[Python-Dev] Twisted and Python 2.5a0r43587

[Thomas Wouters]

On 4/4/06, "Martin v. Löwis" <martin at v.loewis.de> wrote:
>
> Thomas Wouters wrote:
> > And so I could.  test_banana.CananaTestCase.testCrashNegativeLong
> > crashes, because it calls PyString_AsStringAndSize() with an int-ptr as
> > second argument (an adjacent ptr variable becomes garbage.) That's
> > certainly a problem with the Py_ssize_t change. Martin, aren't all
> > output variables (or ptr-variables, rather) supposed to be controlled by
> > the 'PY_SSIZE_T_CLEAN' #define? People aren't going to notice their
> > compiler warnings; I know I didn't :)
>
> No: this is the discussion I had with MAL. You have to watch for
> compiler warnings talking about incorrect pointer types. These
> days, you can apply Fredrik's checker to find out that you are
> using functions that output Py_ssize_t.
>
> Ignoring the warning might cause crashes on 64-bit machines. On
> 32-bit machines, there should be any negative consequence.


I assume you meant "shouldn't be any negative consequences" there ;) I
thought this over during dinner (mmmm, curry) and I agree that we shouldn't
make PY_SSIZE_T_CLEAN change output variables that the compiler can catch.
Too much effort, and in the end it'll just cause extensions compiled for
64-bit python to not use 64-bit values 'silently' (although eventually
something'll get truncated and people will get confused.) This at least
gives a compile-time hint that, on 64-bit platforms, things aren't quite
right.

It's just too bad that it's an easily overlooked hint, and that you can't
get that hint when compiling for 32-bit platforms. Extension writers with
64-bit hardware access and the desire to compile, let alone test, on said
hardware is still rare. If we did make PY_SSIZE_T_CLEAN adjust all
Py_ssize_t*-using functions (which would mean a *lot* of API duplication),
we could add a #warning for every use of the old API calls, so everyone sees
it, even on 32-bit platforms. The warning should contain a pointer to a
document describing how to conveniently support both Python 2.5+ and 2.4-,
though. Too much work for alpha1, in any case, and probably too much work
period. The webpage should be made, though, if just to refer to in the
release notes.

--
Thomas Wouters <thomas at python.org>

Hi! I'm a .signature virus! copy me into your .signature file to help me
spread!
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.python.org/pipermail/python-dev/attachments/20060404/80112663/attachment-0001.htm