Re: Unreviewed^2 testsuite patches

[Mike Stump <mikestump at comcast dot net>]

On Jul 23, 2010, at 2:28 AM, Richard Guenther wrote:
> Increasing the timeout will make us fail notice compile-time regressions
> here.

It is wrong to use the 300 second timeout to notice compile time regressions.  People should avoid large testcases if possible.  Anything that is more than 30 seconds on a slow machine (10 seconds on a fast machine), is, well, too big.  I think the size of the testcase should be reduced to take 10 seconds on a slow machine, and the limit dropped to 12 seconds.  The desire of catching compile time regressions is laudable, but, not at the expense of failures on smaller machines, where the smaller machine is at least 1 GHz.  For machine smaller, they should increase the default timeout in their board file.