[uClibc]Re: concerning getcwd() on /coda

[Ivan Popov]

On Sun, 28 Oct 2001, Erik Andersen wrote:

> that.  So on coda, the inode is some funky non-32bit size?

Not really. The sizes are ok, and it implies for the library (why??) that
it can skip slow search while staying on the same device.
It does not work with coda - you can get different inode numbers in a
directory entry and from stat() of that entry, like passing mount
points - but without device number change.

I suppose dfs may lead to similar effects, when it will be available for
Linux.

> Hmm.  As I think about it, what we _really_ should do is use the
> __NR_getcwd syscall if it is available, then fall back on the
> slow method iff the __NR_getcwd syscall is not available or
> fails.

It would fix the problem with coda, of course.

> Would you mind adjusting your patch to include that bit?

I can try, but it has to wait. On the other side, the current patch is
working properly, we haven't lost much of performance - compared to using
a syscall we are slow anyway :)

>  -Erik

See you,
--
Ivan