2009/1/15 Daniel Stenberg <daniel_at_haxx.se>:
> The only little flaw I could detect is that it treats a strdup() failure as
> "no proxy" which I think is bad. It should rather completely fail the entire
> operation (yeah, that forces the function to be able to return that
> info...).

It turned out to be a REALLY large pain to find a way to return error
information from that function, so I switched to an implementation
that doesn't require memory allocation. This would have been easier if
strspn and strcspn were guaranteed to be available, but I didn't want
to bother with using automake to determine if they were or not, so I
implemented it using strchr and loops. As a special bonus, I synced
the patch up to the current cvs head, which had one conflict with the
previous version...
-Craig