-check-<>-operators): Look for the `syntax-table'
property rather than for the corresponding `category'.
(c-looking-at-decl-block): Remove unused var
`c-disallow-comma-in-<>-arglists'.
(c-forward-<>-arglist-recur): Remove unused var
`orig-record-found-types'.

* lisp/progmodes/cc-langs.el (c-modified-constant): New lang var.
(c-known-type-key): Don't make a list just to throw it away.

* lisp/progmodes/cc-bytecomp.el (cc-bytecomp-unbound-variables)
(cc-bytecomp-original-functions, cc-bytecomp-original-properties)
(cc-bytecomp-loaded-files): Re-set each time the file is loaded.
(cc-bytecomp-obsolete-var, cc-bytecomp-ignore-obsolete)
(cc-bytecomp-obsolete-fun): Delete unused functions.

* lisp/progmodes/cc-mode.el (c-just-done-before-change): New var.
(c-basic-common-init): Initialize it.
(c-common-init): Only use mode-require-final-newline when available.
(c-before-change): Check and set c-just-done-before-change.
(c-after-change): Re-set c-just-done-before-change.
(c-advise-fl-for-region): New macro.
(lazy-lock-defer-rest-after-change, lazy-lock-defer-line-after-change)
(font-lock-after-change-function, jit-lock-after-change):
Advise if needed.

* net/shr.el (shr-dom-print): Fix up `shr-dom-print' after the
dom.el changes.

* lisp/vc/vc.el (vc-find-conflicted-file): Look for conflicted files in the
current "project" rather than just the current directory.
* lisp/vc/vc-git.el (vc-git-conflicted-files): Clarify in which directory
the file names make sense.

* lisp/vc/smerge-mode.el (smerge-swap): New command.

* lisp/net/newst-backend.el (newsticker--get-news-by-url-callback):
Pass correct status to `newsticker--sentinel-work'.
(newsticker--sentinel-work): Use "newsticker--download-error" as
guid in order to prevent multiple "Could not download..."
messages.  Fixes bug#19166.

* lisp/net/newst-treeview.el (newsticker--treeview-item-show): Check
window liveliness before measuring its width.

Fixes: debbugs:19225

* net/eww.el (eww-render): Call `eww-after-render-hook' in the
correct buffer.

* vc/vc.el, vc/vc-cvs.el, vc/vc-rcs.el, vc/vc-svn.el: The 'merge'
backend method of RCS/CVS/SVN is now 'merge-file', to contrast with
'merge-branch'. Prompting for merge revisions is pushed down to the
back ends; this fixes a layering violation that caused bad behavior
with SVN.

* net/eww.el (eww): Leave point in a place that doesn't cause
scrolling when displaying "Loading...".

* vc/vc.el, vc-hooks.el, and all backends: API simplification;
vc-stay-local-p and repository-hostname are no longer public
methods. Only the CVS and SVN backends used these, and the SVN
support was conditioned out because svn status -v is too slow.
The CVS back end retaiin this machibery and the vc-stay-local
configuration variable now only affects it.

...status -v is so slow that it's unusable.

* lisp/emacs-lisp/inline.el: New file.

* vc/vc.el, vc-hooks.el, and all backends: API simplification;
vc-state-heuristic is no longer a public method, having been removed
where it is redundant, unnecessary, or known buggy. This eliminated
all backends except CVS.  Eliminates bug#7850.

* vc/vc-cvs.el, vc/vc-hooks.el, vc/vc-rcs.el, vc/vc-sccs.el: Eliminate
vc-mistrust-permissions.  It was only relevant to the RCS and SCCS
back ends and defaulted to t. Code now always mistrusts permissions -
by actual measurement the effect on performance is negligible. As a
side effect bug#11490 is now irrelevant.

* vc/vc.el, vc-hooks.el, and all backends: API simplification;
vc-workfile-unchanged-p is no longer a public method (but the RCS and
SCCS back ends retain it as a private method used in state
computation). This method was redundant with vc-state and usually
implemented as a trivial call to same. Fixes the failure mode
described in bug#694.

This hasn't made any sense since RCS, and was a dumb stunt then.

* vc/vc.el and all backends: API simplification; init-revision is
gone, and vc-registered functions no longer take an initial-revision
argument.

These changes suffice for temacs on x86-64 with GCC 4.9.2 and
-fsanitize=address.
* alloc.c (valid_pointer_p) [ADDRESS_SANITIZER]:
Return -1 or 0, as the pipe trick doesn't work.
* alloc.c (relocatable_string_data_p, mark_object, sweep_symbols):
* data.c (Ffset):
* print.c (print_object):
When a pointer-check primitive returns -1, do not assume this
means the pointer is valid or that the underlying system has failed.
It could just be that addresses are being sanitized so Emacs can't
test for pointer validity.
* lisp.h (defined_GC_CHECK_STRING_BYTES): New constant.
(USE_STACK_STRING) [GC_CHECK_STRING_BYTES]: Now false, since the
string validity checker doesn't work on stack-based strings.