space-infix-ops: need to exclude "a|0" pattern from warnings

[puzrin]

"a|0" is well known pattern, when result is forced to be signed 32 bit integer (makes sense when writing high speed js). It should not give warning about missed spaces.

Want to back this issue? Post a bounty on it! We accept bounties via Bountysource.

[michaelficarra]

a | 0 works just as well.

[puzrin]

I know. But all write a|0, without space. That help to understand, that it's a type cast, not binary operation.

[puzrin]

Here is example from real code https://github.com/nodeca/pica/blob/adb2b2c78f784b2385c6aa00e973c1b2063e056a/lib/pure/resize.js#L182-L185 . That helps JIT to remove some types check in generated inline caches.

[nzakas]

We definitely can't support this by default, this would have to be some sort of exception if we end up supporting it. I'd like to hear more opinions p.

[lo1tuma]

I don’t think it is a good idea to write such code by hand. JavaScript doesn’t have 32bit integers.
Use a statically typed language that can be transpiled to asm.js or something like sweet.js macros.

[puzrin]

@lo1tuma , be in - sync - with - real - life :) . I'm one of those who write such code, when needed. There is nothing wrong in deep understanding of JITs principles.

[nzakas]

Okay, let's stop the could vs. should conversation, as that's not very helpful.

I took a look at the asm.js spec and it makes a clear distinction between f|0 as a type annotation and f | 0 as a binary operation. This is also the only type annotation for which the default coding style doesn't really apply.

So I think we should support this exception as an option. What should it be called? asmInt?

[puzrin]

Don't know, is this syntax initially related to asm.js or not. Also, _|0 can be used (with leading space). Key point is no space between | and 0.

May be, int32? I'd prefer to avoid promoting terms intersection via naming - there is nothing common between writing speed-optimized monomophic code and asm.js. Writing asm.js manually is mad idea.

As far as i know, we have only one widely known exclusion. Sometime it's possible to cast Uint32 with >>> 0, but i haven't seen it widely in practice, and never used myself. Uint32 is not supported well in different JITs, and attempt to force it can cause boxing.

[nzakas]

int32 isn't specific enough, it makes it seem like it has something to do with numeric literals. int32Hint?

[puzrin]

Argee. int32Hint is much better.

[michaelficarra]

The name should have no semantic meaning. It should refer only to the exceptional syntactic form: bitwiseOrZero.

[nzakas]

bitwiseOrZero is confusing because why not also bitwiseOrOne?

The only real reason to use this option is for type hinting, so having the option refer to this is important IMHO.