[SciPy-dev] scipy testing results

[eric]

Hey Fernando,

> Boy, wasn't that fun and simple :)

Yeah, the CVS is not for the faint of heart.  Things will simplify some, but,
until 0.2 is released, I doubt the build process will be very well documented.

>
> The atlas compilation took a solid 1/2 hour, but it's done. I didn't build
> lapack from scratch, grabbed rpms for that. But I followed the 'completeness'
> instructions for atlas/lapack.
>
> The build_instructions web page should definitely mention that your best bet
> of getting things to work is just to bite the bullet and build atlas
> yourself. The existing prebuilt binaries are just too old/incomplete to be
> useful, and you end up just wasting time trying to use them.
>
> Now things are working, sort of. I can import scipy, but scipy.test() reports
> 5 failures.  I pasted the report at the end in case it's useful for you guys.

The errors look like a combination of the ones I'm seeing and Pearu is seeing.
We'll work out the IEEE ones later.  The linear algebra ones will hopefully get
worked out soon -- or fixed with linalg2 which I still haven't spent any time
on...

> I fully realize that right now we're in a state of flow, but for 'end users'
> we need to consider some centralized packaging for scipy. At least have local
> builds available of lapack,atlas, static fft would be a plus. I think
> 'outsiders' would be quite put off by the current level of difficulty of the
> installation.

If you'll write up the process you followed, I'll post it to the website as
hints on building from the CVS.

I'm glad you got it all at least close to working.

see ya,
eric