split/join vs replace

[ryantenney]

Replace .split().join() in _.template as per discussion on fbd682d and benchmarks here: http://jsperf.com/split-join-vs-replace

[jashkenas]

Thanks for the patch. That's really quite an amazing difference -- merged to master.

[ryantenney]

Thanks for the merge!

"Foolish is the man who attempts to benchmark JavaScript performance in Chrome"

• Me (circa just a moment ago)

Call me a skeptic, but I had a sneaking suspicion that when run in Chrome the benchmark was just spitting out ridiculous numbers thanks to some static analysis and various other voodoo spells and incantations. So I attempted to de-tune the benchmarks in such a way that Chrome wouldn't be able to decide that a particular block of code has no side effect and "optimize" it out of existence. What I ended up with were some more credible numbers that show a substantially more modest improvement of anywhere from 1 - 20% in Chrome and more impressive numbers (5 - 100+%) in other modern browsers. (updated benchmarks here: http://jsperf.com/split-join-vs-replace/2)

Lesson learned.

[jashkenas]

Uh oh, with your new benchmark, it's showing about 2x slower performance for replace() in Safari on the larger string sizes ... perhaps we patched it a bit too soon.

[ryantenney]

I haven't had a similar result, at least not in Safari 5. Unfortunately I don't have any old browsers installed (Safari 5 is the only non-beta browser I have installed).