Tests failing because of wheel installation error

[sjkingo]

@rmb938: there is a failure in the tests because it can't find MarkupSafe to install as a wheel (a dependency of flask). It's definitely on pypi and I can install it successfully.

Build log is here: https://travis-ci.org/sjkingo/virtualenv-api/jobs/119961786

The pip output says it's ignoring indexes and using cache only? What's the --no-index flag to pip for in this instance?

Would you be able to take a look and see if you can fix?

[rmb938]

The --no-index ignores the pypi package repo and instead looks in the --find-links directory to install things from.

Inside /tmp/wheelhouse should be the wheel for the MarkupSafe package but I am guessing it is not there.

I can do some testing this weekend to figure out what is breaking.

[sjkingo]

What's the advantage of using --no-index - wouldn't we want it to check /tmp/wheelhouse first, and if it's not there, fall back to PyPi?

[rmb938]

Well in that test we are testing the wheeling functionality. So we wheel up the package then we try and install it from that wheel.

I see your point where falling pack to PyPi makes sense however that functionality isn't built into pip install. So if we get rid of --no-index everything will go from PyPi instead of trying to install from the wheelhouse directory.

[rmb938]

Something much of changed recently to make that test fail as I believe it was working fine back when we added the wheel functionality.

[sjkingo]

Yes I agree that it was working originally - but with no commits (and hence tests run) in 8 months I'm not sure when or with what version of pip! Defeats the purpose of CI somewhat...

If you wouldn't mind having a look to see why the MarkupSafe wheel isn't being created that would be good.

[rmb938]

Yup I will take a look. If you can give me until tomorrow I should be able to have a solution or a workaround by then.

[sjkingo]

No worries, thank you

[rmb938]

Well I have some bad news. No matter what I do I can't see to replicate the error from travis. The tests run, without errors, running wheel manually works fine and creating a small test script works as well. The only thing I could think of is the travis build is running in a debian container vs my centos machine ... but really that shouldn't matter much if at all.

I will try and do some more testing but if I can't see to replicate it then I am kinda clueless.

[sjkingo]

Thanks for trying to reproduce. Must be something to do with the Travis infrastructure then. I'll have a look and see.

[sjkingo]

Thanks for the push, that fixed it!

[sjkingo]

Oops; wrong thread..

Either way, we're all good now.