From e3229dd57a8b14703afc047b7b0cb076fdf813ee Mon Sep 17 00:00:00 2001
From: KBS <youdie006@naver.com>
Date: Thu, 10 Sep 2026 12:12:30 +0900
Subject: [PATCH] Quote strings that begin or end with non-ASCII whitespace
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

The decoder strips quoteless strings with strings.TrimSpace, which is
Unicode-aware, but needsQuotes anchors on Go RE2 \s, which is ASCII
only. A value starting or ending with U+00A0, U+1680, U+2000-U+200A,
U+205F or U+3000 was therefore written quoteless and came back short:

  Marshal(map[string]interface{}{"k": " ab"})
  // {
  //   k:  ab
  // }
  Unmarshal of that output gives "ab"

needsQuotes already enumerates non-ASCII code points by hand for exactly
this reason; the list just stops short of the spaces TrimSpace removes.
Test with the same function the decoder uses, so the two cannot drift.

hjson-js quotes all of these.
---
 encode.go      |  3 +++
 encode_test.go | 34 ++++++++++++++++++++++++++++++++++
 2 files changed, 37 insertions(+)

diff --git a/encode.go b/encode.go
index 3345a0b..82d2f14 100644
--- a/encode.go
+++ b/encode.go
@@ -134,6 +134,9 @@ func (e *hjsonEncoder) quote(value string, separator string, isRootObject bool,
 	} else if e.QuoteAlways ||
 		hasCommentAfter ||
 		needsQuotes.MatchString(value) ||
+		// The parser trims quoteless strings with strings.TrimSpace, which also
+		// removes non-ASCII whitespace that the needsQuotes regexp does not match.
+		len(value) != len(strings.TrimSpace(value)) ||
 		(e.QuoteAmbiguousStrings && (startsWithNumber([]byte(value)) ||
 			startsWithKeyword.MatchString(value))) {
 
diff --git a/encode_test.go b/encode_test.go
index 9e1480f..acb4ec8 100644
--- a/encode_test.go
+++ b/encode_test.go
@@ -875,3 +875,37 @@ func TestStructComment(t *testing.T) {
 		t.Errorf("Expected:\n%s\nGot:\n%s\n\n", expected, string(h))
 	}
 }
+
+func TestQuoteUnicodeWhitespace(t *testing.T) {
+	// The parser trims quoteless strings with strings.TrimSpace, so the encoder
+	// must quote strings that start or end with a rune unicode.IsSpace reports.
+	tests := []struct {
+		r          rune
+		wantQuoted bool
+	}{
+		{'\u0085', true}, {'\u00a0', true}, {'\u1680', true}, {'\u2000', true},
+		{'\u200a', true}, {'\u2028', true}, {'\u2029', true}, {'\u202f', true},
+		{'\u205f', true}, {'\u3000', true},
+		{'\u200b', false}, {'\u180e', false},
+	}
+	for _, tt := range tests {
+		for _, value := range []string{string(tt.r) + "ab", "ab" + string(tt.r)} {
+			b, err := Marshal(map[string]interface{}{"k": value})
+			if err != nil {
+				t.Errorf("%U: Marshal: %v", tt.r, err)
+				continue
+			}
+			if quoted := bytes.Contains(b, []byte(`k: "`)); quoted != tt.wantQuoted {
+				t.Errorf("%U: quoted %v, want %v (encoded as %q)", tt.r, quoted, tt.wantQuoted, b)
+			}
+			var dst map[string]interface{}
+			if err := Unmarshal(b, &dst); err != nil {
+				t.Errorf("%U: Unmarshal(%q): %v", tt.r, b, err)
+				continue
+			}
+			if got := dst["k"]; got != value {
+				t.Errorf("%U: got %q, want %q (encoded as %q)", tt.r, got, value, b)
+			}
+		}
+	}
+}