On Fri, 21 Sep 2012, Bill Middlecamp wrote:

>> Anyway, I think we need to handle this somehow. I just can't think of the
>> best possible fix right now. I think a good first step would be to write up
>> a test case for the curl test suite that repeats the above scenario so that
>> we have something to work with.
>
> I cannot write a test case because it depends on the server.

No it doesn't. It is quite clear if you read the code that it isn't really
depending on the server. The CWD back to the entry path is made when a CWD is
to be made on a re-used connection. Alas, if you don't do a CWD on the new
connection (like with NOCWD set), libcurl will skip that step and things (can)
go wrong.

It should certainly be possible to write up a test case against the normal FTP
server from the curl test suite.

-- 
  / daniel.haxx.se
-------------------------------------------------------------------
List admin: http://cool.haxx.se/list/listinfo/curl-library
Etiquette:  http://curl.haxx.se/mail/etiquette.html