[llvm-commits] [llvm] r76558 - in /llvm/trunk: lib/CodeGen/VirtRegRewriter.cpp test/CodeGen/X86/2009-07-21-RewriterBug.ll

[David Greene]

On Wednesday 22 July 2009 16:36, Evan Cheng wrote:
> On Jul 22, 2009, at 11:48 AM, David Greene wrote:
> > On Wednesday 22 July 2009 12:54, Evan Cheng wrote:
> >> It's part of the test suite: 256.bzip2.
> >
> > Would it be possible to save the reduced testcase somewhere?  It's
> > nice not
> > to have to sift through a whole application.  If the application
> > starts
> > failing, it would be convenient to run the smaller testcases to see if
> > there's been a regression.
>
> In this case, I am happy not having to commit a test case. If the
> subreg coalescer starts acting up, I am sure we'll catch it with our
> regular testing.

But didn't the testcase exist because nothing else caught the problem?

> > Perhaps we need a second-tier "make check" for larger testcases that
> > only
> > gets run once a week or maybe just when llvm-test fails.  Would that
> > be
> > helpful?
>
> I don't think that adds a lot of value. If it's not the default,
> people will forget about it.

It adds value in that one doesn't have to go and reduce bzip again.  Larger
test runs spread over a period of a few days to a week is common practice.  I
don't see why LLVM wouldn't benefit as well.

                                 -Dave