Closure Compiler Advanced Opts

[kripken]

It looks like only simple but not advanced opts are used. However advanced opts usually give a 50%-100% speedup (+smaller code).

Is there a reason for not using advanced opts? Do you need any help with something there?

[max99x]

Advanced opts don't work in some browsers. I'm afraid I don't remember which browsers were particularly problematic, but I can probably dig it up. For that matter, Opera breaks even with normal closure opts, so repl.it serves it an unclosured version. On a related note, reloop is much slower on Chrome (unlike Firefox), so that's served its special version too. The browser-file mapping is defined here.

Regarding performance, IIRC after the eliminator is run, Closure's effect is minimal in terms of runtime performance (but of course download time is much improved).

[kripken]

Very interesting.

I did some comparisons in the emscripten benchmarks, with either the eliminator, or closure, or both. Mostly the eliminator is about the same as closure, as you said (it's even a bit faster sometimes). However it is significantly slower in the raytrace benchmark, I suspect due to lack of inlining. I tend to suspect Python would benefit from inlining too, but I could be wrong.

Did you try with typed arrays (1 and 2)? How about QUANTUM_SIZE=1? (I assume the latter would not work, though)

[max99x]

IIRC my tests on Python produced negligible differences between Closure simple and advanced, but I'll recheck it next time I rebuild Empythoned. Haven't tried typed arrays or QUANTUM_SIZE=1 yet, since the performance we got was sufficient for a REPL, and I was too lazy to debug the issues that these optimizations might cause. Same as above, I'll give them a shot next time.