[Python-Dev] xreadlines : readlines :: xrange : range

[Guido van Rossum]

> So it looks like the ALLOW_THREADS should be moved out of the for
> loop.  This produced no measureable performance difference on Solaris;
> I'll leave it to GvR to try it on Linux.  I wonder if FreeBSD has some
> unusually slow thread operation?

I kind of doubt that it's Py_ALLOW_THREADS -- it's in the outer loop,
which typically only gets executed once.  It only goes around a second
time when the line is longer than the initial buffer.  We could tweak
the initial buffer size (currently 100, with increments of 1000).

--Guido van Rossum (home page: http://www.python.org/~guido/)