Fix gaps in hashing coverage in the middle and end of even fairly short
strings. E.g., `outline-1`, `outline-2` etc all hashed to the exact
same value but with the patch, there are no collisions among the ~160000
symbols in the Emacs tree.
This change also improves average hashing speed by using fewer mixing
operations.
* src/fns.c (hash_string):
Use unit stride for fairly short strings, while retaining the cap of 8
samples for long ones.
Always hash the last word to ensure that the end of the string is
covered. For strings shorter than a word, use fewer loads and a single
reduction step.
* src/lisp.h (XSYMBOL_WITH_POS_SYM, XSYMBOL): Help the compiler by using
eassume instead of eassert for XSYMBOL postconditions likely to be
useful for optimization later. With gcc 13.2 -O2 x86-64 this improved
speed on my usual “compile all .el files” benchmark by 0.7% and shrank
the text size of Emacs by 0.09%.
* src/fns.c (internal_equal): Turn comment into eassert
that !symbols_with_pos_enabled.
(sxhash_obj): Simplify case of symbol with pos (when enabled).
* src/lisp.h (XSYMBOL_WITH_POS_SYM, XSYMBOL_WITH_POS_POS)
(maybe_remove_pos_from_symbol): New inline functions.
(SYMBOL_WITH_POS_SYM, SYMBOL_WITH_POS_POS): Remove.
All uses replaced by the new functions. This avoids some
double-checking in the source code, simplifies the code overall,
and avoids the need for "Type checking is done in the following
macro" comments to explain unusual code.
When configured with --enable-checking and compiled with gcc -O0,
these macros evaluated arguments multiple times, which made it too
easy to mistakenly write code that behaves differently when debugging.
This patch does not affect performance in normal builds.
In --enable-checking builds with gcc -O0 it slows down my usual
benchmark (remove all '*.elc’ files and then 'make') by 4.4%.
I hope that’s good enough; if not I can complicate the macros to
tune better for debugging builds.
* src/lisp.h (lisp_h_SET_SYMBOL_VAL, lisp_h_SYMBOL_VAL)
(lisp_h_XCONS): Remove, moving each definiens to the corresponding
inline function. All uses removed.
* src/lisp.h (lisp_h_PSEUDOVECTORP, lisp_h_EQ, lisp_h_SYMBOLP):
Refactor by removing these macros, moving each definiens to its only
use. Now that we have symbols with position so that there is no longer
a non-lisp_h_* macro counterpart if DEFINE_KEY_OPS_AS_MACROS, there’s
no need to separate these definiens from their inline function bodies.
* src/lisp.h (lisp_h_BASE2_EQ, BASE2_EQ): Remove. All uses
removed. BASE2_EQ was present only for minor optimization and
with current gcc -O2, BASE2_EQ does not affect performance, so
it’s not worth the hassle.
* src/lisp.h (lisp_h_BASE2_EQ, lisp_h_EQ): Simplify and refactor.
On x86-64 with GCC 3.2 this shrinks temacs text by 0.055% and
after removing all *.elc files speeds up 'make' by 1.0%.
* src/lisp.h (lisp_h_BASE2_EQ, lisp_h_EQ):
Simplify by testing symbols_with_pos_enabled first.
On x86-64 with GCC 13.2 this shrinks temacs text by 1.5%
and after removing all *.elc files speeds up 'make' by 1.2%.
And choose just one binding to display rather than two together.
(https://lists.gnu.org/archive/html/emacs-devel/2024-02/msg00311.html)
* lisp/help.el (help--append-keystrokes-help): New function.
* src/keyboard.c (syms_of_keyboard): Add a symbol for it.
(echo_dash): Use them here.
* src/xdisp.c (display_min_width): Don't return without doing
anything when called from the move_it_* functions. This is needed
to have functions that simulate display layout handle the
min-width display property correctly. (Bug#68374)
* src/androidterm.c (android_note_mouse_movement):
* src/pgtkterm.c (note_mouse_movement):
* src/xdisp.c (get_glyph_string_clip_rects, remember_mouse_glyph)
(expose_area, expose_window, gui_intersect_rectangles): Cast
width or height fields in Emacs_Rectangles to int before summing
with or subtracting them from their coordinate fields, as they
are unsigned outside X, and the sign of the coordinates is thus
not preserved.
Problem reported by Collin Funk (Bug#68996).
* GNUmakefile (.): New macro.
(help): Use ‘$.’ instead of ‘$ ’.
* cross/verbose.mk.android, src/verbose.mk.in (.): New macro.
(AM_V_AR, AM_V_CC, AM_V_CXX, AM_V_CCLD, AM_V_CXXLD, AM_V_GEN):
Use ‘$.’ instead of ‘$ ’.
* lib-src/Makefile.in (install): Use ‘$.’ instead of ‘$ ’.
Commit d005e685e1 should have used
assq_no_signal, but didn't, this commit fixes that.
* src/treesit.c (treesit_traverse_get_predicate): Replace assq_no_quit
with assq_no_signal.
Copyright-paperwork-exempt: yes
* src/fns.c (do_yes_or_no_p, Fyes_or_no_p): Use symbol globals
rather than intern.
(syms_of_fns) <Qyes_or_no_p, Qy_or_n_p>: New symbols.
* src/lread.c (readevalloop): Use symbol global.
(syms_of_lread) <Qinternal_macroexpand_for_load>: New symbol.
This makes both lookups and rehashing cheaper. The index vector size
is now always a power of 2. The first table size is reduced to
6 (from 8), because index vectors would become excessively big
otherwise.
* src/lisp.h (struct Lisp_Hash_Table): Replace index_size with
index_bits. All references adapted.
(hash_table_index_size): New accessor; use it where applicable.
* src/fns.c (hash_index_size): Replace with...
(compute_hash_index_bits): ...this new function, returning the log2 of the
index size. All callers adapted.
(hash_index_index): Knuth multiplicative hashing instead of remainder.
(maybe_resize_hash_table): Reduce first table size from 8 to 6.
Much like the NS port, only the main thread receives input from
the user interface, which is fortunately not a major problem for
packages such as lsp-mode that create Lisp threads.
* configure.ac: Enable with_threads under Android.
* src/android.c (android_init_events): Set `main_thread_id' to
the ID of the main thread.
(setEmacsParams): Set new global variable `android_jvm' to the
JVM object, for the purpose of attaching Lisp threads to the
JVM.
(android_select): [THREADS_ENABLED]: If the caller isn't the
main thread, resort to pselect. Don't check query before select
returns.
(android_check_query): Export.
* src/android.h (_ANDROID_H_): Define new macro and update
prototypes.
* src/process.c (android_select_wrapper): New function.
(wait_reading_process_output): If THREADS_ENABLED, call
thread_select through the Android select wrapper.
* src/thread.c (post_acquire_global_lock): Call
android_check_query; replace android_java_env with the incoming
Lisp thread's.
(run_thread): Attach and detach the thread created to the JVM.
(init_threads): Set the main thread's JNI environment object.
* src/thread.h (struct thread_state) <java_env>: New field.
* java/org/gnu/emacs/EmacsService.java (detectKeyboard): New
function.
* lisp/subr.el (use-dialog-box-p): Don't always return t if a
keyboard is present on Android.
* src/android.c (android_init_emacs_service): Link to new
function.
(android_detect_keyboard): New function.
* src/android.h: Update prototypes.
* src/androidfns.c (Fandroid_detect_keyboard)
(syms_of_androidfns): New function.
Use `ITREE_FOREACH` instead of `overlays_in` if that can save us from
allocating an array.
* src/buffer.c (overlays_in): Mark as static.
(mouse_face_overlay_overlaps): Use `ITREE_FOREACH` instead of `overlays_in`.
(disable_line_numbers_overlay_at_eob): Same, and also change return
value to a boolean.
* src/buffer.h (overlays_in): Don't declare.
* src/editfns.c (overlays_around): Delete function.
(Fget_pos_property): Use `ITREE_FOREACH` and keep the "best so far"
instead of using `overlays_in` and sorting the elements.
* src/lisp.h (disable_line_numbers_overlay_at_eob): Change return
type to a boolean.
* src/xdisp.c (should_produce_line_number): Adjust accordingly.
The assertion failure was raised at lread.c:411 during the
`lread-invalid-bytecodes` test in `test/src/lread-tests.el`.
I suspect we could remove the assertion instead.
* src/xdisp.c (BASE_LINE_NUMBER_VALID_P): New macro.
(try_scrolling): Use it.
(redisplay_window, Fformat_mode_line): Use it to flush the
base_line_number (if it's stale) once at the beginning.
(decode_mode_spec): Don't use (or set) `w->start` and
`w->base_line_number` when operating on another buffer!
* src/xdisp.c (handle_invisible_prop): Skip invisible text
correctly when it starts at position whose resolved bidi level is
above the base paragraph level. (Bug#68446)
We used to print dynamic docstrings "manually" for two reasons:
- References should look like `(#$ . POS)` but `prin1` was unable
to print just `#$` for an sexp.
- `make-docfile` needed to find those docstrings and the object
to which they belonged.
The second point is moot now that we don't use `make-docfile` on
`.elc` files. So this patch lifts the first restriction,
using `print-number-table`.
The rest of the patch then simplifies and regularises the
bytecompiler's generation of dynamic docstrings, which can
now also easily be done for "inner" defvars and other places.
* src/print.c (print_preprocess, print_object): Handle strings in
`print-number-table`.
(Vprint_number_table): Improve docstring.
* lisp/emacs-lisp/bytecomp.el:
(byte-compile--list-with-n): New function.
(byte-compile--docstring-style-warn): Rename from
`byte-compile-docstring-style-warn` and change calling convention.
(byte-compile--\#$, byte-compile--docstrings): New vars.
(byte-compile-close-variables): Bind them.
(byte-compile--docstring): New function.
(byte-compile-from-buffer): Set `byte-compile--\#$`.
(byte-compile-output-file-form): Use `byte-compile--\#$` instead
of special casing specific forms.
(byte-compile--output-docform-recurse, byte-compile-output-docform):
Delete functions.
(byte-compile-file-form-autoload, byte-compile-file-form-defalias)
(byte-compile-file-form-defvar-function, byte-compile-lambda):
Use `byte-compile--docstring` and `byte-compile--list-with-n`.
(byte-compile--declare-var): Add optional `not-toplevel` arg.
(byte-compile-defvar): Add `toplevel` arg. Use `byte-compile--docstring`.
(byte-compile-file-form-defvar): Delegate to `byte-compile-defvar`.
(byte-compile--custom-declare-face): New function. Use it for
`custom-declare-face`.
(byte-compile-file-form-defmumble): Use `byte-compile-output-file-form`
* src/doc.c (Fdocumentation_stringp): New function.
(syms_of_doc): Defsubr it.
(store_function_docstring): Remove left-over code from when we
used DOC for the docstring of some Lisp files.
* lisp/cus-face.el (custom-declare-face): Accept dynamic docstrings.
* lisp/faces.el (face-documentation): Handle dynamic docstrings.
* lisp/help-fns.el (describe-face): Simplify accordingly.
