> 1 - Curl_sendf() sends data non-blocking and it may very well not send the
> entire buffer at once. I suspect you can trigger this problem by sending an
> awful lot of requests with very long paths. A possible solution to this would
> be to use the Curl_add_buffer_init(), Curl_add_bufferf() and
> Curl_add_buffer_send() which is done by the HTTP code to send requests.
> It will probably require some further adjustments of the gopher code to look
> more HTTPish or for the Curl_add_buffer-functions to become slightly more
> generic...

Wilco. I'll tweak that section.

> 2 - One of the primary reasons Gopher dropped off libcurl's chart in the
> first place (removed for sure in January 2006 btw if anyone cares) is that
> we didn't have any test cases for it and we apparently didn't have many
> users of it since it broke [...]
> think we shouldn't underestimate the value of having a few test cases setup
> for it. And given the protocol's simplicity and the "stupidity" of our HTTP
> test server, I believe we can perhaps add a gopher mode for that and have it
> serve gopher for this purpose.

I wasn't sure if you wanted a test case on there, so I didn't implement one,
but adding that to the HTTP test server should be relatively trivial. I'll
attend to both these issues and submit a followup commit patch.

Thanks for the review!

-- 
------------------------------------ personal: http://www.cameronkaiser.com/ --
  Cameron Kaiser * Floodgap Systems * www.floodgap.com * ckaiser_at_floodgap.com
-- Five is a sufficiently close approximation to infinity. -- Robert Firth ----
-------------------------------------------------------------------
List admin: http://cool.haxx.se/list/listinfo/curl-library
Etiquette:  http://curl.haxx.se/mail/etiquette.html