DrRataplan / xq-lsp / 30019998305

93%

push

github

fix: exclude XP-only spec dependencies from QT4 3.1 filter

XQ31_COMPAT treated any "+"-suffixed XP token (XP20+, XP30+, XP31,
XP31+) as sufficient grounds to run a QT4 test case through our
XQuery analyzer. But per qt4tests' own dependency convention, "XP"
tokens mark a test as valid for a *bare XPath* evaluator (no Prolog,
no module) -- a different host language/context than what this tool
implements (a full XQuery 3.1 processor).

Every test case that was only included via an XP token (no
XQ-track alternative) and produced a false positive or false
negative turns out to assert something that's specifically an XPath
restriction, not an XQuery one -- e.g. typeswitch-in-xpath ("Typeswitch
disallowed in XPath") and LetExpr023 ("XPath (prior to 4.0) does not
allow for and let to be mixed"). Both use constructs that are
perfectly valid XQuery, so treating "XP-compatible" as "safe to run
through our XQuery analyzer" was testing a host language we don't
implement.

74 test cases in the corpus were included solely via an XP token: 55
were passing (ordinary function/regex-behavior tests we'll no longer
get QT4 regression coverage for, though nothing about their outcome
was wrong), 10 were false positives, and 9 were false negatives for
XPath-only restrictions that can never apply to an XQuery processor.

Net effect: false positives 69 -> 59, false negatives ~433 -> 426.
This fully resolves the remaining "&" in string literals (issue #57)
and revised "/" tokenization in path expressions (issue #60) false
positives, plus a stray namespace::-axis leak.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

1445 of 1708 branches covered (84.6%)

Branch coverage included in aggregate %.

5350 of 5577 relevant lines covered (95.93%)

324038.32 hits per line