Split out utf8_validate_cp(), adhere to RFC 3629

liberty.c

@@ -2753,6 +2753,11 @@ utf8_decode (const char **s, size_t len)
 
 	// Check the rest of the sequence
 	uint32_t cp = *p++ & ~mask;
+
+	// Overlong sequence (possibly MUTF-8, not supported)
+	if (!cp && sequence_len)
+		return -1;
+
 	while (sequence_len && --sequence_len)
 	{
 		if (p == end)
@@ -2765,6 +2770,13 @@ utf8_decode (const char **s, size_t len)
 	return cp;
 }
 
+static inline bool
+utf8_validate_cp (int32_t cp)
+{
+	// RFC 3629, CESU-8 not allowed
+	return cp >= 0 && cp <= 0x10FFFF && (cp < 0xD800 || cp > 0xDFFF);
+}
+
 /// Very rough UTF-8 validation, just makes sure codepoints can be iterated
 static bool
 utf8_validate (const char *s, size_t len)
@@ -2772,7 +2784,7 @@ utf8_validate (const char *s, size_t len)
 	const char *end = s + len;
 	int32_t codepoint;
 	while ((codepoint = utf8_decode (&s, end - s)) >= 0
-		&& codepoint <= 0x10FFFF /* TODO: better validations */)
+		&& utf8_validate_cp (codepoint))
 		;
 	return s == end;
 }

tests/liberty.c

@@ -331,10 +331,12 @@ test_utf8 (void)
 	soft_assert (utf8_decode (&partial, 1) == -2);
 	soft_assert (utf8_decode (&empty,   0) == -1);
 
-	const char valid  [] = "2H₂ + O₂ ⇌ 2H₂O, R = 4.7 kΩ, ⌀ 200 mm";
-	const char invalid[] = "\xf0\x90\x28\xbc";
-	soft_assert ( utf8_validate (valid,   sizeof valid));
-	soft_assert (!utf8_validate (invalid, sizeof invalid));
+	const char valid[] = "2H₂ + O₂ ⇌ 2H₂O, R = 4.7 kΩ, ⌀ 200 mm";
+	const char invalid_1[] = "\xf0\x90\x28\xbc";
+	const char invalid_2[] = "\xc0\x80";
+	soft_assert ( utf8_validate (valid,     sizeof valid));
+	soft_assert (!utf8_validate (invalid_1, sizeof invalid_1));
+	soft_assert (!utf8_validate (invalid_2, sizeof invalid_2));
 
 	struct utf8_iter iter = utf8_iter_make ("fóọ");
 	size_t ch_len;