On Wed, Sep 9, 2009 at 23:56, Nick Coghlan
Thomas Wouters wrote:
Your idea of making this an API called a 'fork lock' or something sounds good (though I think it needs a better name. PyBeforeFork & PyAfterFork?). The subprocess module, for example, disables garbage collection before forking and restores it afterwards to avoid http://bugs.python.org/issue1336. That type of thing could also be done in such a function.
Unfortunately it's rather hard to make those functions work correctly with the current API -- we can't provide functions you can just use as arguments to pthread_atfork because the global interpreter lock is not re-entrant and we have no way of testing whether the current thread holds the GIL.
I thought "make sure I have the GIL, either by already having it or waiting for it if I don't yet have it" was the entire point of the PyGILState_Ensure() API? 
Hm, yeah. For some reason I was certain it was inappropriate, back when I
was trying to create a pthread_atfork-friendly set of functions. At the time
I was also hip-deep in the awfulness of Python/thread*.c and its unsafe
punning and unwarranted assumptions, so I may have overreacted. I added this
as a feature-request issue ( http://bugs.python.org/issue6923 ) and will
look at it some more.
In the mean time, I fixed the biggest source of issues by checking in the
change to make at least calls to fork() made by Python be safe, also
backported to 2.6.