[gdal-dev] KyngChaos Frameworks: GDAL's numpy vs Enthought's numpy

[Christopher Barker]

Jason Roberts wrote:
> I recall a recent discussion regarding an "ABI breakage" with numpy 1.4.

yup -- that's it. IN short:

numpy 1.4.0 was released without the devs realizing that the ABI had 
broken (or hadn't realized how many problems it would cause). It was 
decided to pull that release, and instead release a 1.4.1 that didn't 
break the ABI, and later on release one that did -- maybe 2.0. I think 
1.4.1 is in "release candidate" stage as I write.

Given all that, I'm surprised that Entought has been released with 1.4.0 
-- are you sure it came that way?

>> GDAL python *should* be able to use a later version of numpy than it was
> built for, it's only the python side of the interface that it cares about

really? I thought it build numy arrays directly in C code? Or is is 
using the "Array Protocal", in whihc case you're right, it shouldn't 
care about the ABI.

Which reminds me -- this is actually a good argument fo why we should 
all use the PEP 3118 extended buffer protocol, rather than the numpy API 
directly, for extension modules ( at least when older python's don't 
need to be supported)

Is it possible that when you use GDAL with Enthought that you've loaded 
two incompatible numpys?

> If you are including the GDAL framework python folder in the Enthought
> python path, then numpy 1.3 is also available from there.

Sure, but if you are using other numpy-using packages together with 
GDAL, you'll need to on;ly use one numpy (well, you could probably 
kludge around it, but it would be ugly...)

I'd wait for/ask Enthought to release a numpy1.4.1 version of EPD, or 
use EPD 5.1

-Chris




-- 
Christopher Barker, Ph.D.
Oceanographer

Emergency Response Division
NOAA/NOS/OR&R            (206) 526-6959   voice
7600 Sand Point Way NE   (206) 526-6329   fax
Seattle, WA  98115       (206) 526-6317   main reception

Chris.Barker at noaa.gov