Ivanov, Alexey A wrote:
-----Original Message-----
From: Geir Magnusson Jr. [mailto:[EMAIL PROTECTED]
Wouldn't the CI systems running on those platforms catch it automatically?
That's what we talk about, isn't it? If someone has fixed a test, it's better
to exclude it from *all* the exclude lists and hoping it won't fail on all
platforms. If it fails on a platform, everybody will be notified about the
failure. Then we decide either to fix it, if it's easy and straightforward, or
to add this test into exclude list for that certain platform.
Am I right?
I think we've gone off track. I think we have enough degrees of freedom
that we can test a few ways of doing this, and find out what works best.
I'm assuming that even if we do the "time waster" process - remove
from exclude lists that you can test, and then send a note to the list -
it's not that big of a burden given we all Know for Certain :) that
there will be very few of these, if any ;)
geir
And remove a test from all exclude lists means you need to edit 8 files rather
than just one.
Regards,
Alexey.
geir
Thanks,
24.11.06, Geir Magnusson Jr.<[EMAIL PROTECTED]> написал(а):
Alexei Zakharov wrote:
Could be - I don't see where the waste of time comes in.
Just a simple arithmetic: 8 (test runs) is bigger (longer) than 1 or
2, + synchronizing costs..
Sorry? The 8 files reflect 8 *different platforms*, which are each
going to run anyway.
geir
Thanks,
24.11.06, Geir Magnusson Jr.<[EMAIL PROTECTED]> написал(а):
Alexei Zakharov wrote:
In this case we can run against the situation when the test will
remain excluded forever for some platform.
Why? Our goal is to get rid of any excluded tests. Consider the
excluded tests as the existence of a regression, and we should work
to
fix ASAP.
We aren't "parking" them there to be ignored - we're putting them
there
so that the build can complete while we work on the fixes. We're in
essence "grandfathering-in" these regressions.
People have their own
problems. IMHO we should do this only if there are suspicions
that the
test is platform dependent. Otherwise it is a waste of time IMO
and CC
can handle this.
Could be - I don't see where the waste of time comes in.
geir
Thanks,
24.11.06, Geir Magnusson Jr.<[EMAIL PROTECTED]> написал(а):
Why not just remove from the platforms you have, and ask others to
update (or not) platforms you don't have as appropriate?
geir
Tim Ellison wrote:
Alexei Zakharov wrote:
Hi,
So we just should choice what is better: to break sometimes
tests run
or to
forget enable test(s) on some platforms.
Yesterday, when I was removing one of the beans tests from
exclude
lists, I feel a bit uncomfortable while updating
exclude.linux.x86_64.xxx since I have no (easy) access to such
systems
and had no plans to run tests on it. IMHO (in the perfect world)
the
fact that I remove or add something from / to
exclude.linux.x86_64
means I've at least ran tests for this platform and obtained
some
result. So let's have a common list, it's easier to deal
with it
psychologically. :-)
As for above question: +1 for being optimists, i.e. to remove
the test
from common list if it passes on all platforms available to
tester.
+1 (otherwise I'll share the psychotherapy costs with you :-)
--
Alexey A. Ivanov
Intel Enterprise Solutions Software Division
