On Tue, 16 Nov 2004 17:34:54 -0800, Dan Fandrich <dan_at_coneharvesters.com> wrote:
> This looks like a clever approach to maintaining backward compatibility while
> still fixing the problem in the common cases. I have one suggestion: rather
> than asserting the fd values, which will terminate the program if it fails,
> why not propagate the error to the user as a CURLE_READ_ERROR or a
> new error message? That gives the user a chance to retry the operation
> after freeing some handles.

That's a good idea, but I'm not sure how to best accomplish that.
Perhaps simply returning -1 from Curl_select() will do the trick?
 
> Also, this part of the patch looks suspicious:
> [...]
> Removing the brackets causes writedone to be set unconditionally.

Good spot. That is a mistake on my part.

-- 
David Phillips <david_at_acz.org>
http://david.acz.org/