On 13/12/14 22:37, Craig Ringer wrote:
> On 12/12/2014 06:02 AM, Josh Berkus wrote:
>>
>> Speaking as the originator of commitfests, they were *always* intended
>> to be a temporary measure, a step on the way to something else like
>> continuous integration.
>
> I'd really like to see the project revisit some of the underlying
> assumptions that're being made in this discussion:
>
> - Patches must be email attachments to a mailing list
>
> - Changes must be committed by applying a diff
>
> ... and take a look at some of the options a git-based workflow might
> offer, especially in combination with some of the tools out there that
> help track working branches, run CI, etc.
>
> Having grown used to push/pull workflows with CI integration I find the
> PostgreSQL patch workflow very frustrating, especially for larger
> patches. It's particularly annoying to see a patch series squashed into
> a monster patch whenever it changes hands or gets rebased, because it's
> being handed around as a great honking diff not a git working branch.
>
> Is it time to stop using git like CVS?
>
> (/hides)
>
Having dealt with other projects that use a more git centric + CI
approach, I can say that trying to do reviews can be just frustrating in
that case too:
- quirky and annoying web interfaces
- changesets "expiring" in the middle of you reviewing them
- pulls and rebases making actually making it harder to see what was
changed in new changeset versions
I think the basic issue is that reviewing is hard, and while our
system-ismation of the workflow is really primitive, and could be much
better (that seems to be being worked on), the *tool* is not really the
problem.
regards
Mark