PATCH to valarray

[Benjamin Kosnik]

> | You need to start checking in test cases for this stuff...
> 
> Agreed.  

yeah... the math parts of the testsuite are much weaker than I'd like
the last time I checked.

on a side note, have you thought about what it would take to merge
libstdc++-v3/libmath to toplevel libgcc-math? I was very pleased to see
the top-level libgccm/libgcc-math development: I think it's long
overdue.

could this allow libstdc++ math config to be much simpler?

> The difference between this and the previous version is essentially
> performance wise (not semantics).  For example, the follow up patch
> is to call calloc() -- instead of malloc + memset -- for PODs and
> malloc() for non-POD.   I just don't know how to test those reliably
> (it is close to the question I asked recently about aligned copy and
> memcpy.) 

replacement version of malloc that sets a boolean when used. See some
of the allocator tests, which test to see if new is used.

this isn't really a regression test, but intead a test of current
libstdc++ behavior, which has proven to be a useful kind of test for
when behavior is changed later....

best,
benjamin