On 11/11/2014 08:14 PM, Mike Holmes wrote:


On 11 November 2014 11:33, Taras Kondratiuk <[email protected]
<mailto:[email protected]>> wrote:

    On 11/11/2014 06:22 PM, Anders Roxell wrote:

        On 11 November 2014 17:12, Taras Kondratiuk
        <[email protected]
        <mailto:[email protected]>> wrote:

            Can we try as an experiment reviewing this series in Gerrit?


        and again, gerrit isn't good for discussions in patches =(


    Why not? It has its pros and cons if compared to ML review.


I also wanted gerrit when B&B in Linaro switched to it, here is what
more experience gerit users told me were the flaws.

 1. you may need a linaro account to send patches to ODP,  I assume this
    is true if we re use Linaro gerrit infrastructure
 2. The major concern was that apparently it is hard to have discussions
    via gerrit, I have never tried beyond using it for the q/a Testing
    repo in Linaro which are small straightforward patches.

I like gerrit because

  * current state of all patches is extremely clear - no grepping your mail
  * mandatory reviewers can be specified - you know who to target to get
    signoffs, on the other side you know you are expected to review
    something (maintainers, area experts get put on this list)
  * Objections to patches are clearly documented and not lost down a
    mail thread - no recycling issues

I personally believe that the discussion issue no.2 is resolved by the
patch submitter taking the initiative and calling, hangout, mailing the
folks with comments and then recording the result in gerrit rather than
bogging the list with the minutia - we could do that now too :)

Item one can be fixed no doubt, but for item two I am not sure it fixes
our fundamental problem, we need tight communication between submitters
any key reviewers before a patch even hits the list.

For 1.0 we should stick with what we have, we cannot afford a round of
assigning gerrit permissions to people and learning the too etcl right now.

Review in ML works nice if everybody follows a set of simple rules:
- Use proper mail client which use '>' for quoting and doesn't corrupt
  patches in replied email.
- Plain text only. Maybe it worth to force ML to drop HTML emails
  completely.
- Bottom posting.
- Leave only necessary context in a reply instead of resending a whole
  patch again.
- Proper subject prefix.
- Resend a new version of a whole series even if one patch was modified.
  Resending only separate patches makes it next to impossible for
  reviewers to reconstruct the latest series.
- Do not resend big series too often, because it clutters ML and
  fragments review comments. People will often miss a new series and
  review an old version of it.

Currently nobody follows these rules and review of big series becomes a
nightmare. Unless everybody start to follow them I'd prefer to use
Gerrit.

_______________________________________________
lng-odp mailing list
[email protected]
http://lists.linaro.org/mailman/listinfo/lng-odp

Reply via email to