Revised semantics for `%wasm` blocks.

[miviwi]

What is the problem?

Currently the code accepted (and then, further parsed) by the compiler in %wasm blocks is not regular WAT syntax, but effectively a dialect.

As a simple example, the following WAT:

(loop $loop (block $a_block
  (i32.sub (local.get $i) (i32.const 1))

  (local.get $i)
  i32.eqz
  (br_if $a_block)

  br $loop))

semantics aside — syntactically it is correct. I've verified this by including the above snippet as part of a WASM (func) and passing it to wasmtime.

Unfortunately under the current grammar rules it'll be regarded as invalid syntax and rejected by the compiler.

Expected behaviour

Both the syntax (%wasm token) and all the current usage implies that any valid WAT should be accepted.

Proposed Solution

It's not as cut-and-dry as correcting the grammar alone, because - as mentioned above - the compiler must itself parse the code inside those blocks. Such a solution would create even bigger problems down the pipe.

So to really solve this we have, as I see it, two options:

1. Live with it for the time being (until version 1..?). Making some kind of small change hinting at the fact that what you put inside those blocks is not straight-up WAT but a limited subset instead.
2. Make all the sweeping changes that accepting all forms of WAT would entail right now, although this would be a very big undertaking without any "immediate" gain.