Re: Windowing related problems

Andrea Campi <[email protected]> writes:

> Hi Jonas and all,
>=20
Hi Andrea,

> I've finally stumbled upon something which looks related to what
> other people have been mentioning on the list. I am attaching two
> different traces (obtained with RR_DEBUG=3D'NET:FILE_NET=3D...') of the
> server.
>=20
Which issue are you referring to, this problem is probably not similar
to the message sending issue discovered in the RR/IDXP profile. It was
a RR/IDXP bug.

*SNIP*

>=20
>=20
> Now this is quite interesting. The last frame gets fragmented at
> the sender (verified with tcpdump), and the 'more' flag set
> accordingly. I have no idea where and why, but some code seems to
> think the sender has filled its window.
>=20

By looking at the log files you attached it looks like roadrunner
somehow ignores the SEQ frame.

After a quick test where I wrote a simple profile that sends a 200 bytes
MSG and answers with a 100 bytes RPY I wasn't able to reproduce this.
I was able to force fragmentation of frames by sending MSG frames
without waiting for the RPY frame. But the message only got fragmented
when the sender really had filled the window.

(I used the current CVS version while doing the test above)

Can you produce a simple (minimal) profile/testcode that triggers the
issue you mentioned?

>=20
> I am quite sure this started happening since Nov 21, when I last
> imported stock sources to my local repository. The problem is,
> I don't really see anything changed that might cause this
> problem (I'm not using rr_mime_* yet).
>=20
> Any idea before I start to put debugging printf's all over the
> place?
>=20
The problem (if it exists) is probably in rr-connection.c or rr-tcpconnecti=
on.c
You can look at the functions rr_connection_get_next_frame, handle_seq_frame
and send_seq_frame.

Thanks!

/ Jonas
--=20
Jonas Borgstr=F6m                  [email protected]
CodeFactory AB                   http://www.codefactory.se/
Office: +46 (0)90 71 86 14