[cfe-commits] Fix search path for clang on latest DragonFly releases [revised patches]

[John Marino]

On 12/20/2012 19:30, Jordan Rose wrote:
>
> I think what Rafael is trying to explain is that any driver patches need
> to have /platform-independent/ tests that /everyone/ will run as part of
> check-clang. Of course this doesn't prove that DragonFly is actually
> working, or that it's not broken in some other way, but it does ensure
> that no one breaks the changes you are making /right now./
> /
> /
> I see you have one changed test file, but it's not testing each of the
> different changes you're making. It would be nice, for example, to
> actually show that clang correctly picks between gcc47 and gcc44
> depending on what's in the filesystem; take a look at the other
> sysroot-based tests for that.

It's testing everything I could think of outside of that, and it's a lot 
better than it was before.  I don't know enough about the test framework 
to add conditions but really it's very simple logic based on the 
existence of the /usr/lib/gcc47 directory.

You say "one file" but what other test files would there be?  every 
routine that I changed has a dragonfly namespace, so the only change 
would be dragonfly.c that I can tell.  In reality, the current 
dragonfly.c is so loose that even changing to gcc47 path passed.  It 
didn't break until I added the other specs.

> As you pointed out, we have a DragonFly test in our test suite...and it
> was totally useless because it didn't test the right things, and thus we
> on trunk didn't know that DragonFly was completely broken. Obviously we
> can't test things end-to-end without a DragonFly bot, but all this
> configuration stuff can at least be in the test suite so we know what's
> expected/necessary.

I think I met the spirit of that, but I am obviously only barely 
familiar with the framework.  I gave it my best effort, and the proposal 
is definitely better than the current state.

John