Rename anyfunc to funcref in js-api tests

[gahaas]

The string anyfunc still occurred in js-api tests. This PR replaces all these occurrences by funcref.

[rossberg]

Hm, removing anyfunc entirely from the JS API would be a breaking change. I think we need to keep supporting it in the Table constructor for backwards compatibility.

[Ms2ger]

Agreed, this is not workable. We could add an alias "funcref" → "anyfunc", though that doesn't seem all that valuable. (Not entirely serious alternative: rename "externref" to "anyextern"?)

[rossberg]

Aliasing seems fine. We could restrict use of the old name to the Table constructor, just like we do for the "initial" limit in the JS Types proposal.

[Ms2ger]

I guess this alias would be rather less complex than the one in JS Types. It looks like we don't expose the value to JS yet, either. If we want to make "funcref" the canonical value, we'll also need to make sure JS Types returns it from the type function.

[lars-t-hansen]

Hm, removing anyfunc entirely from the JS API would be a breaking change. I think we need to keep supporting it in the Table constructor for backwards compatibility.

FWIW, through an oversight Firefox has been accepting 'funcref' exclusively for the last two years, while Chrome has been accepting 'anyfunc' for at least the last three years, judging from git blame. I have received zero bug reports about the incompatibility until Asumu just pointed me to a bug about WPT adding some tests for this (web-platform-tests/wpt@0286e80#diff-631970eae3abaebb8e90fc901c0725f3eea12e792f2c83b45b2b96b647a56dcb)

Since Chrome accepts anyfunc only I will fix this in Firefox, but I think the compat concerns are probably a little overblow. (Of course there could be workarounds for the discrepancy out there, but again, I've received no bug reports.)

[rossberg]

@lars-t-hansen, interesting. I would of course be happy to retire anyfunc. Maybe V8 could accept both for a transition period and implement a counter to see whether anyfunc is used in the wild? (I suppose there simply is very little code that has a need to actually create a table through the JS API.)

[gahaas]

Accepting funcref would be okay for V8. It would be nice though to add funcref to the spec.

[lars-t-hansen]

Well, post in haste, edit at leisure :-/ It turns out that we had a special workaround for parsing the table descriptor, which in addition to funcref also accepts anyfunc. So what I wrote above is probably not worth much as guidance, as it does not apply to tables, and tables are probably the majority use case by far.

Sorry for the noise. (I still think accepting funcref as an alias for anyfunc in the JS API is a good idea.)

[rossberg]

@gahaas, how about you revive this PR, and change it to aliasing the name? For process reasons, could be made part of the JS type reflection proposal.

[gahaas]

Sure, I will look into this.