[Gnu-arch-users] Re: tla on nfs only

[Stefan Monnier]

>> I don't understand what you're talking about.
>> We need:
>> - one stat per work-tree file.
>> - compare this result with a known value (which can come from the
>> work-tree's inode-sigs or from the rev-lib).

> Those are the requirements for determining whether the file's contents
> have changed.

> For determining whether the file's metadata (permissions, file type) has
> changed, the requirements are:
> - one stat of the file in the working directory
> - one stat of the file in the reference tree

No the requirements are the same as for the content: you can check the
meta-data against a stat of the revlib, or you can check it against the
inode-sigs of the worktree.  Admittedly the worktree's inode-sigs don't
include this part of the meta-data yet, but it can't be that hard to
change it.

> Since the reference tree stat is necessary for other reasons,
> it's very cheap.

But the revlib stat is *not* necessary ;-)


        Stefan


PS: All those performance issues could be made less important if tla
    allowed us to work on subtrees.  The <limit> argument to `commit' and
    `changes' is a start but the code does not take advantage of it to only
    look at the relevant files and it only works in some limited cases.
    After all CVS is not really faster when working on whole trees, but
    I almost never use that way.