"Ralf S. Engelschall" wrote:
> > It appears that the intent of setting a file descriptor to non-blocking
> > mode via pth_fdmode() is to do for the thread what normal non-blocking
> > (without pth involved) does for a process.
> >
> > Is this the intent? If so, that's great.. but then there are a few bugs
> > in the implementation.
>
> Yes, that is the intent.
>
> > While pth_read() and pth_write() have been modified to properly handle
> > the PTH_FDMODE_NONBLOCK case, this still needs to be done for other
> > blocking syscalls, e.g. pth_connect() and pth_accept() to name two.
> > They will block the thread even if the fdmode is PTH_FDMODE_NONBLOCK.
>
> Hmmm... yes, that's correct. My fault, I totally forgot these syscalls.
>
> > I'd really like to see this fixed, and would be willing to supply
> > patches if that would help.
>
> Sure, it certainly helps. Feel free to post a patch.
Thanks... it's attached (but untested).
Thanks,
-Archie
__________________________________________________________________________
Archie Cobbs * Packet Design * http://www.packetdesign.com
--- pth_high.c.orig Sat Mar 24 06:49:58 2001
+++ pth_high.c Thu Dec 6 09:53:30 2001
@@ -496,7 +496,7 @@
errno_shield { pth_fdmode(s, fdmode); }
/* when it is still on progress wait until socket is really writeable */
- if (rv == -1 && errno == EINPROGRESS) {
+ if (rv == -1 && errno == EINPROGRESS && fdmode != PTH_FDMODE_NONBLOCK) {
ev = pth_event(PTH_EVENT_FD|PTH_UNTIL_FD_WRITEABLE|PTH_MODE_STATIC, &ev_key,
s);
if (ev_extra != NULL)
pth_event_concat(ev, ev_extra, NULL);
@@ -541,7 +541,8 @@
/* poll socket via accept */
ev = NULL;
while ((rv = pth_sc(accept)(s, addr, addrlen)) == -1
- && (errno == EAGAIN || errno == EWOULDBLOCK)) {
+ && (errno == EAGAIN || errno == EWOULDBLOCK)
+ && fdmode != PTH_FDMODE_NONBLOCK) {
/* do lazy event allocation */
if (ev == NULL) {
ev = pth_event(PTH_EVENT_FD|PTH_UNTIL_FD_READABLE|PTH_MODE_STATIC,
&ev_key, s);
