On Wed, Sep 28, 2011 at 10:28:55PM -0700, Shawn Walker wrote: > On 09/28/11 13:48, Edward Pilatowicz wrote: > >On Tue, Sep 27, 2011 at 04:49:06PM -0700, Shawn Walker wrote: > >>Greetings, > >> > >>The following diff (and webrev) contains the fix for the following issue: > >> > >> 18976 pkg solver needs to go on a dependency diet > >> > >> > >>https://cr.opensolaris.org/action/browse/pkg/swalker/pkg-18976/webrev/ > >> > >>diff -r 32ef5a3b4c12 src/modules/client/pkg_solver.py > >>--- a/src/modules/client/pkg_solver.py Mon Sep 26 12:36:10 2011 -0700 > >>+++ b/src/modules/client/pkg_solver.py Tue Sep 27 16:33:49 2011 -0700 > >>@@ -1129,6 +1129,9 @@ > >> f > >> for da in self.__get_dependency_actions(fmri, > >> excludes) > >>+ if da.attrs["type"] != "incorporate" and > >>+ da.attrs["type"] != "optional" and > >>+ da.attrs["type"] != "exclude" > >> for f in self.__parse_dependency(da, fmri, > >> dotrim, check_req=True)[1] > >> ]) > >> > >>This fix reduced total update operation time on my system by roughly > >>66%. If you exclude image planning execution time, the reduction is > >>91%. Details are below. > >> > > > >first, this is awesome. > > > >my only comment on the fix is that you could also skip "parent" > >dependencies. actually, parent dependencies shouldn't be part of > >possible_set, so we probably should have already been skipping them here > >already. of course it hasn't caused a problem yet since the only parent > >dependencies we have so far are the special "self" ones. if you don't > >want to skip them in this wad let me know and i'll get a separate bug > >filed. > > Actually, it seems like parent dependencies could affect the > possible set if the fmri isn't for "self" and self.__parent_pkgs is > defined. >
well, ultimately, yes it affects what can possibly be installed, but we figure that out later when we trim packages with unsatisfied parent dependencies. if we had non-self parent dependencies, then adding them to be possible_set would actually be kinda broken. > I tried adding parent to the list of dependencies that aren't parsed > and it made no difference in upgrade times (at least for my test > cases, which admittedly weren't with any zones). > > Given my uncertainty surrounding that, I'd rather we file a separate > bug to investigate parent dependencies specifically since we have > already said that we may be able to move the parsing/trimming of > them earlier. > sure. i'll get that filed. ed _______________________________________________ pkg-discuss mailing list [email protected] http://mail.opensolaris.org/mailman/listinfo/pkg-discuss
