Brock Pytlik wrote: > >>>>https://cr.opensolaris.org/action/browse/pkg/bpytlik/7139940-v1 > >>> > >>>dependencies.py: > >>> > >>> - I'm assuming this change is purely for performance, and not really > >>> related to either bug? > >>Well, the removal of the if system_patterns is just removal of completely > >>silly code that got in there accidentally. > >>The other code change is related to the changes made elsewhere. > >>Specifically, is_pkg_installed doesn't work with fmris that don't have > >>publishers. I looked into fixing that (since that seems kinda broken to me) > >>but it involved an unfortunately large amount of change. Since the only > >>place we called get_manifest with a fmri without a publisher was from the > >>dependency code, that seemed like the expedient change for now, with an RFE > >>to be filed for is_pkg_installed. > >Maybe I was misled by "efficiently"? If it doesn't work at all, then > >perhaps you should say so. You might file a bug against this (if there > >isn't one already), along with the note that this code should be tweaked > >again once the fix is in. > > Yeah, I think I wrote that comment when I was still planning on reworking > the lookup code. Even if it's fixed though, I'd leave the comment (or > restore it to say efficiently) since if we know which publisher we're > looking for, that will always be at least as fast as not knowing.
Maybe, maybe not; depends on the nature of the fix. But for now, keep the comment, but rewrite it to indicate that it simply won't work otherwise. > As a side note, I think I was working on the assumption that it was easier > and simpler to just grab all ro data (which in practice was 90+% signing > data) for tests which needed it (which in practice was 90+% of tests that > needed ro data) since having extra data didn't really cost much and kept the > number of flags to a minimum. If that balance has shifted, I'm happy for > someone to change it as desired. It's more a matter of putting the correct metadata on the test (or test class or whatever). I doubt that there's any performance issue with always copying the ro data in all cases, given how small it is right now, but having a tag that indicates that signing is being used is more specific and accurate and meaningful and maintainable should our assumptions change, even if under the covers all it does for now is exactly the same as the more generic need_ro_data. > >>>t_fix.py: > >>> > >>> - line 330: "runing" -> "running" > >>> > >>> - line 336: why this first fix? Isn't this always a no-op? > >>To make sure that the reason the fix fails is because the package isn't > >>signed, and not any other reason. > >I'm surprised that you're worrying about a package that's just been > >installed being in a non-verifiable state. If that's happening, > >something's awfully wrong. > I agree that something would be awfully wrong. If I'd made such a change in > my gate though, having it fail before I change the property instead of after > would probably save my a lot of headache while trying to figure out what > went wrong. I can remove it if you prefer. Please, thanks. Danek _______________________________________________ pkg-discuss mailing list [email protected] http://mail.opensolaris.org/mailman/listinfo/pkg-discuss
