[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Radical idea, test backwards compatibility instead of recompile

I think we've made a mistake. When glibc 2.1.94 was released and it broke
several packages those packages rushed out recompiles against 2.1.94 and
several other packages have been released since compiled against 2.1.94.

glibc 2.1.94 claims to provide backwards compatibility. That's an important
feature for smooth upgrades and for binary compatibility with other
distributions. By recompiling all these binaries so quickly after the new
glibc was uploaded we've failed to actually test this feature thoroughly.

Now that glibc has been fixed (as I understand?) I suggest we should roll back
these other releases and hold off from recompiles against glibc 2.1.94 for a
few weeks. This would give developers time to actually test glibc 2.1.94
against packages compiled with glibc 2.1. Any failures that weren't present
previously should be filed against 2.1.94.

Failing that the only alternative is to have a handful of people install
potato and then upgrade glibc and nothing else. However this is a much much
weaker test that could miss major problems in glibc that only happen to bite
under some circumstances or some programs.


Reply to: