I like it.  We probably need a new timeout event.

Some comments:

- the property should be a guint64, and in nanoseconds.  This is to
conform with other gstreamer properties.

- select() can return EINTR for a number of completely valid reasons.
 It's not necessarily an element error

I have been working on a similar patch for fdsink, but have been
unable to get select to work the way I want. fdsink seams to be a
"chain element" so i wonder if a solution as in fdsrc will work?

_chain() is the equivalent of _get() for sink pads.  A source element
doesn't have a sink pad, so it's irrelevant.

Created attachment 20730 [details] [review]
Diff against CVS HEAD around 2003-07-23

Just add patch as attachment, as i may move it from the remote server.

I'd be happy to apply it to HEAD, if you fix the things I commented
about previously.

Mattias: ping ?

Pong.
Do you think the patch worth including? i can probably fix the things
David commented on, but currently im not sure if have the disk space
to compile the source :)

Created attachment 24389 [details] [review]
New patch against HEAD, includes uint64 for timeout value and EINTR handling.

Timeout is now in nanoseconds.
Does everything look ok?

Also, im curious about one thing:
If I for example set the timeout value to 1 second useing UL-postfix:
g_object_set (G_OBJECT (src), "timeout", 1000000000UL, NULL);
I get this warning about property "fdsrc", not "timeout", weird.
(process:14439): GLib-GObject-WARNING **: gobject.c:882: object class
`GstFdSrc' has no property named `fdsrc'
With ULL-postfix i get no warning, in both cases the timeout value
seams to get set.

in that case, you're passing an unsigned long, instead of a guint64_t,
which probably translates to a unsigned long long on your machine.  You
should use G_GINT64_CONSTANT (1000000000), or less portably,
1000000000ULL.

Patch is applied, with g_error() converted to GST_ELEMENT_ERROR(). 
Thanks.