On Fri, Feb 14, 2014 at 9:16 AM, Richard W.M. Jones <rjones(a)redhat.com>wrote:
It sounds rather ill-defined :-) What counts as large amounts of
memory or CPU?
Yes, I really don't know. CPU isn't so much the concern, anyway. Let
those builder churn away for long periods of time. Bwahahahaha!
But memory, now, that's an issue. All I know for sure is that the x86_64
and i686 builds succeeded, so those boxes had "enough" memory; the ARM
build failed, so that box did not have "enough" memory. Since memory is
exhausted while compiling the test, not while running the test, the
required amount of memory isn't necessarily fixed. It may depend on
architecture, version of the compiler, and possibly other factors that I
don't know about. I really don't know what "enough" is, therefore ...
Can you choose it based on something like the output of `free -m`
and/or `grep -i bogomips /proc/cpuinfo` ? Note the second command
might not produce any output (no output on ARM for sure) so don't rely
... I don't really see how I can do something like this. For now I have
turned the troublesome tests off for ARM only, and left behind a comment in
the spec that if other architectures fail to compile the tests due to
memory exhaustion, the same should be done for them.