Changesets feature complete

[Aaron Bentley]

Michael Ellerman wrote:
> On 5/20/06, Aaron Bentley <aaron.bentley at utoronto.ca> wrote:
> 
>>Hi all,
>>
>>Looks like changesets are feature-complete.
> 
> 
> Nice. Now that you've done all that hard work I'm going to proceed to
> hit you with bugs and niggles ;)

Thanks.  That's exactly what I was looking for.

> If you run 'bzr cset' without redirecting you get the progress bars
> interferring with the changeset output. I imagine it's just a matter
> of buffering the cset output until the pbs are finished?

Hadn't thought of that approach.  But yes, I'll do something about 
progress bars.

> 
> # bzr cset -r 0..-1
> just gives me the last revision as a changeset, I would have expected
> the entire history?
> 
> # bzr cset -r 0..
> bzr: ERROR: Invalid revision-id {None} in
> KnitRepository(u'/home/michael/src/bzr/repo/.bzr/')

Okay, I'll fix those.

> 
> # bzr cset -r 1..-1 > patch
> 
> I was expecting this to be equivalent to the concatenation of all the
> individual changesets between 1 and -1. But it's not quite. The last
> changeset seems to contain stuff from way back in the past, which is
> confusing because it means the message doesn't accord with the diff.
> I'm not sure if that's the expected behaviour or not.

This is by design.  'Changeset' is a really bad word to describe what 
these things have become-- they're more like a collection of revisions 
in a textual format.

When you generate a changeset, you get all the ancestors of the target 
revision that aren't ancestors of the base revision.  If we didn't do 
that, then installing revisions from changesets would introduce ghosts.

So it's correct that the last diff would be from 1..2.  As for the 
message, the message of revision 2 is included in a header section above 
the diff, so there's no disparity between the message and the diff.

Aaron