syntax: allow [ glob in array elements

Inside `arrayElems`, the lexer always tokenized `[` as `leftBrack`, so
the parser treated `foo[0-9]` and `"foo"[0-9]` inside `a=( ... )` as
malformed `[idx]=val` assignments. Only emit `leftBrack` when `[`
begins a new element (after whitespace, `(`, or a newline); otherwise
fall through to `advanceLitNone` so it joins the preceding word as a
glob.

Fixes #1322.

Author: mvdan

syntax/filetests_test.go

@@ -4922,6 +4922,24 @@ var fileTests = []fileTestCase{
 			}},
 		}}}, LangBash),
 	),
+	fileTest(
+		[]string{`a=(foo[0-9])`},
+		langFile(&CallExpr{Assigns: []*Assign{{
+			Name: lit("a"),
+			Array: &ArrayExpr{Elems: []*ArrayElem{
+				{Value: word(lit("foo"), lit("[0-9]"))},
+			}},
+		}}}, LangBash|LangMirBSDKorn|LangZsh),
+	),
+	fileTest(
+		[]string{`a=("foo"[0-9])`},
+		langFile(&CallExpr{Assigns: []*Assign{{
+			Name: lit("a"),
+			Array: &ArrayExpr{Elems: []*ArrayElem{
+				{Value: word(dblQuoted(lit("foo")), lit("[0-9]"))},
+			}},
+		}}}, LangBash|LangMirBSDKorn|LangZsh),
+	),
 	fileTest(
 		[]string{"a]b"},
 		langFile(litStmt("a]b")),

syntax/lexer.go

@@ -325,7 +325,9 @@ skipSpace:
 			}
 			p.next()
 		case '[':
-			if p.quote == arrayElems {
+			// `[` only starts an `[idx]=val` element when it begins a new word;
+			// otherwise it continues a glob like `foo[0-9]`.
+			if p.quote == arrayElems && (p.spaced || p.tok == leftParen || p.tok == _Newl) {
 				p.rune()
 				p.tok = leftBrack
 			} else {