notting at redhat.com
Wed Nov 10 12:09:41 PST 2010
Lennart Poettering (mzkqt at 0pointer.de) said:
> > > It isn't. Most distros clean up that dir already periodically, and with
> > > systemd we will unify that and make it a core part of the basic
> > > system. systemd will apply the same cleanup rules to /var/run/user as to
> > > /tmp.
> > Would we also set $TMPDIR to $XDG_RUNTIME_DIR for legacy applications
> > to use?
> Mhmm, I don't think we should.
> TMPDIR influences tmpnam(), however that function kinda expects a common
> /tmp namespace anyway, so redirecting this is not necessary.
> I think we shouldn't break assumptions about $TMPDIR, for example that
> it might be useful for communicating with other people.
Given that it's settable per-user, I'm not sure why you would
rely on that aspect.
> It's actually one of the things where I believe polyinstantiated /tmp is
> actually going wrong: $TMP is shared between users, which has many
> negative aspects, but also allows users to communicate with each other.
The reason I ask is that I'd expect that apps would code their usage such
that they'd do fallback from XDG_RUNTIME_DIR, to TMPDIR, to /tmp/<whatever>,
and if we set TMPDIR to XDG_RUNTIME_DIR, the 'right' thing happens to the
apps without patching. (For example, orbit and pulseaudio sockets are
currently in $TMPDIR.)
More information about the xdg