Assertion failed: fetched (xrep.cpp:196)

[bwbuchanan]

When a client of an XREP socket disconnects, this assertion fails. Code to reproduce to follow. According to git-bisect, the bug was introduced by commit 10c28c1:

Author: Martin Hurton <hurtonm@gmail.com>
Date:   Sat Jun 19 20:46:16 2010 +0200

    Revive reader on pipe termination

[bwbuchanan]

http://gist.github.com/455010

[ellisonbg]

I see this one as well.

[mato]

I get this with stock 2.0.7 as well as current git, so it doesn't seem to be related to the mentioned commit. As a workaround, setting the IDENTITY of the client (REQ/XREQ) socket makes the problem go away.

Martin Sustrik is away until mid next-week, I will remind him to take a look at this when he gets back.

[bwbuchanan]

The specific test case I linked to above is definitely triggered by this commit in my environment, as confirmed by git-bisect. Reverting that change from current git makes the test case succeed for me. Perhaps this is a race condition, or there are multiple ways to trigger the bug.

[echou]

In xrep::xrecv(), call all readers' check_read() to determine current_reader, then the identity is returned at first. The real messages are returned by successive xrecv calls.

The above code assumes pipe's front message is a normal message. However it can be a DELIMITER message to indicate the termination of the pipe. In this case reader.read() return false (but check_read() returns true). It breaks the assumption and causes the assertion fail.

[hurtonm]

I have fixed this in my private repo (http://github.com/hurtonm/zeromq2/commit/9daf6214dff8a2fe1d3bfadb5b4ab58bb38cf354).
Brian (and others) could you please test and report back? If everything is OK, I will merge upstream.

[bwbuchanan]

Looks fixed to me, thanks!

[mato]

Looks ok here too. Thanks!

[hurtonm]

Thanks for reports. I have merged the fix into master now.