Clarified that implicit return is allowed

[Teemperor]

Current spec doesn't specify what happens in non-void functions without any return.

ml-proto is already doing it as described in the paragraph: https://github.com/WebAssembly/spec/blob/master/ml-proto/test/f32.wast#L2

[rossberg]

This is just a consequence of being an expression language, so probably hinges on #223 being resolved.

[kg]

If I don't end with a return, but my last expression has a type that doesn't match the return type of my function, do we detect that and reject it as an error?
Is it reasonable to do that analysis statically when loading/compiling a wasm module at runtime, since it's 'does the function ever invoke return along this path of execution'? You can certainly do it but I'm not sure how fast it would be.

It might be simpler to just obligate all non-void functions to have a 'last chance' return as their last expression, and prohibit jumping past it with break. Most code generators will produce it anyway and that constraint should be much easier to enforce.

[rossberg]

If I don't end with a return, but my last expression has a type that
doesn't match the return type of my function, do we detect that and reject
it as an error?

Of course. Already the case.

Is it reasonable to do that analysis statically when loading/compiling a
wasm module at runtime, since it's 'does the function ever invoke return
along this path of execution'? You can certainly do it but I'm not sure how
fast it would be.

The function return type is explicit, so this check is completely trivial,
regardless of what returns a function uses. (In fact, it is simpler than
checking returns.)

[titzer]

I agree with @rossberg-chromium; if we resolve #223 then this is a non-issue.