Daniel Stenberg wrote:
>>> ... It must thus at least be converted into a loop of calls.
>>
>> Indeed :-) The question is: do we set this loop in smb or in
>> Curl_client write()? The later will resolve this problem for other
protocols too.

> Yeah, that would be better approach indeed. Feel like taking a stab at
it?
Yes I can.
But looking at the code I noticed another potential problem in
Curl_client_write() with the pause feature:
- Let's start with an unpaused input transfer.
- Curl_client_write() is called for a bunch of data, which might be the
translated in place (encoding, line-ends).
- The callback requests pause, causing the ALTERED data to be pushed.
- If subsequent Curl_client_write() calls are performed, the new data is
pushed UNTRANSLATED.
- When resuming from the pause, the pushed data is translated again,
because flushed by the Curl_client_write() procedure :-(

I'll look to fix this problem together with the long size, because these
are quite related... But it might take a bit longer than a thought. Any
comment ?

>> Another question: what about the idea of dissociating crypto from TLS

>> and adding a virtual lib/crypto subdir as it has been done for TLS ?
> Yes, that is something to consider. But I'm also concerned that it may
just cause _another_ mess instead of the one we currently have and thus
just be a lot of work that doesn't end up substantially better.

Agreed: this must be done carefully.
I'll probably look at it next year and commit this work in a git fork
first.

-------------------------------------------------------------------
List admin: http://cool.haxx.se/list/listinfo/curl-library
Etiquette: http://curl.haxx.se/mail/etiquette.html
Received on 2014-12-09