Return false from matches() when the node-set argument selects nothing

[youdie006]

starts-with, ends-with and contains all return false when their node-set argument selects no node (func.go:354, :381, :408). matches() at func.go:436 returned "", so a boolean function handed back a string.

notFunc (func.go:638) has no case string and falls through to default: return false, so the string is swallowed rather than negated:

not(matches(//no-such, 'x'))       false     <- the three siblings all give true
not(starts-with(//no-such, 'x'))   true
not(contains(//no-such, 'x'))      true
not(ends-with(//no-such, 'x'))     true

boolean(matches(//no-such, 'x'))   false     <- these two are the same X
not(not(matches(//no-such, 'x')))  true

In practice //a[not(matches(@href, '^https'))] drops every <a> that has no href, while the starts-with spelling keeps them. A missing attribute reaches the same branch.

It went unnoticed because asBool (func.go:257) maps "" to false, so a bare predicate like //*[matches(@nope,'x')] is corrected on the way out. not() is the caller that is not defensive.

There is a second way to close this, and it is worth saying out loud: leaving the early return out and letting the regex run against "" also removes the string return, and is arguably closer to string(empty node-set) being "". It differs observably - matches(//no-such, '^$') becomes true rather than false. I went with false because it is what the three sibling functions already do, but the type bug is there under either, so say the word if you prefer the other.

Verification

Base 487996fb. New case sits next to Test_func_matches in xpath_function_test.go; no new file.

row	func.go md5	-run Test_func_matches
pristine	8c32ba3520ab162f1c4a2171b389a9a5	FAIL (Test_func_matches_empty_nodeset)
this PR	fce2401a8e2c34dc9e76ec94f464dea0	ok
return true instead	f1d1a2f9698a37defb59a11fe24598b3	FAIL
revert	8c32ba3520ab162f1c4a2171b389a9a5	FAIL

The pre-existing Test_func_matches passes in every row, including pristine - it never reaches the nil-node branch, so nothing in the suite currently pins this either way. The over-correction is caught only by the new case; I am not claiming it breaks an existing test.

go test ./... ok, go vet ./... silent, gofmt -l clean for both changed files. gofmt -l . does list func_go110.go and func_pre_go110.go, which are unformatted on 487996fb as well and are untouched here.

The same four-line case query: block appears in normalizespaceFunc at func.go:463; "" is right there, matching substringFunc and the convention #136 established for the string functions. Only matchesFunc is changed.

Written with AI assistance (Claude); the measurements above were run locally and I have reviewed the change.

[coveralls]

coverage: 84.163% (+0.4%) from 83.798% — youdie006:matches-empty-nodeset into antchfx:master