[Python-Dev] reference counting in Py3K

[Fernando Perez]

Josiah Carlson wrote:

> Fernando Perez <fperez.net at gmail.com> wrote:

>> Would you care to elaborate on the reasons behind the 'ick'?  I'm a big fan
>> of weave.inline and have used it very successfully for my own needs, so I'm
>> genuinely curious (as I tend to teach its use, I like to know of potential
>> problems I may not have seen).
> 
> 1. Mixing multiple languages in a single source file is bad form, yet it
> seems to be encouraged in weave.inline and other such packages (it
> becomes a big deal when the handful of Python becomes 20+ lines of C).

Agreed.  I only use inline with explicit C strings for very short stuff, and
typically use a little load_C_snippet() utility I wrote.  That lets me keep
the C sources in real C files, with proper syntax highlighting in Xemacs and
whatnot.

[... summary of weave problems]

> Agreed.  I admit that some of my issues would likely be lesser if I were
> to start to use inline now, with additional experience with such things.
> But with a few thousand lines of Pyrex and C working right now, I'm hard
> pressed to convince anyone (including myself) that such a switch is
> worthwhile.

Thanks for your input.  I certainly wasn't trying to suggest you change, I was
just curious about your experiences.  If you ever see this again, specific
feedback on the scipy list would be very welcome.  While I'm not 'officially'
a scipy developer, I care enough about weave that occasionally I dig in and go
in bugfixing expeditions.  With proper bug reports we could improve a system
which I think has a place (especially for scientific computing, with the Blitz
support for arrays, which gives Numpy-like arrays in C++).  I don't see weave
as a competitor to pyrex, but rather as an alternate tool which can be
excellent in certain contexts, and which I'd like to see improve whre
possible.

Regards,

f