On Feb 5, 2015, at 8:00 PM, Xueming Shen <[email protected]> wrote:
> Hi, > > Please help review the fix for #8030179 > > issue: https://bugs.openjdk.java.net/browse/JDK-8030179 > webrev: http://cr.openjdk.java.net/~sherman/8030179/webrev > > This is the regression bug introduced in jdk7 when trying to optimize the > single > byte encoding loop, in which the "optimization" code inappropriately updates > the > "sl" (source limit) value and triggers misbehavior of the sgp parser (in > which it > mistakenly returns "underflow" when it sees a high surrogate but can't see > the next > low surrogate, because the "sl" is changed...) > The fix looks good, but i would like to suggest some changes to the test. We can use TestNG to avoid pulling in a bunch of infrastructure methods. The test cases can be produced by a data provider and we can do a cross product with char sets and character input (for example i have managed to induced stack overflows by just passing sequences of high surrogates). Using the data provider allows the test infrastrcture to report/crunch those tests individually rather than being reported via System.out. I can send a patch to the test if that helps. Paul.
