On 08/08/2014 02:30 PM, Tom Rini wrote:
On Fri, Aug 08, 2014 at 02:19:34PM -0700, York Sun wrote:
On 08/08/2014 02:12 PM, Tom Rini wrote:
[snip]
Oh I'm sure, but maybe we can help get everyone a better script. For example, you've said the way your job goes currently is every commit gets a test. But.. that's not what you need exactly it sounds like. It sounds like you need a given gerrit issue, which can have a few commits to it, to be bisectable and compile tested. buildman will handle this part for you, if you let it.
Tom,
I agree buildman has improvements comparing with MAKEALL. I spent time to make Gerrit, Jenkins and MAKEALL work together. It is not easy as one would expect. For now, if I can make builmand to build the last commit, I can replace MAKEALL with buildman. But that seems not working.
I believe you, I've spent a bunch of time inside of Jenkins myself (and then with IBM's LSF compute setup) and I'm doing some of the same hoop-jumping now too. But I'm hoping maybe we can all take this as a chance to improve the various ad-hoc builder setups many of us have.
Since we use Gerrit for internal review, each commit (proposed patch) triggers a Jenkins event. I can change the script so no build is triggered, but I need to explorer if buildman servers better than MAKEALL in this case. Because Jenkins has no knowledge of the number of patches pushed, I don't think buildman has enough information to find the "top" and start the build.
But anyway, it shouldn't stop us from moving to buildman. I just need the basic feature available before removing MAKEALL.
York