I found working on bug 1316140 that un-applying that patch can require a clobber, because |./mach build| is trusting that the config.status it finds is up to date.

Comment on attachment 8811008 [details] Bug 1317778 - Emit a depfile with python configure dependencies so Make will know when to re-run configure. https://reviewboard.mozilla.org/r/93266/#review94846 There's actually more than configure.py that can affect config.status (and in fact, configure.py doesn't change enough that adding it manually makes a difference). We should probably make configure emit a proper dependency file and include it from client.mk.

Comment on attachment 8811008 [details] Bug 1317778 - Emit a depfile with python configure dependencies so Make will know when to re-run configure. https://reviewboard.mozilla.org/r/93266/#review97418 ::: configure.py:36 (Diff revision 2) > if sandbox._help: > return 0 > > - return config_status(config) > + mk = Makefile() > + rule = mk.create_rule() > + rule.add_targets([os.path.join(config['TOPSRCDIR'], 'configure'), It will be better for make to: - use a literal '$(TOPSRCDIR)', because in some corner cases, config['TOPSRCDIR'] might be different, and that would make the dependencies unused. - use mozpath instead of os.path, because forward slashes are better. - and in fact, the right target for the dependencies is not $(TOPSRCDIR)/**/configure, but $(OBJDIR)/config.status. This should also go into the config_status function.

Comment on attachment 8811008 [details] Bug 1317778 - Emit a depfile with python configure dependencies so Make will know when to re-run configure. https://reviewboard.mozilla.org/r/93266/#review98002 ::: client.mk:333 (Diff revision 3) > $(TOPSRCDIR)/testing/mozbase/packages.txt \ > $(OBJDIR)/.mozconfig.json \ > $(NULL) > > +# Include a dep file emitted by configure to track Python files that > +# may incluence the result of configure. typo ::: configure.py:36 (Diff revision 3) > if sandbox._help: > return 0 > > - return config_status(config) > + return config_status(config, sandbox) > > - > +def config_status(config, sandbox): Mmmmm it would be better if we could avoid the signature change, so that we don't break comm-central. How about a set_config() at the end of moz.configure to be able to grab _all_paths from config instead of sandbox? But I won't stop you if you don't feel like it's necessary ; after all, that only requires a one-liner in comm-central to be fixed.

Comment on attachment 8811008 [details] Bug 1317778 - Emit a depfile with python configure dependencies so Make will know when to re-run configure. https://reviewboard.mozilla.org/r/93266/#review100554 Sorry for the delay.

Pushed by cmanchester@mozilla.com: https://hg.mozilla.org/integration/mozilla-inbound/rev/202f46bb664d Emit a depfile with python configure dependencies so Make will know when to re-run configure. r=glandium