in test https://github.com/FasterXML/jackson-benchmarks/blob/master/src/main/java/com/fasterxml/jackson/perf/json/JsonNoInternReadVanilla.java
it uses JsonFactory and it uses protected final transient ByteQuadsCanonicalizer _byteSymbolCanonicalizer = ByteQuadsCanonicalizer.createRoot(); that has hard-coded enabled intern = true - see createRoot(int seed) - therefore it does it not matter for test - INTERN_FIELD_NAMES id enabled or not On Saturday, November 19, 2016 at 6:20:18 PM UTC+1, Vladimir Dolzhenko wrote: > > Initially the issue https://github.com/FasterXML/jackson-core/issues/332 > has been raised - there is full investigation of the problem and how it > affects the app - explicitly and implicitly. > > My concerns that it is not a super big issue - rather inaccurate using of > internal jvm stuff w/o reason - basically misusage of it - I demonstrated > it using several tests that quite close to real one (at my current job, and > quite close to cases we used on my 2 prev jobs where we used json). > > Applying patch definitely would not gain 2x or 3x performance boost but > makes it a bit better - as string deduplication can trigger EscapeAnalysys > as well and those temporary strings will be dropped. > > In the same time - flag INTERN_FIELD_NAMES is turned on by default - I > pretty sure that people quite quite rear doing extra features tuning > (relying on default behaviour covers the most cases). > > In the ticket group of people joined me and my effort to stop using > String.intern - and they have quite similar uses cases (in terms of number > of unique property names). > > Sure - I think InternCache has to be rewritten later - using just 180 > fields and drop all of known strings is not the best strategy. > > My point or make it less suffer of String.intern or disable the default > behaviour of INTERN_FIELD_NAMES - or both of them. > -- You received this message because you are subscribed to the Google Groups "jackson-dev" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. For more options, visit https://groups.google.com/d/optout.
