Re: Creating a tag

[law at redhat dot com]

 In message <Pine.LNX.4.33.0206192012190.16343-100000@kern.srcf.societies.cam.a
c.uk>, "Joseph S. Myers" writes:
 > The logs from the generation of the last snapshot
 > http://gcc.gnu.org/ml/gccadmin/2002-q2/msg00174.html
 > show it taking five hours to delete and recreate the gcc_latest_snapshot 
 > tag.
It's certainly expensive.

 > Could anoncvs access to the tree on gcc.gnu.org be disabled (without
 > breaking cvsweb)
CVSweb works by reading the RCS files directly, it's independent of anoncvs

 > (for example, by removing anoncvs from the gcc group so
 > it can't create lock files, and chowning to someone else the directories
 > in the gcc tree owned by anoncvs)?
I'd prefer a less invasive route -- chowning the directories will create a
lot of IO traffic that you really need to avoid.  And I don't like the
idea of stomping on the group file automatically in some script.

It would be better to somehow control anoncvs at connection time; for
example, the wrapper could check a toplevel lock file.

 > What's the status of the new machine that was being planned some time ago,
 > and will it have enough memory that the whole CVS tree can be expected to
 > stay cached?  (I don't know whether this is relevant to the speed issue,
 > it just seems plausible since tag operations need to write to every ,v
 > file in the tree, and operations on branches need to read them all.)
I believe it's still being set up -- Chris has lots of other responsibilities
that come before setting up the new sourceware/gcc machine.

jeff