Validate failures

[Simon Marlow]

On 28/02/2012 13:07, Ian Lynagh wrote:
> On Tue, Feb 28, 2012 at 10:44:35AM +0000, Simon Marlow wrote:
>> On 28/02/2012 09:03, Simon Marlow wrote:
>>> The safeHaskell failures disappeared on my second validate run, so I'm
>>> not sure what happened (I had pulled into the tree first, and there were
>>> no patches between the two runs that should have affected this). Perhaps
>>> there's some missing cleaning somewhere? Anyway I wouldn't worry about
>>> it unless the failures come back.
>>
>> I just remembered: my second validate was with --fast, so it seems
>> the safeHaskell failures only show up without --fast.  Probably some
>> bad interaction with BINARY_DIST=YES?
>
> It's because ghc has a different program name (ghc-stage2 vs ghc)
> depending on whether it is in in-tree GHC or not.
>
> The testsuite would normalise this on stderr, but ghc/InteractiveUI.hs
> showException is printing the exception on stdout. I didn't get as far
> as working out whether that's the right thing for it to do or not.

I think our policy in GHCi is to print everything to stdout, except for 
program output that is explicitly sent to stderr of course.  It doesn't 
seem to make sense for an interactive session to send some output to 
stdout and some to stderr, risking strange interleaving due to buffering 
and whatnot.

Cheers,
	Simon