Tests fail in rails 3 mode on ruby 1.9.3 with v0.9.7.4


(will) #1

You can see them here

https://travis-ci.org/driveless/discourse

the branch is pretty nearly vanilla v0.9.7.4 with changes specific to my setup on top

I also had to add a magic comment to one file that you can see the effect of here

https://travis-ci.org/driveless/discourse/builds/14078116

Update:

I’ve narrowed it down to just rails 3 mode. Tests run fine on both ruby versions in rails 4 mode but fail in 3


(Sam Saffron) #2

We will be dropping rails 3, no point fixing this really


(will) #3

Ok got it.

Still getting 2 test failures in rail4 mode though

https://travis-ci.org/driveless/discourse/builds/14087207


(will) #4

And its because that constant does not exist in ruby 1.9.3

They use an argument error there. That makes since to me.


(will) #5

Made a pull request


(Sam Saffron) #6

Thanks, fixed now, this does raise an interesting issue though.

We still want to support 1.9.3 for at least another 6 months, but weak refs are broken so the test suite will fail once in a while.

We need to “allow” the test suite to retry certain specs so we can turn 1.9.3 back on in travis.


(will) #9

Ahh yeah I got one of those weird fails with a mutex in the Category controller spec once

So I see what you are saying


(Sam Saffron) #10

If we can figure out how to build in a retry, I would be more than happy to add it back to travis

Let me ask on parley :slight_smile:


(Sam Saffron) #11

Ok, I think this can work, mind working on a PR that uses GitHub - dblock/rspec-rerun: Re-run (retry) failed RSpec examples. in Rails 1.9.3 mode ? at least that way it will not erratically fail.


(will) #12

Having trouble getting my app deployed in rails 4 mode on cloud66 because for some reason 2 asset manifest files are generated

one with the hash digest at the end of the name and one without.

and their cap deploy task fails because of it

Can’t figure out whats wrong with it.

Once I get that figured out I’m up for it


(Sam Saffron) #13

I thought we fixed that:

https://github.com/discourse/discourse/commit/908ba32d379ab91738f63a267ac41299e673dbbe


(will) #14

So mad I couldn’t find that on my own lol.

I literally spent the past three hours googling permuations of “rails two manifest files”

Of course its literally the next commit after the release was cut lol

but isn’t

next if File.extname(file) == ".json" # Avoid duplicate manifest

a bit aggressive though.

Wouldn’t that skip any json file even one that was like a mock for some javascript object or something of that nature?

Just curious


(Sam Saffron) #15

Yeah its a tad aggressive, but I don’t see why we would bundle any json files, besides that whole file is one big hack. We need a smarter way of letting plugins register unhashed assets.


(will) #16

Got it. And I finally got my app working correctly in my staging environment and was able to go to production with no hiccups

:smile:

I’ll take a look at doing the rspec reruns tomarrow doesn’t look too tough.


(Sam Saffron) #17

This topic was automatically closed after 1 day. New replies are no longer allowed.