Re: undo bug?

[Stefan Monnier <monnier@iro.umontreal.ca>]

>>> Couldn't we do away with undo-boundaries and always undo until the next
>>> recorded position of `point'?
>> In theory we could, but that may break various packages: of all the
>> supported undo records, the "undo-boundary" is probably the only one
>> that is used by external packages ;-(

> What I had in mind was that `undo-boundary' would simply `record-point'
> instead of inserting nil.

Yes, that's indeed what I had understood from your earlier message and
what I was responding to.  You're remove the undo boundaries (i.e. the
nil entries) and replace them by point-record entries.  That would work
fine, except that there are packages out there that look for the
nil entries.

>> Instead the current code tries to only insert a point-record when
>> it's actually necessary.

> Sometimes.  For example I use the patch below to handle the (pretty
> annoying) problem that when I redo an earlier insertion `point' ends up
> _before_ the inserted text (my `undo' binds `undo-is-redo' to `equiv'
> when calling `undo-more').

Have you tried to analyze the source of the problem?  Maybe there are
cases where we incorrectly decide not to put a point-record even tho it
is needed.


        Stefan