This results in a healthy reduction of .elc file size (> 300 KiB saved)
as well as less memory use.
* lisp/emacs-lisp/bytecomp.el (byte-compile-get-constant):
Use 'equal-including-properties' as equivalence relation for all
constants, not just strings.
* etc/NEWS: Mention it. Strictly speaking it's not an incompatible
change because it doesn't break any earlier promises, but for people
having code that made incorrect assumptions, this provides a useful
explanation.
* lisp/emacs-lisp/bytecomp.el (byte-compile-get-constant):
No reason not to use 'assoc' to search for strings as well.
Previously we kept searching even after having found the string!
Calls to these functions were previously rewritten in terms of other
functions both in the optimiser and during codegen, for no good reason.
This also resulted in poor diagnostics: wrong-arity calls to 'atom' and
'nlistp' produced doubled but slightly-different warnings, and no
warnings at all for '/='.
Using compiler macros fixes the problems. The generated code is the same.
* lisp/emacs-lisp/bytecomp.el (byte-compile-negated)
(byte-compile-negation-optimizer): Replace with...
(bytecomp--define-negated): ...this compiler macro defining macro.
* lisp/emacs-lisp/bytecomp.el (byte-compile-cmp):
Don't assume that N-ary comparisons have been normalised, which is done
in the optimiser. Reported by Pip Cet.
* lisp/emacs-lisp/bytecomp.el (byte-compile--check-arity-bytecode)
(byte-compile-warn-about-unresolved-functions, byte-compile-form)
(byte-compile-free-vars-warn):
Delay checking whether a warning is enabled until we are about to emit
such a warning, since most warnings are enabled most of the time but
finding something to warn about is rare.
* lisp/emacs-lisp/byte-opt.el (byte-opt--nary-comparison)
(byte-optimize-constant-args, byte-optimize-member)
(byte-optimize-append):
* lisp/emacs-lisp/bytecomp.el (byte-compile-warnings)
(byte-compile-out-toplevel):
* lisp/emacs-lisp/cconv.el (cconv-convert):
Replace various hand-written 'any' and 'all' expressions with
calls to the new functions, for readability and speed.
Implement a better fix for bug#79634.
* lisp/emacs-lisp/byte-run.el (eval-and-compile):
* lisp/emacs-lisp/bytecomp.el (byte-compile-initial-macro-environment)
<eval-and-compile>: Preserve the surrounding lexical context (the part
available during macroexpansion, i.e. which vars are dynbound).
* lisp/emacs-lisp/rx.el (<pcase> rx): Remove workaround.
* test/lisp/emacs-lisp/macroexp-tests.el
(macroexp--dynbound-eval-and-compile): New test.
The "Macro %s unrecognized, won't work in file" message
happened only for macros defined to `nil`, and the code handling
"unrecognized" code was split with the function part is
`byte-compile-file-form-defmumble` and the macro part in
`byte-compile-file-form-defalias`, so straighten it out
and simplify a bit.
* lisp/emacs-lisp/bytecomp.el (byte-compile-file-form-defmumble):
Don't modify `byte-compile-*-environment` here.
(byte-compile-file-form-defalias): Do it here instead.
Use `macroexp-quote`. Merge the two "unrecognized macro" warnings.
* lisp/emacs-lisp/bytecomp.el (byte-compile-file-form-defalias):
Turn messages into warnings to make them more visible to the user and
provide a source location (bug#78792).
* lisp/emacs-lisp/bytecomp.el (byte-compile-file-form-defalias):
Quote the function name in '(macro . function-name), since we eval
it later.
* test/lisp/emacs-lisp/bytecomp-tests.el
(test-eager-load-macro-expand-defalias): New test.
(Bug#78792)
This fixes a regression from Emacs 29, and is the second attempt after
the later reverted 8b0f5b0597.
* lisp/emacs-lisp/bytecomp.el (byte-compile-fdefinition):
Make it work for functions that aren't compiled.
* test/lisp/emacs-lisp/bytecomp-tests.el (bytecomp-tests--f):
(bytecomp-tests--warn-arity-noncompiled-callee): Add test.
This is a regression from Emacs 29.
* lisp/emacs-lisp/bytecomp.el (byte-compile-fdefinition):
Make it work for functions that aren't compiled.
* test/lisp/emacs-lisp/bytecomp-tests.el (bytecomp-tests--f):
(bytecomp-tests--warn-arity-noncompiled-callee): Add test.
* lisp/emacs-lisp/bytecomp.el (byte-compile--reify-function): Hoist
subexpression out of `if`.
(byte-compile-variadic-numeric, byte-compile--cond-vars)
(byte-compile--cond-switch-prefix, byte-compile-file-form-defalias):
Obey `lexical-binding` when evaluating the code we're compiling.
(byte-compile--maybe-guarded): New function, extracted from
`byte-compile-maybe-guarded`.
(byte-compile-maybe-guarded): Use it so we can edebug the code.
This user option was introduced in 2017, with the rationale:
"This is a workaround for when `byte-compile-cond-jump-table'
accidentally generates wrong code (hasn't happened so far in my
tests), and should be removed once we're sure there are no issues
with it." https://lists.gnu.org/r/emacs-devel/2017-02/msg00223.html
There hasn't been any bugs in this area within the last 6 years, and the
few we did see before that were quickly resolved. Let's declare this
variable obsolete now, while keeping its behavior when set to nil. We
can always revert the obsoletion if it turns out to be needed after all.
* lisp/emacs-lisp/bytecomp.el
(byte-compile-cond-use-jump-table): Declare obsolete.
The docstrings of cl-nintersection and cl-nset-difference have been
inconsistent with their manual entries since the beginning of
emacs.git history (bug#76017). This patch settles on the weaker and
thus backward-compatible requirement that only their first argument
be safe to mutate.
* lisp/emacs-lisp/bytecomp.el: Include only first argument in
mutates-arguments property.
* lisp/emacs-lisp/cl-seq.el (cl-nintersection, cl-nset-difference):
Make docstring consistent with manual in that the second argument is
not modified.
* test/lisp/emacs-lisp/cl-seq-tests.el (cl-nintersection-test)
(cl-nset-difference-test): Simplify.
(cl-nset-difference): Pass fresh list as second argument, otherwise
destructive modifications to it could go undetected.
* lisp/emacs-lisp/bytecomp.el (byte-compile-cond-jump-table): Don't
request our hash tables be purecopied. Adjust comment.
* lisp/progmodes/elisp-mode.el (elisp--local-variables-completion-table):
Use 'defconst' rather than 'defvar' now the purespace problem is gone
* lisp/rfn-eshadow.el (file-name-shadow-properties): Remove obsolete
comment.
This allows the byte compiler to give correct positions, those
of the invoking forms, when an error or warning is caused by
the innards of the invoked macros.
This fixes bug#73725 and bug#73746.
* lisp/emacs-lisp/macroexp.el (sub-macroexp--posify-form)
(macroexp--posify-form): New functions.
(macroexp-preserve-posification): New macro.
(macroexp-macroexpand): Use the new macro to preserve a calling
form's position.
(macroexpand-all): Add a new parameter `keep-pos' which
instructs the function to preserve an invoking form's position.
* lisp/emacs-lisp/byte-opt.el (byte-optimize-form): Use the new
macro to preserve source positions.
* lisp/emacs-lisp/bytecomp.el (byte-compile-preprocess): Call
macroexpand-all with the new argument `keep-pos' non-nil.
