On Thu, Jul 27, 2017 at 12:39:32PM +0200, Christophe Fergeau wrote:On Thu, Jul 27, 2017 at 11:54:08AM +0200, Victor Toso wrote:
On Mon, Jul 24, 2017 at 10:47:34AM -0400, Frediano Ziglio wrote:
Not really familiar with GitLab merge requests but on GitHub they
remain open till closed so this would help with old ones.
The big change on moving to full PR is the way of commenting patches.
Unless PR are just used for tracking and are replicated to ML but
maybe is hard to keep them consistent. I think is possible to configure
PRs to send changes to ML. This would make the history persistent on
the ML.
Could we try for a period and see how does it go?
+1, but how should we approach that? I think we need to move from
freedesktop to either gitlab or github first otherwise we can get
confused on what's going on.
Just to be sure, you are both suggesting switching to pull requests, and
doing the reviews in gitlab web UI?
The initial problem is that some reviews are not done in a timely
manner. Being able to easily get a list of pending reviews was brought
forward as a potential solution to this problem, and apparently, you
both think that switching from email based reviews to a web based review
system would help in getting more reviews faster? (iow, it would make
you more efficient at reviewing code, and you vastly prefer that over
email).
I'm not necessarily opposed to trying things out, I'm just trying to get
a clear view of what we are expecting to get out of the change.
IME, if you're not getting fast enough reviews on patches, the problemisn't usually the tools, it is the lack of reviewer bandwidth. On OpenStackwe used Gerrit for reviewers, had tonnes of pretty graphs, metrics, reportson what reviews were stalled, how long reviews were waited for, etc, etc.It didn't help getting reviews done in a timely manner. You still had togo and manually "ping" people to get attention on your reviews they hadnot looked at or forgotten about.
+1 on your analysis, but I the problem I heard about was not that we don’t go fast enough, but rather that some patches get dropped.
Added to that the web UI was so inefficient to deal with (particularly for
patch series), that people ended up writing command line tools to do code
review in a "email like" user interface and then posting the results back
to gerrit.
This is why I’m really advocating baby steps. Stick to mail for now, just have one shared place where we agree to post “large” things under review. I suggested pushing branches with some naming conventions because we are on freedesktop which has no merge request. If we switch to gitlab or github, a merge request or pull request would be perfectly fine.
Small one-liners are usually reviewed instantly, so as Christophe pointed out, it’s probably not worth it for small things.
|