Fix issue #4859: allow loading RTL text plugin before first style load.

[ChrisLoer]

Fixes issue #4859: this was a regression I introduced in 171db61. Unfortunately I still haven't figured out a good way to exercise this code in our test suite, since it depends on running in the browser. :/

/cc @anandthakker @mollymerp

[mollymerp]

Hmmm would it be possible to unit test to ensure that the callback is called with argument of type Object with those properties?

[ChrisLoer]

@mollymerp I had to hack things up a bit to do the stubbing (I'm not very experienced with Sinon), but I extracted the hard-to-test createBlobURL and put in a test that verifies a bogus pluginBlobURL gets passed down to the workers. Do you think that makes sense?

[mollymerp]

Thanks for adding that test!
Would adding a calledWith check to make sure the correct arguments are passed to the test here https://github.com/mapbox/mapbox-gl-js/blob/master/test/unit/style/style.test.js#L59 have caught this specific regression? Sorry if this is a dumb question – I'm not too familiar with the RTL plugin 😬

[ChrisLoer]

@mollymerp A calledWith check on the original test (https://github.com/mapbox/mapbox-gl-js/blob/master/test/unit/style/style.test.js#L59) wouldn't have worked originally because it didn't have a stubbed out createBlobURL (one confusing thing here is that the the URL we ultimately care about is the "blob URL", not the "plugin URL" that gets passed in to setRTLTextPlugin). I'll try adding a setRTLTextPlugin call to the test now that we've got the stubbing mechanism.

[ChrisLoer]

I had to add a clearRTLTextPlugin method for the tests to allow both tests to modify the plugin... global singletons don't play well with tests. :/

[mollymerp]

Thanks and sorry for the testing troubles!