Re: Patch to fix complaints from valgrind about make check for randist

["Jonathan G. Underwood" <j dot underwood at open dot ac dot uk>]

Brian Gough wrote:
> Jonathan G. Underwood writes:
>  > Another (trivial) patch to fix valgrind errors for make check in randist.
>
> Thanks, I'll add those in.  Most of the errors I've seen with valgrind
> were related to "long double". I decided that it didn't seem to handle
> that type correctly as the same code doesn't generate any errors with
> other types.

Ok, so I'm not going insane :) I've just spent a couple of days looking 
hard at that - it generates a number of complaints for the vector and 
matrix code for long double complex vectors and matrices. I looked hard 
at the code and tried a number of things to debug it, but can't see what 
is causing it. Interestingly, different errors arise if one uses no 
optimisation, but again it's all with the long double stuff. I did start 
to wonder if it had to do with alignment on boundaries for long doubles, 
though (since the gsl_complex structures used for the vectors and 
matrices seem to be generating the issues). But, I haven't investigated 
this further yet. It does concern me though, as I've never previously 
known valgrind to generate errors that were worth ignoring.

For reference I am using fedora core 4, so gcc 4.0.1, valgrind 2.4.0, 
glibc 2.3.5. I may try valgrind 3.0 to see if anything changes there.

Jonathan.