Re: [Emacs-diffs] fix/undo-point-in-wrong-place 6b3cfe4 4/4: Prepare for record now separate function.

[Stefan Monnier]

>     PT is the position of point that will naturally occur as a result of the
>     undo record that will be added just after this command terminates.  */

This comment is invalidated by your change.

> +      record_point (beg + SCHARS (string));

Hmm I thought the sign on sbeg took care of this case already (i.e. the
record_point should only record something when point was neither at the
beginning nor at the end of the deleted string).

As for the source of the bug (i.e. what change caused the new behavior):
in the old code, undo-boundary was called right before every command
(whether there was a need to push a boundary or not), so contrary to the
comment in the code, last_boundary_position was actually recording
"position of point at beginning of the command" rather than "position of
point last time we inserted a boundary".

So the hunk below should recover the old behavior (well, more or less:
it wouldn't compile as is, but I hope you get the idea).  But to fix it
right, we should rename these vars and adjust their comment to better
reflect the way they're really used.


        Stefan


diff --git a/src/keyboard.c b/src/keyboard.c
index 849066c..125091e 100644
--- a/src/keyboard.c
+++ b/src/keyboard.c
@@ -1448,6 +1448,8 @@ command_loop_1 (void)
             /* Ensure that we have added appropriate undo-boundaries as a
                result of changes from the last command. */
             call0 (Qundo_auto__add_boundary);
+            last_boundary_position = PT;
+            last_boundary_buffer = current_buffer;
 
             call1 (Qcommand_execute, Vthis_command);