Check fails on cygwin, gcc 4.2.1

[Michael Abshoff]

Oliver Jennrich wrote:
> Michael Abshoff <Michael.Abshoff at fsmath.mathematik.uni-dortmund.de>
> writes:
>
>> Oliver Jennrich wrote:
>>>

Hello,

>>
>> Hello Oliver
>>
>> Just to be clear: the gmp version is 4.2.1, not 4.1.1, so my suggestion
>> to
>> try the current release is moot. The gcc you used is gcc 4.1.1 and your
>> Cygwin install is roughly 7 months old. Could you try with current
>> Cygwin?
>
> It is not quite clear what you mean with a current cygwin.  2007-01-31
> is the date of the most recent cygwin.dll, and this, I'm afraid, the
> most recent available.

That is correct regarind the latest official release, but you can install
daily snapshots. Because the last release is 6 months it might have been
worth a try.

> gcc of course is *not* the actual cygwin compiler, as this is still on
> 3.4.4.
>
> Anyway, I tried to find the source of the problem and it seems that
>
> int memcmp(const void *S1, const void *S2, size_t N);
>
> returns a '1' even if N=0, which should not happen.
>
> The problem goes away, however, if I turn down the level of
> optimization, i.e. calling gcc with -O3 reproduces the problem,
> calling it with -O2 shows no problem at all, so I assume a problem
> with loop unrolling.
>

Intersting.

> So it seems to be a bug (or at least a very odd behaviour) of gcc, so
> I will report it there...
>
>

Cheers,

Michael

> --
> Oliver Jennrich                                    Tel:  +31-71-565-6074
> LISA Project Scientist
> ESA/ESTEC, Keplerlaan 1, Postbus 299, 2200 AG Noordwijk, The Netherlands
>