Santiago M. Mola wrote:
It's not as simple as that. A package may fail tests because compiler
bugs, build environment misconfiguration, problems in a library which
is being used, a setup problem or, of course, a bug in the package
which shows up in rare cases and haven't been spotted by upstream yet.
May happen indeed.
When the package can be critical for the system, upgrading to a buggy
version will eat people's dogs. I feel a bit safer when I run the test
phase for my package manager, and I wouldn't install it if it's
failing any test. I don't think that's too paranoid.
The main point is that it could be overly bothersome, if you depend on
certain applications you won't just use the standard testsuite but also
run your batch of compliance checks, but that isn't common.
Upstream clearly states that a gmp build which tests have failed
shouldn't be used. I bet they deny support for users who fail to
follow that indication ;-)
gmp isn't a key component if you aren't using math/sci applications
using it. You may point openssl as something you may want to have a
round of checks before is too late, same for openssh.
Still there are people perfectly happy w/out those since they do not use
those packages that for me and possibly many other are vital.
I won't be happy to have gcc have its batch of tests run, just to see
later it fails on ffmpeg because the tests do not catch those
conditions, I have better way to break gcc than those you have in the
regression test =/
Changing the default features would just at best have people that do not
care switch to -test, people that care already about that won't be
affected and just create an annoyance.
Putting it in an eapi makes not much sense as well since you may change
the defaults as you wish since they aren't causing incompatibilities.
To sum up:
- having the test feature on by default isn't good for anybody but
paranoids and lazy developers, paranoids have that already on, lazy
developers will switch it off for them and let people do the automated
test for them.
- having that mandated by the eapi doesn't have sense since it doesn't
change anything by itself.
lu
--
Luca Barbato
Gentoo Council Member
Gentoo/linux Gentoo/PPC
http://dev.gentoo.org/~lu_zero
--
gentoo-dev@lists.gentoo.org mailing list
