Re: Proposal to deploy GitLab on


On Wed, May 17, 2017 at 1:05 PM, Carlos Soriano <csoriano protonmail com> wrote:
Hey Jehan,

Knowing that core contributors like you and GIMP maintainers will have
access to the repo, are the sporadic contributions still many enough enough

Yes we still have regular one-time contributions. If anything, we are
the ones who don't review them fast enough, though we have been
getting better now and try to review external patches in a more timely

for fetching a remote being inconvenient? Is it because it takes
considerably more time to fetch a repo than download and applying a patch?

It does take more time indeed. But the most annoying part is having to
switch branches. When you checkout another branch, autotools gets
confused and will re-build much more than what it should have if I
just applied the patch (actually for good reasons sometimes; for
instance often the branch would be based on older commits than master
HEAD). So you transform a 10-second builds into a 10-minute build
(this is *not* exageration; if the patch is on a plugin or even on
most of the core for instance, the rebuild will be very quick; but if
it starts rebuilding libgimp*, then we are doomed!).
When it's a separate remote, I even wonder if git will still make the
link between the 2 remotes? Will it try to rebuild everything from
scratch? This would be absolutely terrible.

What I would do to test a patch is:
- wget
- git apply (this won't make a commit so I won't push it by mistake)
- test it. If it looks good…
- git checkout -- .
- git am
- Optionally fix minor stuff and amend, edit the commit message if needed.
- git push

If the patch looks really simple and obviously good from the basic
visual review, I would just ignore the `git apply` steps. Just git am,
test, push. This can all be done in 15 minutes. In these 15 minutes,
the procedure and rebuild could take just 2 or 3 minutes; the 10+
additional minutes are because I do thorough tests even for small

If I am forced to checkout another branch, the procedure + build would
be suddenly extremely long and boring.

Now I don't say that there is no alternative. I guess what I would do
is: fetch the remote (don't checkout it), then cherry-pick only the
commit. This way, I avoid a stupid rebuild of useless stuff. It's
still not as good as previously since it will still take longer, and I
lose the `git apply` step, which is the step which allows me to work
and test patches on master without fearing making a stupid push
mistake. Now here too there are workarounds, like I could git reset
immediately to get rid of the commit (still keeping the code), but
that makes a lot of workarounds now! ;P

So yeah, that's not as bright and simple as it could be.


Carlos Soriano

-------- Original Message --------
Subject: Re: Proposal to deploy GitLab on
Local Time: May 17, 2017 12:47 PM
UTC Time: May 17, 2017 10:47 AM
From: jehan marmottard gmail com
To: Tristan Van Berkom <tristan vanberkom codethink co uk>,
desktop-devel-list <desktop-devel-list gnome org>


On Wed, May 17, 2017 at 12:33 PM, Sébastien Wilmet <swilmet gnome org>
On Tue, May 16, 2017 at 11:15:51PM +0900, Tristan Van Berkom wrote:
I don't share your optimism about gitlab bug tracking, nor do I share
in the mentioned frustration with bugzilla.

Me too, I like bugzilla (but not for doing code reviews).

What would be the pain points if GitLab is used only for git and code
reviews, and we keep bugzilla for the bug tracker? Have you considered
that option?

We would loose automatic links between bug tracker tickets and pull
requests. When a pull request is merged, we would need to close manually
the bugzilla ticket if everything is done. And when someone requests a
pull, the person would need to add a comment manually on bugzilla so
that other people know that the bug is being worked on.

Mmh I think that's not practical if the links must be done manually.

Maintaining the bugzilla instance would also require sysadmin time, and
development efforts to rebase the patches to new bugzilla versions.

I don't know, I'm excited about the idea to use a similar contribution
workflow as in GitHub, but less excited about having a bug tracker
similar to the GitHub one. (I've never used GitLab, but I'm familiar
with GitHub, and after seeing some screenshots it seems that the GitLab
bug tracker is similar to GitHub's).

I like bugzilla too and guess it probably does more than github/lab
bug trackers. But I also know there are annoying parts. Like someone
noted that searching projects in the long list of GNOME projects is
terrible experience (I even have a browser keyword so that I don't
have to do this anymore, because it was so annoying; but obviously new
contributors would not have such shortcuts).

Also the fact that the reports actually have less options is not bad
IMO. One gets lost in all these bz options. Simplicity is good
sometimes. :-)
gitlab has cool features too, like it's much easier to mention someone
to have them take a look at a report, for instance.
And finally, as you say, code review is much better. I like that you
can annotate line per line (easier for the reviewee in particular to
understand our review).

Bottom line: I definitely don't think we should keep both bz and
gitlab in the end.

The only thing I am annoyed at is this forking workflow. Both as a
contributor, and as a code committer/reviewer. Having to fetch a new
remote for every single-commit contribution out there is terrible.


desktop-devel-list mailing list
desktop-devel-list gnome org

ZeMarmot open animation film
desktop-devel-list mailing list
desktop-devel-list gnome org

ZeMarmot open animation film

[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]