Tests and spec disagree about call_indirect table specifier about reference-types

[rossberg]

The test is correct, and consistent with the text format in the draft spec. I'll fix the overview.

Having the table index first is way more consistent with the other instructions, where the index of the main subject always is first, and never after a non-trivial thing like a typeuse.

IME, having the index first also is significantly easier to parse, because there are so many ways to expand a typeuse into an arbitrary number of items.

[sbc100]

Ok that makes sense. And just to be clear the table specifier is required (not optional) in the text format after this proposal?

[rossberg]

No, it's optional for backwards compat. Same in other places and with memory indices in the multi-mem proposal.

[sbc100]

Is that optional-ness reflected in the draft spec? I didn't see anything when reading it.

I think it was the fact that it was optional that was causing difficulties for the wabt parser but I'll have another go now.

[rossberg]

Yes, see Abbreviations below.

[sbc100]

Great, sorry I didn't spot that. I will update wabt, which currently implement the "table as last arg" approach.

[rossberg]

In case, just steal the grammar from the reference interpreter. :)