Re: github mirror of lilypond?

[Han-Wen Nienhuys]

On Sun, Jan 19, 2020 at 12:14 PM Jonas Hahnfeld <address@hidden> wrote:
> General things first: The UI of Gerrit v3 seems to be much nicer than
> its previous version. Back then I found it incredibly hard to just look

thanks, I'll pass on the compliments.

> > In the above, I sent a stack of 7
> > commits for review at once.
>
> That's where "Relation chain" and "Submitted together" come into play,
> right? Funnily enough, "Submitted together" only shows predecessors...

Yes, if you submit the change of the top of the stack, you typically
submit all the predecessors.

> From a high level, I think Gerrit expects reviewers to accept each
> change individually. Does this really model the development style of
> LilyPond?

It's rather the opposite. Since Gerrit has the actual git commits, you
can rebase, and cherry pick changes from the UI, so you can submit
changes individually, even if their predecessors are missing
approvals.

Rietveld has no notion of dependent changes at all.

> Being relatively new, I got the perception that we usually
> split up a change into logically grouped commits. Eventually I want
> them to be viewed and submitted as one push, not one by one as the
> individual commits are accepted.

Yes, you can do that too. see

https://gerrit.googlesource.com/gerrit/+/9e47119d65b7ac7ab4db568456e17b556c207f59

which merges in 4 changes in one go.

> In that regard I really like how Pull Requests / Merge Requests in
> GitHub / GitLab / gitea work: You can look at commit granularity if you
> want to, but in the end you accept the whole thing. That's also good
> because changes to a predecessor might require reworking later patches.

I find PRs unwieldy for complex changes, because the GH interface
doesn't let you compare different commits in the PR to each other. For
example, the change

  https://git.eclipse.org/r/c/146568/

went through 92 rounds of review before it was merged. The GH UI isn't
really suited to this much back and forth on a review.

That said, as said before, I like gerrit, but I think the benefits of
switching to a platform (like GH or GL) outweigh the limitations of
their review experience for LilyPond.


> > Working with a single commit requires using commit --amend and with
> > multiple commits requires using rebase -i, both of which Git newbies
> > seems a little scared of.
>
> Right, but as most other tools you can probably just upload and update
> a patch file somewhere? That would be the easiest solution for new
> contributors.

Gerrit has a complete git server inside. You don't upload patches, but
you do a git push of your to-be-reviewed code instead. Other people
can then clone your code (see the command under the "download"
button.)

> Jonas
>
> >
> > On Sun, Jan 19, 2020 at 4:34 AM Karlin High <
> > address@hidden
> > > wrote:
> > > On 1/18/2020 4:59 AM, Jonas Hahnfeld via Discussions on LilyPond
> > > development wrote:
> > > > I strongly dislike Gerrit, it's really hard to learn and even after
> > > > some time I still can't figure out how to use it correctly.
> > >
> > > This topic is outside my expertise. But I understood that Gerrit and
> > > Rietveld are both descendants of Google's proprietary internal-use
> > > Mondrian code review tool.
> > >
> > > <
> > > https://www.gerritcodereview.com/about.html
> > > >
> > >
> > > I see quite a few patches lately that show Jonas Hahnfeld as the author...
> > >
> > > <
> > > https://codereview.appspot.com/user/hahnjo
> > > >
> > > <
> > > https://git.savannah.gnu.org/cgit/lilypond.git/log/?qt=author&q=Jonas+Hahnfeld
> > > >
> > >
> > > ...so I'm wondering if Rietveld has you suffering in silence ;) or if
> > > the fork and rewrite mentioned on that Gerrit "about" page has greatly
> > > diverged the user experiences of these respective products?
> > > --
> > > Karlin High
> > > Missouri, USA
> >
> >
> >
> >



-- 
Han-Wen Nienhuys - address@hidden - http://www.xs4all.nl/~hanwen