doseq & for macros should support chunked seqs

This is fantastic news, thanks!

Fixed, http://github.com/clojure/clojurescript/commit/44389e2f233ccc26ed5a1c75b1e043054e61b586

fixing chunked-seq? made the perf difference quite small for regular seqs. We can take futher enhancement in other tickets/patches.

Reviewing the patch some more, it might be worth inlining chunk-append and chunk via macros. Also in the chunked seq branch, since we've hinted the result of chunk-first as ^not-native we should make sure to just call the protocols fns directly where possible.

Interesting! I'd be happy to drop the bitmask stuff as it complicates the implementation, though looking at those benchmarks it doesn't appear that it would make much of a difference. It does seems that Safari 6 does greatly prefer the bitmask approach. We'd have to do quite a bit more testing before deciding of course.

Ah, good idea, I'll give it a try.

Incidentally, about making satisfies? faster: I had a look at the jsPerf benchmark linked to from CLJS-246 – here's a direct link: http://jsperf.com/direct-vs-chain/8 – and it seems like JS engine perf profiles may be shifting. In Chrome 24 "property" and "property many interfaces" are about equally fast and somewhat faster than "bit" & "bit m.i.". Chrome 21 saw no difference, recent versions of Firefox seem to be similar to Chrome w.r.t. relative results, Opera seems to prefer bit, but only slightly, Safari has a slightly stronger preference for bit than that.