On Wed, 16 Feb 2022 18:45:29 GMT, Ian Graves <igra...@openjdk.org> wrote:
> This is a fix in the buggy way CIBackRef traverses unicode characters that
> could be variable-length. Originally it followed the approach that BackRef
> does, but failed to account for unicode characters that could be 2
> chars-long. The upper bound (groupSize) for the traversing loop is set by the
> difference between group start and stop indexes. This works for single char
> characters and it also works for case-sensitive comparisons because
> byte-by-byte comparisons are acceptable, but it doesn't work for a comparison
> where some kind of normalization (i.e. case) is required. This fix adjusts
> the upper bound for the loop that traverses the character when a two-char
> character is encountered.
>
> An alternative was to check the length of the group size by scanning the
> group in advance and converting to code points, but this could potentially
> result in multiple scans and codepoint conversions of the same matcher group
> which could be long. The solution that adjusts the loop bounds on the fly
> avoids this case.
Looks good with a minor suggestion.
src/java.base/share/classes/java/util/regex/Pattern.java line 5104:
> 5102: j += Character.charCount(c2);
> 5103:
> 5104: if(xIncr > 1) {
You can eliminate `xIncr` by comparing `c1 >=
Character.MIN_SUPPLEMENTARY_CODE_POINT` here.
-------------
PR: https://git.openjdk.java.net/jdk/pull/7501
