[EMAIL PROTECTED] wrote:
Craig Berry wrote:

!To really stress test this with multiple threads going after the same
!directory stream I'd need an SMP machine to test on, which I don't have.
!But I think we are at least somewhat safer than we were before.

Can you specify which thread option to configure.com to use?

$ @configure -"Duseithreads"


and answer yes to the question "Enable multiple kernel threads and upcalls?"

Also: has anyone made a tar ball out of [EMAIL PROTECTED]
(or something later that incorporates your patch) that is
accessible via either anon ftp or http?

I don't think it's in maintperl yet, and I haven't seen a bleadperl snapshot in a long time. I'll try to set you up with something off-list.

I could give it a try on an SMP machine.
> Is there any ad hoc testing to try?

There would be value in simply running the standard test suite with the
above configuration, but offhand I don't see anything in it that would
really stress readdir_r. To do that you'd need to write something that
does an opendir, then starts multiple threads, each of which does
readdir on the same directory stream. Doing this on a fast mirrored disk
or RAM disk would help factor out I/O completion as a serializing influence.



Reply via email to