[Python-Dev] stack check on Unix: any suggestions?

[Martin v. Loewis]

> Does anyone have suggestions for how to detect unbounded recursion
> in the Python core on Unix platforms?

I just submitted patch 101352, at

http://sourceforge.net/patch/?func=detailpatch&patch_id=101352&group_id=5470

This patch works on the realistic assumption that reliable stack usage
is not available through getrusage on most systems, so it uses an
estimate instead. The upper stack boundary is determined on thread
creation; the lower stack boundary inside the check. This must allow
for initial stack frames (main, _entry, etc), and for pages allocated
by on the stack by the system. At least on Linux, argv and env pages
count towards the stack limit.

If some systems are known to return good results from getrusage, that
should be used instead.

I have tested this patch on a Linux box to detect recursion in both
the example of bug 112943, as well as the foo() recursion; the latter
would crash with stock CVS python only when I reduced the stack limit
from 8MB to 1MB.

Since the patch uses a heuristic to determine stack exhaustion, it is
probably possible to find cases where it does not work. I.e. it might
diagnose exhaustion, where it could run somewhat longer (rather,
deeper), or it fails to diagnose exhaustion when it is really out of
stack. It is also likely that there are better heuristics. Overall, I
believe this patch is an improvement.

While this patch claims to support all of Unix, it only works where
getrlimit(RLIMIT_STACK) works. Unix(tm) does guarantee this API; it
should work on *BSD and many other Unices as well.

Comments?

Martin