[Python-Dev] Time for the yearly list.append() panic (original) (raw)

Tim Peters tim.one@home.com
Fri, 25 May 2001 18:11:43 -0400

[Tim]

Long-term we should teach PyMalloc about Python's realloc() abuses and craft a cooperative solution.

[MAL]

That's what I think too. There's really not much point in trying to work around poor malloc() implementations when we've already got the cure built into Python...

The point here is that a simple localized patch could kill off a Frequently Irritating Complaint without further ado: on my personal cost/benefit scale, it's all I can afford to do now. PyMalloc likely won't solve it as-is x-platform, without new work to accommodate extreme realloc() abuse.

I just wish Vladimir would resurface again to complete his great work

I'd like him to come back even if he doesn't .

(AFAIK, pymalloc still has problems with threads).

It has lock macros that haven't been #define'd to do anything yet. But part of the potential value of the Python core using its own allocator is to exploit the global interpreter lock to not lock in the allocator. Messy issues. Python should grow a cheaper platform-specific flavor of internal lock too. (Jeremy pointed out some code the other day that jumps through hoops to simulate a reentrant lock on top of a Python lock; an irony is that on Windows, the native lock is reentrant already, and Python jumps through hoops to make it act as if it weren't )