Il giorno 10/lug/2013 20:35, "Iustin Pop" <[email protected]> ha scritto: > > On Wed, Jul 10, 2013 at 11:17:45AM +0200, Michele Tartara wrote: > > On Wed, Jul 10, 2013 at 11:10 AM, Klaus Aehlig <[email protected]> wrote: > > > > > > > And I guess I should also add the order dependency > > > $(BUILT_PYTHON_SOURCES). > > > > > Don't you think ? > > > > > > > > > > > > > For this, it's really not clear to me why it's there. I actually think > > > it's > > > > completely useless and just a remainder of old times. The files that are > > > > part of BUILT_PYTHON_SOURCES are already dependencies of test/hs/htest > > > > through the rule for building src/Ganeti/Constants.hs. I think it can be > > > > removed from the rule from hs-tests as well. > > > > > > Are you saying we should not mention a dependency needed in the rule > > > just because it's an indirect dependency anyway? What if at some point > > > we change the dependencies of the build rules that pulls the needed > > > indirect dependency in? > > > > > > > Of course not. > > It's just that I cannot see where this dependency is needed by this rule > > directly. If it is, of course it should be stated, but as far as I can see, > > we need the python constants (part of the BUILT_PYTHON_SOURCES) just to > > build the haskell constants that are needed to build test/hs/htest. > > If this is a case, it shouldn't be a direct dependency of hs-tests-%. > > > > On the other hand, if the dependency is there for some other reason... > > please, enlighten me, because I can't see why! :-) > > I believe the dependency needs to remain. It is no for the constants > file, but because some Haskell tests execute Python code directly (the > OpCode compat tests, etc.) and as such the Python code should be runable > when/before this task is run, separately from the constants.py to > Constants.hs change. > > That is, unless something changed since I last looked at the code (which > is a while ago). > > As always, best way to test this is to built everything, then remove the > generated python sources, and then run the tests with -j, so that you > 'force' as much as possible any potential race conditions. > > regards, > iustin
Yes, you are right. In fact, in the end the dependence was left in the final revision of the patch (and added for the new rules). Thanks for the additional background info on why it's actually needed, though. Cheers, Michele
