mercurial-1.3.1 self-test fails on fedora

[Alejandro Santos]

On Fri, Jul 24, 2009 at 11:40 AM, Martin Geisler<mg at lazybytes.net> wrote:
> Mads Kiilerich <mads at kiilerich.com> writes:
>> On 07/24/2009 02:22 PM, Neal Becker wrote:
>>> + /usr/bin/python run-tests.py
>>> .............................s....................sssssss.sssssss..s...ssss.ssssssss.............................................s......................s.
>>> ERROR: test-http output changed
>>
>> The Mercurial test suite is not unit tests testing small parts of the
>> code in artificial situations in 100% reproducible environments.
>> Instead it tests the whole Mercurial system in real use cases, and it
>> thus tests differently and catches many bugs that unit tests wouldn't
>> have caught - including some bugs that isn't there.
>
> Well put! I see failures in the HTTP tests quite often here, especially
> after upgrading my machine. But they always go away when I re-run the
> tests so I attribute them to weird stuff like network ports not being
> closed fast enough or stuff like that.

I also see lots of HTTP errors on the test suite when the script is
stopped and hit by a Ctrl+C. There is some dummy background script
created by one of the test cases that won't die.

In that case a dirty-and-not-so-quick solution is to kill the
background task and restart the "lo" network interface ("ifdown lo;
ifup lo" on Debian based systems) so the sockets with a TIME_WAIT
state on the loopback are (forced to be) closed.