Doug,
I found a way to avoid breaking FTP. I can just disable suspending the
socket transport when mSelectFlags includes PR_POLL_READ. This should
only effect FTP and would result in the CPU-hogging behavior of FTP as
it is now.
Darin
Doug Turner wrote:
> yeah, see: <http://bugzilla.mozilla.org/show_bug.cgi?id=65220>
>
> http://bugzilla.mozilla.org/show_bug.cgi?id=65220
>
> The problem is that nsIChannel's are not bidirectional (eg you can not
> read and write at the same time even if you are lockstepping). In FTP,
> I do a AsyncRead while doing a AsyncWrite. Things should be okay, but
> they are not. The write socket transport is stuck in the select list
> resulting in the sockettransport going into a busy waiting state. I am
> working on fixing nsIChannel and all callers and should be done with it
> in a week or so. (I estimated a week, but I think i may run over
> because some of the terrible usage patterns I am finding).
>
> Furthermore, Darin will be landing some AsyncWrite (and friends) changes
> today which will definitely break FTP. This breakage will be fixed my
> may nsIChannel changes.
>
> We should probably disable the ftp test on tinderbox for the next week
> or so.
>
> CC-ing the netlib newsgroup.
>
>
>
> [EMAIL PROTECTED] wrote:
>
> That was doug.
>
> >
> > These timeouts starting with a major ftp checkin. Don't remember
> who did
> > it, but it was some kind of minor carpool.
> > Due to the slowdown, the corresponding bug got reopened, but
> obviously
> > not fixed :-(.
> >
> > Axel
>
