An empty pattern that matched nothing: match() and search() in OjG

An empty pattern, '', made match() and search() select nothing in OjG, a Go library for JSON. The empty string is a valid I-Regexp (RFC 9485), so RFC 9535 says match(@, '') selects "" and nothing else, and search(@, '') selects every string. PR #257 fixes both. @ohler55 merged it about one hour and thirteen minutes after I opened it.

When it happens

data := []any{"", "a"}
fmt.Printf("%q\n", jp.MustParseString(`$[?match(@, '')]`).Get(data))
fmt.Printf("%q\n", jp.MustParseString(`$[?search(@, '')]`).Get(data))
// []
// []

These should print [""] and ["" "a"].

How the pattern check works

evalStack in jp/script.go reads the pattern with rs, _ := right.(string) and runs the match only when 0 < len(rs). The _ drops the ok that says whether the value is a string. So the length test had two jobs: skip a pattern that is not a string, and, as a side effect, skip ''.

Why the gap existed

The match and search cases have had that guard since March 2023. Back then match read the first and last byte of the pattern to find anchors, which needs a non-empty pattern. In October 2026 that code was replaced and the guard stayed. search never read those bytes. No match or search row in TestScriptEval used an empty pattern.

The fix

Trimmed jp/script.go diff, hunk header cut to @@. match and search each have this pair of lines; one is shown.

@@
-				if rs, _ := right.(string); 0 < len(rs) {
+				if rs, ok := right.(string); ok {

For a non-empty string, 0 < len(rs) and ok are both true. For a pattern that is not a string, or is missing, both are false, so it still gives no match. Only '' changes. match wraps it as ^(?:)$, which matches only "". search compiles an empty regexp, which matches every string.

The test

Six rows in TestScriptEval. Without the fix three fail: match of "", and search of "" and of "abc". Three pass either way: match of "a", a number pattern against "", and a missing search pattern against "abc". go test ./jp -run 'TestScriptEval$' -count=3 -v failed 3 of 3 before the fix and passed 3 of 3 after.

What the maintainer did

@ohler55 approved it and squash-merged it into develop. The merged files match my branch.

The lesson

• A length test on a value from a type assertion does two jobs. rs, _ := x.(string); 0 < len(rs) asks “is it a string” and “is it non-empty” at once. Test ok for the first.
• Give the wrong-type case a row an empty pattern would match. With the guard made always true, match(@.x, 5) against "" fails. The same row against "a" would not.
• When the code a guard protected is replaced, read the guard again.

If a function of yours guards a type assertion with len, reply on X.