On Thu, 10 Aug 2006, Armel Asselin wrote:

> here is a patch, it works well for me, but i could not pass the test suite on
> it (i am on Windows, how should i do?)

Thanks! I don't believe it works fine on Windows at all. Parts of it used to
work on Windows, but I don't think anyone has tested for a good while.

> it does not optimize first 'TYPE' request on new connections to avoid relying
> on server respecting the TYPE A default (as they should as per rfc959).

Sounds good, since we then won't stop functioning on servers we currently work
with just by assuming they are standards compliant...

Let me just comment on the code style. You did several lines like this:

+ if ((result=ftp_nbsendtransfertype(conn, data->set.ftp_ascii, FTP_TYPE))) return result;

... while curl code in general always is narrower than 80 columns and splits
it up like this:

  result=ftp_nbsendtransfertype(conn, data->set.ftp_ascii, FTP_TYPE);
  if(result)
    return result;

I happen to find it much easier to read and the generated code is exactly the
same. You also didn't follow the {}-style or even the 2-space indent
tradition.

Apart from that the actual functionality seems fine. I'm having this patch on
hold a bit though since I'm working to get a monster-huge patch applied that
will remove the tight connection between 'connectdata' structs and
'SessionHandle' structs (and thus your patch will need to be slightly
adjusted).

-- 
  Commercial curl and libcurl Technical Support: http://haxx.se/curl.html