[pypy-dev] Two bugs in PyPy 1.2

[Armin Rigo]

Hi Maciek, hi Fredrik,

On Fri, Mar 12, 2010 at 11:23:24PM -0700, Maciej Fijalkowski wrote:
> Actually I fixed the one about fatal rpython assertion. There is also
> a problem, which seems to be windows only that I can't reproduce.
> 
> from mpmath import fp
> for i in range(1000): fp.zeta(0.5+37235j)

I fixed this in r72263.  The issue was hard to pinpoint because it was
caused by a fast-path malloc: on the slow path (i.e. rarely), it would
call random C code without saving the XMM registers.  That's wrong:
there is -- on Windows only? -- a path through that C code that trashes
the XMM registers.


A bientot,

Armin.