Blank space refused before a segment and in a slice: JSONPath parsing in OjG

$ .a and $[1:5 :2] were parse errors in OjG, a Go library for JSON, though RFC 9535 allows blank space before a path segment and between the parts of a slice. PR #258 accepts both. @ohler55 merged it about forty minutes after I opened it.

When it happens

_, err := jp.ParseString("$ .a")
fmt.Printf("%q\n", err)
// "parse error at 2 in $ .a"

$[1:5 :2] failed with invalid slice syntax at 7.

How the parser reads a path

nextFrag in jp/parse.go reads the next segment. A blank falls to its default case, which ends the expression. readSlice took the byte after the end value and after the second colon as it stood. After the first colon it skipped blanks, then read a number whatever came next, so $[1: ] tried to read a number at ].

Why the gap existed

The default case and readSlice date from June 2020, and readSlice read the bytes after the end value and the second colon raw in its first version. TestParse had blank space only inside brackets and around filter operators. Its one slice row, $[1\n:\t5], put it around the first colon.

The fix

Trimmed jp/parse.go diff: hunk headers cut to @@, no context lines, the second-colon change (two sites) shown once.

@@
+		if !first && !lastDescent && spaceMap[p.buf[p.pos]] == 's' {
+			// Blank space can come before a segment but not after the last one.
+			start := p.pos
+			if b := p.nextNonSpace(); b != '.' && b != '[' {
+				p.pos = start
+			}
+		}
@@
-		b = p.buf[p.pos]
-		p.pos++
+		b = p.skipSpace()
@@
-	} else {
+	} else if b != ']' {
@@
+		if spaceMap[b] == 's' {
+			b = p.skipSpace()
+		}

The last block is added after the end value and, one tab less deep, before the closing ]. Blank space is taken only when . or [ follows; otherwise nextFrag puts p.pos back.

The test

Twenty-one rows in TestParse. Twelve must parse: six in slices, three between segments, three in filters and function arguments. Nine must stay errors, among them $ *. Eight give the same error as before. $[1:5 3] now reports position 8 where it said 7.

What the maintainer did

@ohler55 approved it with no comment and squash-merged it into develop. The merged files match my branch.

The lesson

• Test an optional part when it is absent and blank space follows. $[1: ] has no end value, but the check for one ran before the blanks were skipped.
• A lookahead that skips must put the position back when it declines. Without p.pos = start, $ * parses as $.* and $.a b fails at 5, not 4.

If a parser of yours skips blank space before it knows what follows, reply on X.