Wireshark-dev: Re: [Wireshark-dev] Proposed Gerrit workflow (was: Re: Notes from Sharkfest '13)
From: Marc Petit-Huguenin <[email protected]>
Date: Sat, 22 Jun 2013 10:17:02 -0700
Hash: SHA256

On 06/22/2013 09:43 AM, Bálint Réczey wrote:
> Hi Marc,
> 2013/6/22 Marc Petit-Huguenin <[email protected]>:
>> -----BEGIN PGP SIGNED MESSAGE----- Hash: SHA256
>> On 06/22/2013 03:47 AM, Bálint Réczey wrote:
>>> Hi All,
>>> 2013/6/21 Marc Petit-Huguenin <[email protected]>:
>>>> -----BEGIN PGP SIGNED MESSAGE----- Hash: SHA256
>>>> On 06/20/2013 04:52 PM, Guy Harris wrote:
>>>>> On Jun 20, 2013, at 2:58 PM, Marc Petit-Huguenin 
>>>>> <[email protected]> wrote:
>>>>>> On 06/20/2013 02:17 PM, Gerald Combs wrote:
>>>>>>> Advantates: - I'm not sure that an in-house equivalent (e.g. 
>>>>>>> Gerrit plus a private repository) would be better than what
>>>>>>> Github offers.
>>>>>> Yes, Gerrit is better than github:
>>>>> Presumably you mean "Gerrit plus a private repository is better
>>>>> than github", as Gerrit, as far as I can tell, is just software
>>>>> that works with a Git repository.
>>>> Yes, although managing repositories being what Gerrit do, Gerrit
>>>> without a least one repository would be a very boring application.
>>> :-)
>>> I have started describing a Gerrit based workflow which IMO would fit
>>> to the project at http://wiki.wireshark.org/Development/Workflow .
>>> Please check it and share your opinion.
>> "Code is building and tests are passing on all platforms. (Tests
>> automatically start when at least one Core Developer gives +1 or +2 to
>> prevent overloading or cracking the build servers.)"
>> Why do not build and test all patchsets submitted?  Is that a limitation
>> of the build servers?  Having Jenkins automatically verify your patchset
>> is IMO one of the nice feature of Gerrit, and it will lower the workload
>> of core devs if building and testing are done before they start looking
>> at the patchset.
> Build can be triggered by patchset submissin, too, but it would require
> more build server resources. Usually not the first version of the
> changeset will be accepted especially from new contributors and this means
> more builds.

My view is the opposite: New contributors patchsets will probably be rejected
anyway (does not build, does not pass test, etc...), so having the system
doing that lowers the burden on core developers, who they can focus on more
high level problems.

> Note that Core Developers would not have to wait since they can give +1 for
> their own changesets.
> The other reason behind requiring a +1 from someone we trust is that 
> otherwise it would be easy to prepare a changeset which does unspeakable
> things to the build servers which we don't want to happen. Without
> requiring +1 we would have to prepare build systems to cope with malicious
> commits.

That is a good point (basically because of the halting problem).  But builds
are done in isolation (i.e a git clone is done each time), so apart using too
much resources or never ending, there is no harm that can be done to the
infrastructure.  And there is a Jenkins plugin to abort a build if it is stuck.

- -- 
Marc Petit-Huguenin
Email: [email protected]
Blog: http://blog.marc.petit-huguenin.org
Profile: http://www.linkedin.com/in/petithug
Version: GnuPG v1.4.12 (GNU/Linux)