Update text syntax to use (memory $i)

[alexcrichton]

This commit intends to apply #17 to the test suite by using (memory ...) instead of bare indices which can be confused with other proposals
such as SIMD which use bare indices for other purposes.

[rossberg]

Thanks! This makes sense, at least for load/store instructions. However, for memory.* instructions it's a bit odd, since in all other cases where an entity.xyz instruction is followed by an $entityidx, we don't require an extra keyword. So for consistency, I'd leave it out for those.

I would prefer holding off landing this until a corresponding change to the interpreter is implemented, since that should always stay green on the test suite.

[rossberg]

Now that SIMD has landed, I created a PR for merging upstream (#26). This required adapting the parser to handle both memory indices and lane indices for lane load/stores. In doing so, I did not actually encounter any grammar ambiguities with that. In the interest of overall consistency with other memory instructions and the likes, I hence decided not to implement the syntax change suggested in this PR. Instead, #26 adds a few syntax tests to memory-multi.wast that check all combinations of immediates for lane load/stores.

(I had to do a minor grammar transformation to avoid yacc conflicts, see parser.mly#461 and parser.mly#500. However, using (memory $x) would in fact have required much more substantial transformations.)

So I'd be inclined to close this PR. Does that sound okay?

[alexcrichton]

Looking at this again I agree it's not ambiguous. If you'd like to keep the syntax the way it is then that seems like what it'll be then.

[rossberg]

Okay, thanks. I'm closing this PR then.