Wireshark-dev: Re: [Wireshark-dev] Proposed Gerrit workflow (was: Re: Notes from Sharkfest '13)
From: Bálint Réczey <[email protected]>
Date: Tue, 25 Jun 2013 12:17:14 +0100
Hi Michael,

2013/6/24 Michael Tuexen <[email protected]>:
>>> The current process puts responsibility on the core developer who
>>> commits a change. Personally, I don't think it is bad if this breaks
>>> the build on some buildbot, I only this it is bad if the committer
>>> doesn't care. This worked out pretty well in the past, I think. I only
>>> run the head version of wireshark and can usually build it (thanks to the Mac OS X buildbots).
>>> If a core developer didn't want to take responsibility for a patch, he
>>> could contact others to get feedback on questions. This also worked in
>>> the past since you contacted people who also are interested in the subject.
>> I'm generally satisfied with the quality of trunk and I'm proud to be part of the project.
>> I also think if we could break trunk even less often, it would be even better and Gerrit would help that and would also help discussing the patches.
>>> The same responsibility applies for changes being compiled by the
>>> buildbots. Each such change comes from a core developer. I'm
>>> hesitating to allow an arbitrary patch to compile on the buildbots
>>> where we have no one being responsible for it in any way. Some of the
>>> buildbots run older software, some of them are not hardened in any way.
>> If all the buildbots are running a newly cloned VM and limits network usage of the VM, I think we can be safe.
> If we can change the buildbots running like that, you might be true. However,
> the current way the Mac OS X buildbots are running is different. The versions
> running currently on the buildbots are Leopard (PPC) and Snow Leopard (Intel)
> and would only allow the server versions to be virtualised. That would mean
> that we upgrade the buildbots to at least Lion...
Yes, this is a severe limitation of older OS X versions.

The solution could be limiting some tests to run only changes after integrating
them to the target branches based on time needed to run them or for other
reasons like this license problem.

I would prefer upgrading at least one of the OS X bots to use VM to catch bugs
earlier, though.