Reduce memory allocations and improve performance in JSONObject

[basil]

Parsing https://updates.jenkins.io/update-center.json is extremely slow (hundreds of times slower than jq, for example). It consistently takes about 8 seconds and allocates about 170 GiB of RAM over the course of the parsing procedure. Profiling showed lots of regular expression compilation like

  java.lang.Thread.State: RUNNABLE
	at java.util.regex.Pattern.compile(java.base@11.0.5/Pattern.java:1757)
	at java.util.regex.Pattern.<init>(java.base@11.0.5/Pattern.java:1428)
	at java.util.regex.Pattern.compile(java.base@11.0.5/Pattern.java:1068)
	at net.sf.json.regexp.JdkRegexpMatcher.<init>(JdkRegexpMatcher.java:38)
	at net.sf.json.regexp.JdkRegexpMatcher.<init>(JdkRegexpMatcher.java:31)
	at net.sf.json.regexp.RegexpUtils.getMatcher(RegexpUtils.java:39)
	at net.sf.json.util.JSONTokener.matches(JSONTokener.java:111)
	at net.sf.json.JSONObject._fromJSONTokener(JSONObject.java:912)
	at net.sf.json.JSONObject.fromObject(JSONObject.java:156)
	at net.sf.json.util.JSONTokener.nextValue(JSONTokener.java:348)
	at net.sf.json.JSONArray._fromJSONTokener(JSONArray.java:1131)
	at net.sf.json.JSONArray.fromObject(JSONArray.java:125)
	at net.sf.json.util.JSONTokener.nextValue(JSONTokener.java:351)
	at net.sf.json.JSONObject._fromJSONTokener(JSONObject.java:955)
	at net.sf.json.JSONObject.fromObject(JSONObject.java:156)
	at net.sf.json.util.JSONTokener.nextValue(JSONTokener.java:348)
	at net.sf.json.JSONObject._fromJSONTokener(JSONObject.java:955)
	at net.sf.json.JSONObject.fromObject(JSONObject.java:156)
	at net.sf.json.util.JSONTokener.nextValue(JSONTokener.java:348)
	at net.sf.json.JSONObject._fromJSONTokener(JSONObject.java:955)
	at net.sf.json.JSONObject._fromString(JSONObject.java:1145)
	at net.sf.json.JSONObject.fromObject(JSONObject.java:162)
	at net.sf.json.JSONObject.fromObject(JSONObject.java:132)

and string allocation like

   java.lang.Thread.State: RUNNABLE
	at java.lang.String.<init>(String.java:207)
	at java.lang.String.substring(String.java:1933)
	at net.sf.json.util.JSONTokener.matches(JSONTokener.java:110)
	at net.sf.json.JSONObject._fromJSONTokener(JSONObject.java:912)
	at net.sf.json.JSONObject.fromObject(JSONObject.java:156)
	at net.sf.json.util.JSONTokener.nextValue(JSONTokener.java:348)
	at net.sf.json.JSONArray._fromJSONTokener(JSONArray.java:1131)
	at net.sf.json.JSONArray.fromObject(JSONArray.java:125)
	at net.sf.json.util.JSONTokener.nextValue(JSONTokener.java:351)
	at net.sf.json.JSONObject._fromJSONTokener(JSONObject.java:955)
	at net.sf.json.JSONObject.fromObject(JSONObject.java:156)
	at net.sf.json.util.JSONTokener.nextValue(JSONTokener.java:348)
	at net.sf.json.JSONObject._fromJSONTokener(JSONObject.java:955)
	at net.sf.json.JSONObject.fromObject(JSONObject.java:156)
	at net.sf.json.util.JSONTokener.nextValue(JSONTokener.java:348)
	at net.sf.json.JSONObject._fromJSONTokener(JSONObject.java:955)
	at net.sf.json.JSONObject._fromString(JSONObject.java:1145)
	at net.sf.json.JSONObject.fromObject(JSONObject.java:162)
	at net.sf.json.JSONObject.fromObject(JSONObject.java:132)

There are two issues here: repeatedly compiling a pattern where a simple .startsWith("null") would have sufficed, and repeatedly copying a massive string just to search a few characters in it. See flame graphs before and after.

I added a new unit test. This has also been shipping in production in our fork of json-lib to Jenkins users in 2.456 since May without any reported issues.

[aalmiray]

Thank you 😄