* src/eval.c (funcall_subr): Help the compiler by reducing aliasing
problems, and compensate for a missed-optimisation bug in LLVM where
switches sometimes forget to use variable range information (reported
in https://github.com/llvm/llvm-project/issues/76085).
Reported by OGAWA Hirofumi.
* lisp/emacs-lisp/bytecomp.el (displaying-byte-compile-warnings):
Move most of the innards to...
(bytecomp--displaying-warnings): ...this new function, for ease
of maintenance.
* lisp/emacs-lisp/bytecomp.el (byte-compile-file):
Wrap early warning about missing lexbind declaration in
`displaying-byte-compile-warnings` so that it doesn't cause the
creation of a compile-log buffer with the wrong mode.
* lisp/emacs-lisp/bytecomp.el (byte-compile-toplevel-file-form)
(byte-compile-form):
* lisp/emacs-lisp/macroexp.el (macroexp--expand-all):
Use `macroexp--with-extended-form-stack` instead of explicit
push and pop.
* src/sfnt.c (sfnt_interpret_compound_glyph_2): Don't grow the
compound decomposition context for saving compound points into,
since the space for them already exists.
* src/sfnt.c (sfnt_move): Correct commentary.
(sfnt_interpret_control_value_program): Reset dual projection,
freedom and projection vectors, in addition to the reference
points, zone pointers and loop counter.
Any non-trivial EIEO object in particular is impossible to read in the
*trace-output* buffer without this. Functions, hash-tables, etc now
print as they do in backtrace buffers.
* lisp/emacs-lisp/trace.el (cl-print): Require it
(trace-entry-message, trace-exit-message): Use cl-prin1-to-string
The time stamp format in the chat buffer may now
differ from the format in the log files.
* doc/misc/rcirc.texi: Document new variable
* lisp/net/rcirc.el (rcirc-log-time-format):
Custom variable for the format. (bug#67597)
Copyright-paperwork-exempt: yes
Now with have :noconfirm no need to check for '(set modify) otherwise
we fail as well in kmacros when register-use-preview is t.
The conditions should not be hard coded in
register-read-with-preview-fancy but in the cl-defmethod
register-command-info for each command.
* lisp/register.el (register-read-with-preview-fancy): Remove now
unneeded condition.
Using post-command-hook in minibuffer-setup-hook instead of a timer
allow running exit-minibuffer without delay and ensure the serie of
commands used in a kmacro run synchronously.
* lisp/register.el (register-read-with-preview-fancy): Do it.
It is now the default with a value of register-use-preview eq to
basic.
To change this one have now to customize register-use-preview to
another value.
* lisp/register.el (register-preview-delay): Remove obsolescence.
(register--read-with-preview-function): New.
(register-use-preview): New option basic, it is now the default.
(register-preview-default-1): New the register-preview-default used by
`register-read-with-preview-fancy`.
(register-preview-default): Restored (same as Emacs-29).
(register--preview-function): Generic fn that return the right
function for register--preview-function.
(register-preview): Restored (same behavior as Emacs-29).
(register-preview-1): Used by `register-read-with-preview-fancy'.
(register-read-with-preview-basic): The old
register-read-with-preview.
(register-read-with-preview-fancy): The new
register-read-with-preview.
This happen when register-use-preview is nil or never.
This reproduce what we had previously in 29.1 but with filtering in
the preview and default registers are provided for the commands of
type 'set'.
This is implemented with cl-defmethod to keep the code as much as
possible configurable.
* lisp/register.el (register-preview-info): New slot.
(register-command-info): Add new methods for copy-to-register,
point-to-register, number-to-register,
window-configuration-to-register, frameset-to-register and
copy-rectangle-to-register.
(register-read-with-preview): Bind noconfirm.
* lisp/progmodes/project.el (project--read-file-cpd-relative):
Don't abbreviate at all, only suffixes are shown anyway.
And expand-file-name is slightly faster.
* lisp/progmodes/project.el (project--transplant-file-name):
Don't abbreviate here (bug#67901, bug#63829).
(project--read-file-cpd-relative): Do it here instead. The reader
functions should decide on the preferred format themselves.
* lisp/vc/vc.el (log-view-vc-prev-revision)
(log-view-vc-prev-fileset): New dynamic variables (bug#55871).
(vc-print-log-renamed-add-button): Extract from
'vc-print-log-setup-buttons'. Bind the above variables to convey
the current revision and fileset to 'vc-print-log-internal'.
(vc-print-log-internal): Use it also here, to print a "back"
button above the log when 'log-view-vc-prev-fileset' is set.
`math-read-preprocess-string` is one of the bottlenecks of `calc-eval`
and was unnecessarily slow even with no substitutions made.
This affected org-mode in particular, where `calc-eval` is called
repeatedly to recalculate tables.
Reported by Raffael Stocker who also wrote the unit tests here.
* lisp/calc/calc-aent.el (math--read-preprocess-re-cache): New.
(math-read-preprocess-string):
Use math--read-preprocess-re-cache, first computing it if necessary.
* test/lisp/calc/calc-tests.el (calc-math-read-preprocess-string):
New test.
bug#67884
To prevent errors, the same case-fold-search setting used for
originally selecting the candidate should be used when scoring it.
* lisp/minibuffer.el (completion--flex-score): Fix
* src/sfnt.c (sfnt_mul_f26dot6_fixed): Correct rounding
constant.
(sfnt_make_interpreter, SSW, WCVTF, sfnt_compute_phantom_points)
(sfnt_interpret_simple_glyph, sfnt_interpret_compound_glyph_1)
(sfnt_vary_interpreter, sfnt_check_ssw): Account for the
fractional bits in a f26dot6 when computing the interpreter
scale factor.
* src/sfnt.c (sfnt_dot_fix_14): Correct typo in final division.
(sfnt_compute_phantom_points): New parameters S1 and S2, into
which unrounded phantom point coordinates are saved.
(sfnt_interpret_simple_glyph, sfnt_interpret_compound_glyph_2)
(sfnt_interpret_compound_glyph_1): Adjust correspondingly.
* src/sfntfont.c (sfntfont_get_glyph_outline): Delete redundant
branch.
Let the caller tell us clearly where is the base of the backtrace,
if it's not `debug`. This is done by passing a new `:backtrace-base`
keyword argument to `debug`.
Then use this info systematically in all the places where we access
the real C-level backtrace, to try and avoid inconsistencies and brittle
code that tries to enumerate the expected frames we're in.
* src/eval.c (get_backtrace_starting_at): Add support for offsets in the
`base` argument.
(Fbacktrace_debug): Add optional `base` argument.
* lisp/emacs-lisp/debug.el (debug, debugger-frame, debugger-frame-clear):
Use `debugger--backtrace-base` when calling `backtrace-debug`.
(debugger-setup-buffer): Use `debugger--backtrace-base`
when calling `backtrace-get-frames`.
(debugger-frame-number): Drop `skip-base` arg, assume it's never nil.
Add sanity check.
(debugger--backtrace-base): Use the `:backtrace-base` info
in `debugger-args`.
(debugger-eval-expression): Adjust call to `debugger-frame-number`.
(debug--implement-debug-on-entry): Pass appropriate `:backtrace-base`.
* src/sfnt.c (sfnt_deltap): Enable deltap instructions again,
since they now serve a purpose.
(sfnt_compute_phantom_points): Round phantom points to match the
behavior of other TrueType scalers.
