[tanjiroK-coder]

MultiPartName.unquote strips identifier quotes with ^["\[]+|["]]+$, and Table.setName and Column.setName call it whenever a name contains a dot, so it runs on identifiers straight out of CCJSqlParserUtil.parse. The trailing alternative has no left anchor, so replaceAll retries it at every offset inside a run of quote characters, and each attempt consumes the rest of the run before failing on $. A quoted identifier can legally hold any number of doubled "" escapes, which makes that run as long as the caller wants: parsing SELECT * FROM "a""""...""".b" takes 440 ms with 10k quotes, 3.9 s with 20k, and hits the parser timeout at 40k, while the same statements parse in under 10 ms with this change. I found it while looking for other regexes applied to token images after the earlier hint and server-name fixes. unquote and isQuoted now scan the leading and trailing quote characters directly, with the same character sets and the same leading-then-trailing order, so results do not change; LEADING_TRAILING_QUOTES_PATTERN stays because it is a public constant. The new tests compare both methods against the old pattern on a set of edge cases and parse the crafted statement under a 2 second preemptive timeout.

[manticore-projects]

Thank you!