Tokenize the space value when stringify-ing

[ericf]

Using a token for the space value, the Node 0.10 CPU issue is avoid while also keeping the file size closer to the no-space size.

Using the test script in #3, the results of this method add a bit of CPU time, but its worth the reduction in size:

1 space:

[48885]      803 ms: Scavenge 109.4 (142.2) -> 106.3 (142.2) MB, 0 ms [Runtime::PerformGC].
memory usage: { rss: 140365824, heapTotal: 127084240, heapUsed: 117960976 }
memory usage: { rss: 140394496, heapTotal: 127084240, heapUsed: 118006704 }
str.replace takes 4 ms.

Tokenized space:

[48879]      880 ms: Mark-sweep 156.3 (190.6) -> 107.8 (148.2) MB, 41 ms [allocation failure] [GC in old space requested].
memory usage: { rss: 151355392, heapTotal: 160808928, heapUsed: 120760240 }
memory usage: { rss: 151359488, heapTotal: 160808928, heapUsed: 120818752 }
str.replace takes 2 ms.

/cc @redonkulus @kaesonho

[juandopazo]

wat

[ericf]

This is the max radix and the largest magnitude random number that produces an 8 char string. I need 8 chars so I can wrap it with @ special char to end up with a 10 char string which is the max length for space param to JSON.stringify().

[lingyan]

Thanks for all the efforts of optimizing this library!

I would suggest to make these configurable, so that people can benchmark with their own data set and decide what they want to do. An additional string.replace() is a little scary :)

Some people might actually want to serialize to a human readable format. It would be nice to allow configure the number of space for indentation.

The other thing to make configurable is whether to use the replacer, maybe a boolean option like isSimpleData. I'm sure there is a better name ;) That way, if people know their data set does not have regexp or function, they can turn replacer off to be much more performant.

[ericf]

This change was only made to get around the Node 0.10 V8 issue. I'm on board with making the space argument configurable, but adding isSimpleData goes against a primary reason this package exists and libraries or frameworks build on top of this would always set it to false because they don't know.

In the benchmarks the extra str.replace slows things down a bit but seemed within reason for the byte size savings.

[lingyan]

Sure. Ok with not adding isSimplData. Making indent and whether to do token-then-replace configurable would be great.

[ericf]

Closing this because this hack is going too far and the new plan is to revert the changes released in v1.1.0.