- made a separate buffer for uploads (due to the non-blocking stuff)
- added two connectdata struct fields for non-blocking uploads

2) in the curl_dns_cache_entry structure, timestamp should be
a time_t instead of an integer (although I doubt it matters).

The default cache timeout for this is 60 seconds, which is arbitrary and
completely subject to change :)

    - Use a global dns cache (via setting the tentatively named,
    CURLOPT_DNS_USE_GLOBAL_CACHE option to true)
    - Use a per-handle dns cache, by default
    - Use a pooled dns cache when in the "multi" interface

we return a set of headers not more. This enables FTP operations that don't
transfer any data, only perform FTP commands.

is not dealt with when we find an end-of-response line, as there might be
important stuff even after the correct line. So on subsequent invokes, the
cached data must be used!

attempting to use EPSV before the standard PASV.

for each *_perform(). It makes things a lot easier, as the first one that
detects the error get to write the final error reason...

to hold 512 bytes (for user+password+hostname)

HTTP server we return error

libcurl should use in its request

valid time. we now store the filetime as a long to know for sure it can
hold -1 (there exist some unsigned time_t cases)

seriously decreases the amount of used stack space

the SessionHandle. It was previously wrongly put in UserDefined

be posted in a minute to the libcurl list.

cache

it.

already been transfered no longer returns error but instead is OK. The
reasoning behind this is of course that no extra actions need to be taken
and it is as if a transfer had been successfully performed.