s2wasm emits invalid drops or types

[kripken]

e.g. in test/dot_s/alias.wast,

  (type $FUNCSIG$v (func))
  (func $__exit (type $FUNCSIG$v)
    (return
      (i32.add [..])
    )
  )
  (func $__needs_exit (result i32)
    (drop
      (call $__exit)
    )

So __exit is declared as not returning, but does have a return for some reason, and is dropped in __needs_exit. I'm not sure if the drop is wrong or the type is wrong or both (not sure how to read the .s file to tell).

cc @dschuff @sunfishcode

[dschuff]

from alias.s it looks like __exit is void (i.e. no .result) but it also has return $pop4. Also the drop for the call to __exit isn't in the .s, so I'm not sure exactly where it came from. It's also a known (to me) issue that s2wasm emits a bunch of redundant drops (the code I added in the last update to put them in for 0xc is really dumb) so I had already planned to look into the drop code once Binaryen's 0xc stabilized, so I'll look at this too.

[kripken]

I did some poking around,

• Manually adding .result i32 to __exit helps, it then prints out a proper wast when drop validation is ignored. This suggests the wasm backend is emitting it wrong?
• However, it still doesn't validate properly in memory with drop validation, as the type of the call node is none (and it is dropped, which is wrong). One odd thing there is that the call isn't dropped in the .s file, which maybe it should? Anyhow, somehow s2wasm needs to know the type of the function called and set the type of the call node properly. Perhaps this can be done in a post-pass (asm2wasm does it that way), but explicit drop in the .s file (with the type) would be another way.