decorators vs GIL

Neil Hodgson nhodgson at bigpond.net.au
Sun Aug 8 03:12:47 CEST 2004


mudd at vex.net:

> I'm guessing the GIL is beyond any of us.  If it was easy, or even
> possible, someone would have already eliminated it.

   Greg Stein did produce a free threading version of Python.
http://python.tpnet.pl/contrib-09-Dec-1999/System/threading.README
   It performed much slower than GIL based Python, at something like 60% of
the GIL version with one processor. This is the expected result when
replacing a large scale lock with many finer grained locks: there is more
locking overhead. It is possible that techniques can be found to reduce the
free threading cost so it would be great for people to continue working on
this.
http://mail.python.org/pipermail/python-dev/2001-August/017099.html
   The barrier to keeping this code in standard Python as an option is the
amount of maintenance effort it represents both for itself and its impact on
the GIL version.

   Neil





More information about the Python-list mailing list