On Sun, 16 Nov 2008, koettermarkus_at_gmx.de wrote:

> I hope you don't mind if I throw in my own results with the modified
> ftp-multi-upload.c

Oh certainly not! We can use as many eyes as possible on this!

> * STATE: DONE => COMPLETED handle 0x748f8b0; (connection #-5000)
>
> gdb told me it hangs in
>
> #0 0x00007f77fe9c4433 in select () from /lib/libc.so.6
> #1 0x00000000004014cc in main ()

Yes, this looks similar to what I managed to reproduce but I've not followed
it through since I didn't think it was close enough to Andreas Wurf saw.

It is logical that it gets stuck in main and it is the same for what Andreas
Wurf reports and what I saw but that doesn't necessarly indicate the same bug.
Although I'm not of course saying it isn't.

> STOR file01
> 150 Ok to send data.
> CWD /
> -------------------

Oh, this looks like the same thing. Or at least similar. I think I'll rerun my
tests and see if I've simply just not been analyzing my debug output good
enough.

I wonder if that 'CWD /' command that shouldn't have to be there is an
indication of the bug or why the heck that appears there.

> Digging in the code, I found "Expire cleared" in lib/multi.c Curl_expire(),
> I came across the same function trying to debug my 'loosing timeouts' for
> http in hiperfifo.c.
>
> 2 different plattforms, 2 different protocols, same bug?

Hm, that would then truly indicate a generic timeout bug. But I fail to see
how the failure to read an FTP server response is due to a timeout flaw.

Thanks a lot for the additional clues!

-- 
  / daniel.haxx.se