On 4/10/2012 1:26 PM, Garrett Wollman wrote: > <<On Tue, 10 Apr 2012 08:17:14 -0400, Jeffrey Altman > <jalt...@secure-endpoints.com> said: > >> Its quite simple really. You ask each of the builder owners to run make >> check on their system before you turn it on. We know that "make check" >> does not work across the board. It never has. That is why it wasn't >> turned on previously. > > It's not quite so simple, really. Currently, "make check" does a lot > of building in the process of testing things, and these test harnesses > really should be built and tested the same way as other parts of the > code base are built and tested -- otherwise the code will just rot, as > it clearly has already.
I don't see how requiring that builder owners demonstrate that "make check" works on their system prior to adding "make check" to that system in the buildbot is contrary to testing the test harnesses. When we are processing 50+ patchsets a day through buildbot, why should we break everything that is unrelated to the test harnesses while waiting to determine that the test harnesses can in fact complete successfully? I will also note that there are other aspects of the OpenAFS build process that are not tested by buildbot: 1. documentation 2. installer packaging They are not turned on because doing so significantly increases the build times for each patchset. The rate of change of documentation at this point is also so low that it is more efficient to test each patchset manually. Jeffrey Altman
signature.asc
Description: OpenPGP digital signature
