> What steps will reproduce the problem?

run 'make check'

> What is the expected output? What do you see instead?

Expected: | == Running test_all == | ............... |
| OK (15 tests)

Result: | == Running test_all == | Received more requests than expected | ..FFFFF........ | | There were 5 failures: | 1) test_serf_closed_connection: test/test_context.c:444: expected <0> but was <20014> | 2) test_serf_setup_proxy: test/test_context.c:512: expected <0> but was <48> | 3) test_keepalive_limit_one_by_one: test/test_context.c:700: expected <0> but was <48> | 4) test_keepalive_limit_one_by_one_and_burst: test/test_context.c:860: expected <0> but was <48> | 5) test_serf_progress_callback: test/test_context.c:978: expected <0> but was <48> | | !!!FAILURES!!! | Runs: 15 Passes: 10 Fails: 5

>What version of the product are you using? On what operating system?

0.3.0 on FreeBSD 7.2

> Please provide any additional information below.

apr_pollset_poll() uses kqueue, and returns APR_POLLIN and APR_POLLOUT into separate events. As a consequence replay() is called twice, which messes up the tests.

The attached file contains a patch to fix the issue.

I believe that r1375 on serf trunk fixes this issue.

The patch attached here seemed to solve the issue most of the time, but the problem turned to be a fundamental flaw in the design of the test framework.

Can you (issue reported) see if r1375 fixes your problem in freebsd?

Thanks for the report & patch!

Please revise r1520.

Reverting r1520 on the 1.0.x branch unblocks the test suite on current FreeBSD and Debian GNU/kFreeBSD.

http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=632573 http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=621692#77