Short-circuiting conditionals (&&, ||)

[kg]

In a discussion about control flow it came up that asm.js currently doesn't support && or ||. The resulting control flow trees are pretty complicated sometimes. I think we should make sure short-circuiting boolean logic is expressible in webasm so that the resulting AST is denser and shallower. Thoughts?

EDITED: clarity

[jfbastien]

+1

I also suggested that the polyfill should use short-circuit. This implies either adding them to asm.js (I remember discussing this with @sunfishcode and @kripken, but can't recall the outcome), or having the polyfill not be strictly asm.js.

[lukewagner]

Polyfill isn't a problem because a && b === a ? b : 0 and a || b === a ? 1 : b, which is what Emscripten generates instead. Given ?:, the question of whether to add && and || goes back to the same directness vs. minimality question we're having in the other PRs. FWIW, in FF, those ternary forms are optimized to produce the expected control flow.

[kg]

Ah, so ternaries work in asm.js but not && or ||? Interesting, less problematic.

[sunfishcode]

Also worth noting is that when we implement the CFG structuring technique described in #44, the problem of complicated control flow will be reduced in general.

[titzer]

Consensus is that we have statements = expressions, so building short-circuiting conditionals can be done pretty easilly.

[kripken]

Where is the discussion leading to statements = expressions? I must have missed it.

[lukewagner]

Even without stmts=exprs, AstSemantics.md currently has a ternary. asm.js does not have short-circuiting because ternary was sufficient.

[kripken]

More generally, we have an open issue on statements =?= expressions, #223, which does not have any conclusion, nor any activity for almost 4 months. That's why I'm surprised to hear @titzer say "Consensus is that we have statements = expressions".

(No concerns from me on closing this particular issue about && and ||.)