Ticket #807 (closed issue: wontfix)
Python2.5 + 64-bit linux incompatible with Numeric
Reported by: | spascoe | Owned by: | spascoe |
---|---|---|---|
Priority: | critical | Milestone: | PROD Step2 |
Component: | community | Version: | |
Keywords: | Cc: |
Description (last modified by selatham) (diff)
Python2.5 introduces a new C-type (Py_ssize_t) that is 8 bytes on 64-bit architectures. On Python2.4 and below the C-type int was used instead (4-bytes). As a result both cdat_lite and Numeric will not work on a combination of Python2.5 and 64-bit linux. For instance,
>>> import Numeric >>> a = Numeric.arrayrange(10) >>> a[1:] zeros((0,), 'l')
This isn't what we expect. CDAT crashes completely when accessing a variable. Note this even effects the OpenSUSE 10.2 RPM python-numeric-24.2-36.rpm. We can either fix the source of both cdat_lite and Numeric (not difficult, about 2 hours work) and rely on a custom version of Numeric or we can target Python2.4.
Change History
comment:2 Changed 14 years ago by mggr
- Owner changed from mggr to spascoe
Handing ownership of the ticket to Stephen as he has a much bigger clue than me on this one and is likely to do better on the CDAT mail ;) (thanks for the heads-up Phil!)
comment:4 Changed 14 years ago by spascoe
This bug has been reported to the CDAT mailing list. See this sourceforge page.
I would like to see a message on the cdat mailing list making it clear that CDAT is irreversibly broken at python 2.5 unless it uses numpy (and pointing out that python 2.5 is the default on modern ubuntu).