testsuite results

[Daniel Fischer]

Running the testsuite with today's HEAD (perf build, but without profiling 
to keep time bearable) resulted in:

====================
OVERALL SUMMARY for test run started at Do 12. Mai 13:34:13 CEST 2011
    2765 total tests, which gave rise to
    9300 test cases, of which
       0 caused framework failures
    1587 were skipped

    7467 expected passes
     229 expected failures
       9 unexpected passes
       8 unexpected failures
====================

Pretty cool, I can't remember having so few unexpected failures before.

====================
Unexpected failures:
   T5084(normal)

That's  the compiler not complaining about an INLINE-pragma on a class 
method without default implementation. Patch is in ghc-generics branch, not 
yet in master, according to #5084. Anyway it's nothing serious (was a 
feature request, not a bug).

   dph-diophantine-opt(normal,threaded1,threaded2)

These are due to a missing Show instance for [:Int:], a library issue.

   dph-words-opt(normal)

Fails with "dph-words-opt: libraries/vector/Data/Vector/Generic.hs:369 
(slice): invalid slice (1,2,2)".
No idea whether that's a library or a compiler issue.

   hpc_markup_multi_001(normal)
   hpc_markup_multi_002(normal)
   hpc_markup_multi_003(normal)

Those are due to hpc looking in the wrong directory for the tix files, 
patch exists, but is not yet in the master branch, according to #5069.

So, of the eight unexpected failures, six are due to trivia (they *might* 
fail for other causes when the trivia are fixed, but there's no reason to 
expect that), one is a feature request whose test reached testsuite/master 
before the implementation reached ghc/master and only one may (but need 
not) indicate a compiler bug at present, that's rather awesome.

====================

Unexpected passes:
   mc01(hpc,ghci)
   mc06(hpc,ghci)
   mc08(hpc,ghci)
   mc11(hpc)
   mc16(hpc)
   mc18(hpc)

All these involve the new MonadComprehensions extension, they're expected 
to work and do so for the normal and optasm ways, maybe they should also be 
expected to work for hpc and ghci.


Additionally, sometimes conc016(threaded2) passes unexpectedly; which 
thread first gets its exception to the other one is impossible to predict:

-- NB. this test is delicate since 6.14, because throwTo is now always
-- interruptible, so the main thread's killThread can be legitimately
-- interrupted by the child thread's killThread, rather than the other
-- way around.  This happens because the child thread is running on
-- another processor, so the main thread's throwTo is blocked waiting
-- for a response, and while waiting it is interruptible.


Summing up: Yay!