Lift Json Serialization write optimization and patch

[Chris Webster]

I was profiling an application using lift-json (2.6.2) and the profiler was showing a significant portion of time was spent in JsonAST.appendEscapedString (see https://github.com/lift/framework/blob/master/core/json/src/main/scala/net/liftweb/json/JsonAST.scala#L864). The StringBuilder.append method invoked was StringBuilder.append(Object) as both String and Char are being returned. The implementation of StringBuilder.append(Object) when applied to a Character is essentially:

* call Character.toString - which produces a one element char[] and then creates a new String with the char array, which in turn does a defensive copy of the buffer.

* the string created about is used to add to the StringBuilder

Here is a change to JsonAST.appendEscapedString to use the StringBuilder.append(Char) which directly inserts the character into the buffer avoiding the operations described above (https://github.com/lift/framework/compare/master...chriswebster:appendStringOp?expand=1). The patch attempts to minimize the change to the code but get the performance gain, by ensuring the StringBuilder.append(Character) method is used when copying unconverted characters. 

Here is a microbenchmark (https://github.com/chriswebster/lift-jsonBenchmark/blob/master/src/main/scala/LiftJsonBenchmark.scala) demonstrating the performance gain. 

Current code:

285499.476 ±(99.9%) 5230.757 ops/s [Average]

[info]   (min, avg, max) = (279162.240, 285499.476, 289016.681), stdev = 3459.822

[info]   CI (99.9%): [280268.719, 290730.233] (assumes normal distribution)

After change:

545955.912 ±(99.9%) 9927.185 ops/s [Average]

[info]   (min, avg, max) = (533569.676, 545955.912, 556373.272), stdev = 6566.219

[info]   CI (99.9%): [536028.727, 555883.098] (assumes normal distribution)

Which represents about a 90% performance increase when writing values. Please consider this for inclusion into the lift json code base. 

Chris 

[Antonio Salazar Cardozo]

I think we could accept a pull request for this.

One question: if case c: Char is used, does that not select the correct overload

of StringBuilder.append?

Thanks,

Antonio

[Chris Webster]

Hi, 

  The javap output from case c:Char and case c for the invocation of StringBuilder.append is the same (both generate StringBuilder.append(Object) invocations). I captured the output here:

https://gist.github.com/chriswebster/dec25f150fbba24ceba2 . Let me know if pull request is ok. 

Chris 

[Matt Farmer]

I, too, think this is small enough that we could accept a PR for. It represents a significant benefit to other developers.

—
Matt Farmer | Blog | Twitter

--
--
Lift, the simply functional web framework: http://liftweb.net
Code: http://github.com/lift
Discussion: http://groups.google.com/group/liftweb
Stuck? Help us help you: https://www.assembla.com/wiki/show/liftweb/Posting_example_code

---
You received this message because you are subscribed to the Google Groups "Lift" group.
To unsubscribe from this group and stop receiving emails from it, send an email to liftweb+u...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Chris Webster]

Thanks!

 PR submitted https://github.com/lift/framework/pull/1706

Chris