[Python-Dev] Re: [Python-checkins] CVS: python/dist/src/Lib/test test_dumbdbm.py,NONE,1.1

Thomas Wouters thomas@xs4all.net
Sat, 20 Jan 2001 19:37:10 +0100

On Sat, Jan 20, 2001 at 12:00:05PM -0500, Guido van Rossum wrote:

> > Well, there is one more issue, which we can't fix terribly easy: test_fcntl
> > tries to flock() the file. flock() doesn't work on all filesystems (like
> > NFS) :P If we cared a lot, we could try several alternatives (current dir,
> > /tmp, /var/tmp) in the specific case of flock, but personally I don't want to
> > bother, and real sysadmins (who should care about the test failure) are
> > more likely to build Python on a local disk than in their NFS-mounted
> > homedirectory. At least that's how we do it :-)

> These days, I would think that it's a pretty sure bet that the
> system's tmp directory is not on NFS.  Then we could just use
> tempfile.mktemp() in that module, right?  Or does the /tmp filesystem
> on Linux (which AFAIK is a RAM disk implemented in virtual memory so
> it uses swap space when it runs out of RAM) not support locking?

Actually, most Linux distributions don't care enough about /tmp to make it a
RAM-based filesystem. At least Debian and RedHat don't :) (There's a good
reason for that: Linux's disk-data cache rocks if you have enough RAM, so
there's no real gain in using a ramdisk) BSDI does (optionally) have such a
/tmp, and probably the other BSD derived systems as well. But that doesn't
mean it doesn't support locking, so that's not a real excuse.

But like I said, I don't care enough to worry about it. I'll look at it
before alpha2.

Thomas Wouters <thomas@xs4all.net>

Hi! I'm a .signature virus! copy me into your .signature file to help me spread!