On 02/13/11 09:04 PM, Tim Foster wrote: > Hi there, > > Based on feedback from the last round of code review, I've got another > cut of the pkgdepend bypass-generate functionality. I have addressed > the comments I received from the previous round of code review, thanks > again for that. > > Here are webrevs, both complete, and incremental from the first code > review (since that was the one that most people looked at; the > incremental does unfortunately show some diffs unrelated to my changes, > as I've merged with the latest pkg-gate tip) > > http://cr.opensolaris.org/~timf/pkgdep-python-webrev-2 > http://cr.opensolaris.org/~timf/pkgdep-python-webrev-0-vs-2
src/modules/flavor/base.py: 63-64: I know I requested this change, but I might still be unclear on this. You can set pkg.depend.runpath on a manifest, and/or on an action. If set for an action, it completely overrides the manifest setting, but PD_DEFAULT_RUNPATH doesn't get expanded on a per-manifest basis because its value is actually flavor-dependent? So the error could stem from the setting of either a manifest attribute or an action attribute, but it will always be detected and reported per-action? 357-374: Factoring this out is nicer than implementing it for each caller, but doesn't address my earlier question, which (though perhaps not well phrased) was really about performance. The callers of insert_default_runpath() must still calculate the default runpath, which I had assumed to be expensive, since it's done on a per-action basis. On a related note, is there a performance difference between simply attempting to get the index PD_DEFAULT_RUNPATH and handling the resulting ValueError, versus testing for PD_DEFAULT_RUNPATH in run_paths? I'm not insistent on changes here, I'm just sensitive to the fact that we're calling pkgdepend generate a lot (800?) of times in the OS/Net pkg build, and it would be a shame to slow it down needlessly. --Mark > The main functional differences are: > > * pkg.depend.bypass-generate attribute values are regular expressions > > * when checking for bypasses, we expand the pkg.debug.depend.file and > pkg.debug.depend.path values into pkg.debug.depend.fullpath > values, then omit files which match our regular expression > (rather than trying to split off dependencies as before, which > was slightly hairy) > > I've done sparc & x86 nightly builds on ON with these bits, using an ON > gate which employs both new pieces of functionality, checking that we're > generating the same dependencies (without the majority of the pkgdepend > errors, bugs filed for everything that's left - 3 failures on sparc, 2 > on x86) I have done a full pkg-gate test run too. > > All comments would be most welcome, thanks. > > cheers, > tim > > > _______________________________________________ > pkg-discuss mailing list > [email protected] > http://mail.opensolaris.org/mailman/listinfo/pkg-discuss _______________________________________________ pkg-discuss mailing list [email protected] http://mail.opensolaris.org/mailman/listinfo/pkg-discuss
