Package: emacs;
Reported by: Eshel Yaron <me <at> eshelyaron.com>
Date: Tue, 29 Oct 2024 17:28:02 UTC
Severity: normal
Found in version 31.0.50
Done: Eli Zaretskii <eliz <at> gnu.org>
Bug is archived. No further changes may be made.
Message #22 received at 74091 <at> debbugs.gnu.org (full text, mbox):
From: Eli Zaretskii <eliz <at> gnu.org> To: Eshel Yaron <me <at> eshelyaron.com> Cc: 74091 <at> debbugs.gnu.org Subject: Re: bug#74091: 31.0.50; string-pixel-width in mode line disables region Date: Thu, 31 Oct 2024 13:41:02 +0200
> From: Eshel Yaron <me <at> eshelyaron.com>
> Cc: 74091-done <at> debbugs.gnu.org
> Date: Thu, 31 Oct 2024 12:09:20 +0100
>
> Invoking this command twice in a row in subr.el deactivates the region,
> while the same without the argument to kill-all-local-variables keeps
> the region active.
>
> So the problem seems to be in a lower level than string-pixel-width...
Killing local variables makes the global value of deactivate-mark be
in effect when the command loop decides whether to deactivate the
region after a command finishes.
> As I'm sure you know, applying a crude fix without fully understanding
> the problem is likely to hide other subtle bugs that may then be harder
> to investigate.
That's not a crude fix, that's what the ELisp manual tells us to do:
-- Variable: deactivate-mark
If an editor command sets this variable non-‘nil’, then the editor
command loop deactivates the mark after the command returns (if
Transient Mark mode is enabled). All the primitives that change
the buffer set ‘deactivate-mark’, to deactivate the mark when the
command is finished. Setting this variable makes it buffer-local.
To write Lisp code that modifies the buffer without causing
deactivation of the mark at the end of the command, bind
‘deactivate-mark’ to ‘nil’ around the code that does the
modification. For example:
(let (deactivate-mark)
(insert " "))
> >> And in both the old
> >> implementation and in the new one, the modification is in a different
> >> buffer, is that expected to disable the mark in the original buffer?
> >
> > The variable deactivate-mark only becomes buffer-local if set;
> > otherwise the global value will be changed.
>
> Could you perhaps elaborate? I see that running a command that modifies
> a different buffer does not deactivate the region in the current buffer,
> which is basically what I would expect.
You are asking me to elaborate about what? about the local value of
deactivate-mark or about why you see what you see (in a scenario you
haven't described)?
Look, you are welcome to keep debugging this if you are interested. I
invested enough of my time into figuring out why the region was
deactivated by C-n, and the solution I installed satisfies me. But
you are welcome to keep digging, and let me tell you what I found to
save you some non-trivial tinkering:
. The region is deactivated because of this in our command loop:
if (!NILP (Vdeactivate_mark))
/* If `select-active-regions' is non-nil, this call to
`deactivate-mark' also sets the PRIMARY selection. */
call0 (Qdeactivate_mark);
This is consistent with what the ELisp manual says, see the above
citation.
. Deactivate-mark is set non-nil by the low-level subroutines that
modify the buffer, again according to the manual. Two such
buffer-modification calls were present in string-pixel-width: one
in string-pixel-width itself, when it inserts the string into a
work buffer, the other in work-buffer--release where it erases the
work buffer. This happens in in prepare_to_modify_buffer_1:
signal_before_change (start, end, preserve_ptr);
Fset (Qdeactivate_mark, Qt);
. Here is a Lisp-level backtrace from one call which sets
deactivate-mark in your recipe, as collected by GDB:
Lisp Backtrace:
"string-pixel-width" (0x65a8940)
"progn" (0x65a8c40)
"eval" (0x65a8f90)
"posn-at-point" (0xa084200)
"line-move-visual" (0xa084170)
"line-move" (0xa084108)
"next-line" (0x65aea10)
"funcall-interactively" (0x65aea08)
"call-interactively" (0xa084078)
"command-execute" (0x65af798)
As you see, next-line calls posn-at-point, which formats the mode
line (to calculate is height), and that invokes the :eval form.
Here's the C backtrace which captures the details missing from the
above Lisp backtrace:
Thread 1 hit Hardware watchpoint 4: Vdeactivate_mark
Old value = XIL(0)
New value = XIL(0x30)
0x00bd2da5 in store_symval_forwarding (valcontents=..., newval=XIL(0x30),
buf=0x8f4c48) at data.c:1430
1430 *XOBJFWD (valcontents)->objvar = newval;
(gdb) bt
#0 0x00bd2da5 in store_symval_forwarding (valcontents=..., newval=XIL(0x30),
buf=0x8f4c48) at data.c:1430
#1 0x00bd3da3 in set_internal (symbol=XIL(0x6210), newval=XIL(0x30),
where=XIL(0xa0000000008f4c48), bindflag=SET_INTERNAL_SET) at data.c:1759
#2 0x00bd37c5 in Fset (symbol=XIL(0x6210), newval=XIL(0x30)) at data.c:1630
#3 0x00b5b757 in prepare_to_modify_buffer_1 (start=1, end=1, preserve_ptr=0x0)
at insdel.c:2073
#4 0x00b5b784 in prepare_to_modify_buffer (start=1, end=1, preserve_ptr=0x0)
at insdel.c:2083
#5 0x00b57e1b in insert_from_string_1 (string=XIL(0x800000000bc07ad8), pos=0,
pos_byte=0, nchars=3, nbytes=3, inherit=false, before_markers=false)
at insdel.c:1023
#6 0x00b57c15 in insert_from_string (string=XIL(0x800000000bc07ad8), pos=0,
pos_byte=0, length=3, length_byte=3, inherit=false) at insdel.c:974
#7 0x00be5dbf in general_insert_function (insert_func=0xb57249 <insert>,
insert_from_string_func=0xb57b92 <insert_from_string>, inherit=false,
nargs=1, args=0xa084250) at editfns.c:1336
#8 0x00be5e59 in Finsert (nargs=1, args=0xa084250) at editfns.c:1372
#9 0x00c71edc in exec_byte_code (fun=XIL(0xa0000000008f45b0),
args_template=513, nargs=1, args=0x65a8948) at bytecode.c:1417
#10 0x00c019a3 in funcall_lambda (fun=XIL(0xa0000000008f45b0), nargs=1,
arg_vector=0x65a8940) at eval.c:3238
#11 0x00c017be in apply_lambda (fun=XIL(0xa0000000008f45b0),
args=XIL(0xc00000000071c2d0), count=640) at eval.c:3201
#12 0x00bff56b in eval_sub (form=XIL(0xc00000000071c2e0)) at eval.c:2631
#13 0x00bf7cde in Fprogn (body=XIL(0xc00000000071c2b0)) at eval.c:439
#14 0x00bfec17 in eval_sub (form=XIL(0xc00000000071c2f0)) at eval.c:2535
#15 0x00bfe644 in Feval (form=XIL(0xc00000000071c2f0), lexical=XIL(0x30))
at eval.c:2448
#16 0x00c010df in funcall_subr (subr=0x128a1c0 <Seval>, numargs=2,
args=0x65a8f90) at eval.c:3149
#17 0x00c00a02 in funcall_general (fun=XIL(0xa00000000128a1c0), numargs=2,
args=0x65a8f90) at eval.c:3026
#18 0x00c00d92 in Ffuncall (nargs=3, args=0x65a8f88) at eval.c:3079
#19 0x00bfbd5f in internal_condition_case_n (bfun=0xc00c46 <Ffuncall>,
nargs=3, args=0x65a8f88, handlers=XIL(0x30),
hfun=0x9dfc51 <dsafe_eval_handler>) at eval.c:1687
#20 0x009dfd7d in dsafe__call (inhibit_quit=true, f=0xc00c46 <Ffuncall>,
nargs=3, args=0x65a8f88) at xdisp.c:3093
#21 0x009dfef9 in dsafe_eval (sexpr=XIL(0xc00000000071c2f0)) at xdisp.c:3129
#22 0x00a3033e in display_mode_element (it=0x65a93d0, depth=2, field_width=0,
precision=0, elt=XIL(0xc00000000071c300), props=XIL(0), risky=false)
at xdisp.c:28039
#23 0x00a308f2 in display_mode_element (it=0x65a93d0, depth=1, field_width=0,
precision=0, elt=XIL(0xc00000000071c290), props=XIL(0), risky=false)
at xdisp.c:28125
#24 0x00a2e976 in display_mode_line (w=0xbb0b428,
face_id=MODE_LINE_ACTIVE_FACE_ID, format=XIL(0xc00000000071c310))
at xdisp.c:27550
#25 0x009d54b2 in pos_visible_p (w=0xbb0b428, charpos=1, x=0x65adfec,
y=0x65adfe8, rtop=0x65adffc, rbot=0x65adff8, rowh=0x65adff4,
vpos=0x65adff0) at xdisp.c:1732
#26 0x00a65e8e in Fpos_visible_in_window_p (pos=XIL(0),
window=XIL(0xa00000000bb0b428), partially=XIL(0x30)) at window.c:2018
#27 0x00b2810d in Fposn_at_point (pos=XIL(0), window=XIL(0xa00000000bb0b428))
at keyboard.c:12552
Translation: posn-at-point called pos-visible-in-window-p, , which
called pos_visible_p, which called display_mode_line. That
eventually called the :eval form, and inserted the string via
insert_from_string_1, which called prepare_to_modify_buffer_1,
which set deactivate-mark to t.
That's what I saw and what led me to my solution, according to what
the ELisp manual says.
GNU bug tracking system
Copyright (C) 1999 Darren O. Benham,
1997,2003 nCipher Corporation Ltd,
1994-97 Ian Jackson.