JDK 9 RFR: 8176168: Performance drop due to SAXParser SymbolTable reset

[huizhe wang]

Hi Aleksej,

The changes look good. Some changes to the notes in the test may make it 
a bit clearer on what scenarios are tested: move the current notes for 
the method testResetEnabled to parseAndCheckReset, and then add scenario 
description to the three test* methods, for example, testNoFeatureSet 
verifies two scenarios: resetSymbolTable is and is not requested through 
the System property, while testResetEnabled tests that the feature is 
set  and testResetDisabled unset by using SAXParserFactory's setFeature 
method regardless of whether the System property is set.

Best regards,
Joe

On 4/16/2017 3:02 PM, Aleks Efimov wrote:
> Hi,
>
> Please, help to review the fix [1] for a performance regression in 
> SPECjvm2008-XML benchmark. The cause of this regression was the 
> unconditional reset of SAX parsers symbol table during each reset() 
> operation introduced by JDK-8173390.
> Proposed fix introduces new system property (CCC request is still 
> pending approval) that provides ability to control the symbol table 
> reset functionality. By default the reset table functionality is 
> disabled, similar to pre JDK-8173390 behavior. JAXWS parsers pool 
> implementation was updated to utilize new property to reset symbol 
> table only in JAXWS use-cases that helped to restore the performance 
> levels to pre JDK-8173390 level.
>
> Modified regression test and XML related JCK tests passes on build 
> with proposed changes.
>
> With Best Regards,
> Aleksei
>
> [1] http://cr.openjdk.java.net/~aefimov/8176168/00/