Issue2708

I noticed that rather recently (some months ago) test_urllib2net started to fail on TravisCI. I went through the github commit history to identity the satrting point of failure.
It seems to be this commit:
https://github.com/jythontools/jython/commit/2f4c2928b617a10aa99bb5de98926015500481d8
where it consequently starts failing.
One commit earlier (https://github.com/jythontools/jython/commit/52eb2fbae077036f9488a2ee9b78ed1630f2d306) the test still went well.

Routinely breaking tests make TravisCI less useful for us, so this should be cleaned up.

Given that the "breaking" PR is titled "Refactor PyJavaType.init for readability (no material change)." I suspect this issue might be induced some change in Travis CI. Traditionally, networking tests were not so reliable anyway. However also in this case we should consider to skip this test or better only the broken bit.

Any ideas what actually goes wrong?

It is a fairly big refactor ... and it does not quite live up to the description. But the thing I know I broke here (sorry):
https://hg.python.org/jython/rev/623eaa3d7834
I fixed here:
https://hg.python.org/jython/rev/2adcf0e29407

test_urllib2net passes for me at both these revisions on Java 7 & 8. Also at the tip on 11. All on Windows.

I'd been attributing those failures to the network environbment in which the CI is run. I agree it is useless if it just shows red all the time for reasons we think are spurious, but there is a bit to do to fix it. I can't say I really follow the scripts. Our CI needs revisiting on quite a few points.

The CI (Travis, not looking at the others) performs some fatally failing runs due to some config issues. However, there are also some okay runs and for these the only failure is test_urllib2net since some time.
So. fixing (or muting) this issue, turns at least some runs green, providing a better look than currently.

     [exec] 379 tests OK.
     [exec] 2 tests skipped:
     [exec]     test_codecmaps_hk test_curses
     [exec] 1 test failed:
     [exec]     test_urllib2net
     [exec] 1 fail unexpected:
     [exec]     test_urllib2net
     [exec] Re-running failed tests in verbose mode
     [exec] Re-running test 'test_urllib2net' in verbose mode
     [exec] test test_urllib2net failed -- multiple errors occurred

Unfortunately the "verbose" rerun of test_urllib2net is not very enlightening. I suspect it will be tedious to find the cause.

In another run I found this:

     [exec] Re-running test 'test_urllib2net' in verbose mode
     [exec] test test_urllib2net failed -- Traceback (most recent call last):
     [exec]   File "/home/travis/build/jythontools/jython/dist/Lib/test/test_urllib2net.py", line 294, in test_ftp_basic
     [exec]     u = _urlopen_with_retry(self.FTP_HOST)
     [exec]   File "/home/travis/build/jythontools/jython/dist/Lib/test/test_urllib2net.py", line 27, in wrapped
     [exec]     return _retry_thrice(func, exc, *args, **kwargs)
     [exec]   File "/home/travis/build/jythontools/jython/dist/Lib/test/test_urllib2net.py", line 23, in _retry_thrice
     [exec]     raise last_exc
     [exec] URLError: <urlopen error ftp error: 425 Security: Bad IP connecting.>
     [exec]

Looks like there was a Travis infrastructure change around this time that causes severe instability to anything using the ftp protocol.

https://blog.travis-ci.com/2018-07-23-the-tale-of-ftp-at-travis-ci

https://github.com/travis-ci/travis-ci/issues/9352

https://github.com/travis-ci/travis-ci/issues/9112

https://github.com/travis-ci/travis-ci/issues/9391

Network calls are often flaky on CI infra. Perhaps the test can be skipped in CI, or tweaked to remove ftp on CI, or similar.

PR https://github.com/jythontools/jython/pull/142

Corresponding CPython bug https://bugs.python.org/issue17160

Fix by Adam (thanks) now at:

https://hg.python.org/jython/rev/f3baccd1fc34