[
Date Prev][
Date Next][
Thread Prev][
Thread Next][
Date Index][
Thread Index]
[
List Home]
Re: [mdt-ocl.dev] Playing with gerrit
|
Hi
The earlier discussion was leaving me a bit discouraged, but..
On 18/07/2013 12:23, Laurent Goubet wrote:
For big features, we still develop on branches, but we try never to
push them on the eclipse repository. Not only are branches dangerous
(you have to remember never to rebase a branch that has been pushed to
a remote repository), but reviewing a long stream of commits seems
useless : what we wish to review is the end result (tip of the dev
branch), and how it changes from the existing state (master at the
time of review, not at the branching time). Reviewing such branches is
tedious and error-prone. Plus, IMHO, merging makes for a very messy
and illegible history.
I very much agree. If you look at the OCL history, I don't think you
will find a non-fast forward merge that I have done in the last year. I
always rebase onto master if I can, but too often cherry pick one by
one. The many branches on OCL are actually far more controlled than you
might think. In the long term there is just the master and maintenance
branches, with some old side branches for attempted improvemnts that
have yet to be completed/archived. In the short term there are sometimes
multiple improvments underway but they get linearised before I push to
master.
When we think that the new feature is ready, we prepare it for review
by squashing the commits into a single one with a proper commit
message, which we push for review. What you see in the review I've
linked here (https://git.eclipse.org/r/#/c/13903/) is the result of
such a squash.
For a single change I certainly prefer to compare the start/finish; I
don't have time to study the mistakes that were made along the way.
Unfortunately Adolfo's second try at a review request failed to squash
correctly, so I'm a bit worried that an intelligent user is making such
a mess of a trivial change. I'm not impressed by the web interface to
Gerrit and apparently there is no working EGIT integration, so I had to
change to Eclipse, navigate to the file to verify that the Gerrit
display was wrong.
It's 'just' a matter of better discipline to try to avoid two
improvements getting tangled, which is very difficult unless you have a
very fast review response, and that is counterproductive because often I
like an improvement to have a little time to settle since it was often
triggered by another problem, and may deserve further work.
Regards
Ed Willink