Re: Calling Lisp from undo.c's record_* functions

[phillip.lord@russet.org.uk (Phillip Lord)]

Stefan Monnier <monnier@IRO.UMontreal.CA> writes:

>> As far as I can tell, changing a text property will result in changes to
>> the undo list, but doesn't call prepare_to_modify_buffer. Rather
>> modify_text_properties appears to call prepare_to_modify_buffer_1
>> directly. Bit surprising -- I wouldn't have expected a function called
>> "blah_1" to be called directly from anywhere other than "blah".
>
> Ah, so maybe the call to run_undoable_change should be in
> prepare_to_modify_buffer_1 rather than in prepare_to_modify_buffer.
> I haven't actually looked at the code.


Yeah, that's on the other branch

fix/segfault-undoable-change-prepare-for-buffer

I think I have managed to replicate this error now -- not unit testable,
but forcing a GC in undo-auto--undoable-change like so:

(defun undo-auto--undoable-change ()
  "Called after every undoable buffer change."
  (garbage-collect)
  (add-to-list 'undo-auto--undoably-changed-buffers (current-buffer))
  (undo-auto--boundary-ensure-timer))

then running make check seems to segfault pretty reliably, usually
during bootstrap. fix/segfault-undoable-change-prepare-for-buffer seems
to behave rather better (even with the forced gc).

This seems like the better solution to me.

Phil