Adding the autopkgtest-devel list to involve the current maintainers. On Sun, Mar 16, 2014 at 11:27:06PM +0100, Stephen Kitt wrote: > (Hi Ian, I'm adding you to the discussion since I'm talking about changes to > DEP8. I've snipped the exchanges but I hope you can get the gist of it > without needing to spend time looking anything else up!) > > On Sun, 16 Mar 2014 17:39:20 -0300, Antonio Terceiro <terce...@debian.org> > wrote: > > On Sun, Mar 16, 2014 at 08:39:38PM +0100, Stephen Kitt wrote: > > > On Sun, 16 Mar 2014 15:41:15 -0300, Antonio Terceiro <terce...@debian.org> > > > wrote: > > > > ... this is an important point. You have to make sure that the any > > > > tests will run against the code that is _installed_ and not against > > > > the code that was just built. Also, it would be really nice on the > > > > test infrastructure if you could build strictly the bits you need to > > > > the tests instead of building everything. e.g. the ideal would be > > > > build _only_ the unit test files (assuming they need to be built) > > > > and not all the other code (since you are supposed to run the tests > > > > against the installed version of the package) > > > > > > Indeed, thanks to Jakub for pointing that out. I've reworked the > > > upstream tests to build using the installed package. > > > > > > Your point concerning building only the required bits effectively > > > means that we shouldn't use "build-needed" in the tests/control > > > "Restrictions" field, but manually control the build only for the > > > purposes of running the tests, am I right? > > > > most probably, yes. > > > > It's always a compromise. Sometimes it's reasonably easy to patch the > > upstream test suite to make it not expect locally-built binaries, not > > use local includes etc. Sometimes it may be a lot harder, so YMMV. > > Right, and in this particular instance it's not particularly difficult. > > What bothers me is that the current DEP8 spec says that packages can rely on > having their source tree in the built state by stating "Restrictions: > build-needed", but effectively that imposes too much of a burden on the > testing infrastructure. (That's not a complaint, I don't think we should > require another buildd network to run tests, at least not until we've got as > much test code as binary-targeted source code.) It's the kind of expectation > that makes sense in a "traditional" CI setting (e.g. Jenkins with Maven for > Java projects, where the project is built and its tests run in the same > environment), but with DEP8's aim of testing the installed binaries it seems > less useful to me. Wouldn't it make sense to change DEP8 to encourage > building whatever is strictly required for the tests, and perhaps drop > "build-needed" altogether? > > Regards, > > Stephen
-- Antonio Terceiro <terce...@debian.org>
signature.asc
Description: Digital signature