Unable to duplicate claimed performance numbers

[johnezang]

Unable to reproduce claimed performance numbers. Timing based on the twitter_public_timeline.json file in Sams json-benchmarks, using (approximately) gcc-4.2 -arch x86_64 -DNS_BLOCK_ASSERTIONS -Os:

JSONKit        | decode: min:      415.000 us, avg:      424.730 us, char/s:  57775527.98, decodes:   1000
json-framework | decode: min:     1675.000 us, avg:     1687.973 us, char/s:  14537554.81, decodes:   1000
CoreJSON       | decode: min:     1230.000 us, avg:     2641.324 us, char/s:   9290416.47, decodes:   1000

It's not clear to me how CoreJSON can significantly improve on the times turned in by YAJL-objc. After parsing overhead is accounted for (which libyajl handles), the next major contributor to overall parsing time is in object instantiation, which is fairly considerable. This primarily how JSONKit is able to turn in better times than the other major libraries- it has a recently instantiated object cache to significantly reduce this overhead.

[mirek]

interesting, will check

[mirek]

Thanks for pointing this out, numbers in json were out of 2^32 range which were not handled properly. Updated stats, respect for JSONKit speed.