Thoughts on push performance

[Andrew Bennetts]

John Arbash Meinel wrote:
[...]
> ~  b) Further, if you look at update_revisions we have:
[...]
> ~                if overwrite:
> ~                    self.generate_revision_history(stop_revision)
> ~                else:
> ~                    self.generate_revision_history(stop_revision,
> ~                        last_rev=last_rev, other_branch=other)
>
> ~     So I *think* what is happening is that if you supply --overwrite it might
> ~     be generating the revision history 2 times.

Btw, _basic_push strangely does *not* pass the overwrite to update_revisions.
Instead it catches the DivergedBranches error itself, if overwrite is set.

This code is in general a bit tangled; depending on combinations of flags
different code is responsible for raising DivergedBranches, and different code
is responsible for catching it.  I don't think it needs to be like that.

Can you see any reason why we shouldn't just do this to _basic_push:

-        try:
-            target.update_revisions(self, stop_revision)
-        except errors.DivergedBranches:
-            if not overwrite:
-                raise
-        if overwrite:
-            target.set_revision_history(self.revision_history())
+        target.update_revisions(self, stop_revision,
+                overwrite=overwrite)

?

I'm sure if this code was easier to follow it'd be easier to optimise...

-Andrew.