duckduckgo / zeroclickinfo-goodies

DuckDuckGo Instant Answers based on Perl & JavaScript
https://duckduckhack.com/
Other
981 stars 1.76k forks source link

Figure out what the error is on Travis. #242

Closed jagtalon closed 10 years ago

ghost commented 10 years ago

The tests that fail to run on Travis with 5.14.4 seem to run OK with Perl 5.14.4 on osx. So I could not replicate this behaviour.

jagtalon commented 10 years ago

@bpaschen Couldn't replicate it on my machine either. Thanks for trying it out, though!

mintsoft commented 10 years ago

@jagtalon I notice that 5.16 seems to pass now, however 5.14 is bombing-out with an issue installing Xslate. I can't reproduce it on my 5.14 or get any more actual information out of Travis however.

The thought occurs, on a related not, do we actually need 5.12 or 5.10 testing? They're pretty old (i.e. debian oldstable and older)

ghost commented 10 years ago

The 5.10 and 5.12 errors are problems with wrong versions of some locale module. I wrote a shell script which checks the perl version and then pulls neccessary packages from backpan. I don't know if that is an acceptable fix though. I did that because apparently cpanm is not able to install specific module versions.

Nevertheless, then the Xslate problem appeared on 5.14 and 5.12 and I don't know what is causing it. Any ideas? Debugging Travis is like rolling a rock up a hill.

mintsoft commented 10 years ago

@bpaschen I'm hoping we can just remove 5.12 and 5.10 all together and sweep the problem away; but it's not my call.

I'm totally at a loss with regards to 5.12's Xslate issue. I just can't reproduce it at all. We really need the output of the /home/travis/.cpanm/work/1387287344.4157/build.log to see wtf is going on.

ghost commented 10 years ago

@mintsoft As far as I remember the problem also popped up on 5.14.

crazedpsyc commented 10 years ago

@bpaschen cpanm can install specific versions with a direct link, or at least the author: cpanm ETHER/Moose-2.1005

Is Xslate still the issue? Can't see that one in recent build logs, all I see is Locale::SubCountry requiring a newer warnings than travis has.

mintsoft commented 10 years ago

@crazedpsyc I think that might be an old build with Locale::SubCountry

The latest seems to be Xslate: https://travis-ci.org/duckduckgo/zeroclickinfo-goodies/jobs/15664991

crazedpsyc commented 10 years ago

Weird. It was current yesterday. Anyway, feel free to add an extra cpanm -v Text::Xslate or so, to show the details. It should be automatically tested once you submit a pull request.

moollaza commented 10 years ago

Ping. What's the status of this? Are there still any errors?

mintsoft commented 10 years ago

@moollaza Sadly yes on 5.14, 5.12 and 5.10.

I managed to re-create the issue with perlbrew on a completely empty machine and fixed it by explicitly installing Xslate & ExtUtils::ParseXS.

Whilst I was testing that I've come to the conclusion that the regex on dzil is wrong, I think it should be as in: https://github.com/mintsoft/zeroclickinfo-goodies/commit/0ee1e6aa01cccfdddf21995902e96e6daa93d48f

This could be why the dependencies aren't installing properly? I think Travis is actually relying on cpanm --quiet --notest App::DuckPAN to install the dependencies.

crazedpsyc commented 10 years ago

This was because ExtUtils::ParseXS was too old on travis, but Xslate did not complain -- going to add a ticket over there. 5.10-12 are failing because Locale::SubCountry is nuts, so I also added a ticket over there.

5.14 is now failing because "iso639 eng" did not trigger anything. Closer!

mintsoft commented 10 years ago

Excellent, @crazedpsyc have you managed to get the 5.14 build to get through to actually running tests on travis?

crazedpsyc commented 10 years ago

@mintsoft yep, that's what my last sentence meant :) One of the goodies tests fails, but at least it gets there.

mintsoft commented 10 years ago

@crazedpsyc I notice that Travis now passes and 5.14/5.12 have been removed; is that permanent or just until the modules get fixed up?

yegg commented 10 years ago

@mintsoft we run 5.16 in production and the other ones were always failing, so the thought was to make this useful we'd stick to what we support (and know works).

jagtalon commented 10 years ago

This looks like it's fixed. :)