Fix error-code regression in trailing-comma rules and harden tests

The new "member ','" / "element ','" reduces are installed as bison
default actions in their states, so a scanner error token arriving
right after a comma (control character, invalid UTF-8, unpaired UTF-16
surrogate) executed the semantic action before the parse error was
detected, overwriting the scanner's specific error code with
JSON_ERROR_SYNTAX when the flag was unset. Preserve an already-set
errcode, mirroring php_json_yyerror(), so json_decode("[1,\x01]")
keeps reporting JSON_ERROR_CTRL_CHAR exactly as before this feature.

Also terminate // comments at a bare CR in addition to LF, matching
how the scanner already treats CR ("\r"? "\n" line terminators, CR as
whitespace) and how JSONC editors treat line comments.

Extend the tests to pin: the preserved error codes above, empty
containers whose only content is a comment, CRLF and bare-CR comment
termination, comment bodies being opaque to UTF-8 validation, a
comment in colon position, [1,} state mismatch, a comment between an
opener and a lone comma, JSON_THROW_ON_ERROR with trailing commas, and
CRLF line counting inside block comments.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TUEZ4aW8xZqzutZBAxX9UJ

Author: claude

ext/json/json_parser.y

@@ -122,7 +122,9 @@ members:
 	|	member ','
 			{
 				if (!(parser->scanner.options & PHP_JSON_ALLOW_TRAILING_COMMAS)) {
-					parser->scanner.errcode = PHP_JSON_ERROR_SYNTAX;
+					if (!parser->scanner.errcode) {
+						parser->scanner.errcode = PHP_JSON_ERROR_SYNTAX;
+					}
 					zval_ptr_dtor_nogc(&$1);
 					YYERROR;
 				}
@@ -187,7 +189,9 @@ elements:
 	|	element ','
 			{
 				if (!(parser->scanner.options & PHP_JSON_ALLOW_TRAILING_COMMAS)) {
-					parser->scanner.errcode = PHP_JSON_ERROR_SYNTAX;
+					if (!parser->scanner.errcode) {
+						parser->scanner.errcode = PHP_JSON_ERROR_SYNTAX;
+					}
 					zval_ptr_dtor_nogc(&$1);
 					YYERROR;
 				}

ext/json/json_scanner.re

@@ -123,7 +123,7 @@ std:
 	EXP     = ( INT | FLOAT ) [eE] [+-]? DIGIT+ ;
 	NL      = "\r"? "\n" ;
 	WS      = [ \t\r]+ ;
-	CMT_SL  = "//" [^\n\x00]* ;
+	CMT_SL  = "//" [^\r\n\x00]* ;
 	CMT_ML  = "/*" ( [^*\x00] | ( "*"+ [^*/\x00] ) )* "*"+ "/" ;
 	EOI     = "\000";
 	CTRL    = [\x00-\x1F] ;

ext/json/tests/json_decode_comments.phpt

@@ -23,7 +23,19 @@ decode_dump("{\"a\" /*x*/ : /*y*/ 1 /*z*/, // t\n \"b\":2}");
 decode_dump('"// not a comment"');
 decode_dump('"/* not a comment */"');
 decode_dump("//only\n// comments\n[]");
+// empty containers whose only content is a comment
+decode_dump("[/*x*/]");
+decode_dump("{/*x*/}");
+decode_dump("[ //x\n]");
+decode_dump("{ //x\n}");
+// CRLF and bare CR both terminate a line comment
+decode_dump("// crlf\r\n[1, 2]");
+decode_dump("// bare cr\r1");
+// comment bodies are opaque: bytes that are not valid UTF-8 are skipped
+decode_dump("/* \xFF */1");
 var_dump(json_validate("// c\n{\"a\": [1, 2]}", 512, JSON_ALLOW_COMMENTS));
+var_dump(json_validate("{/*x*/}", 512, JSON_ALLOW_COMMENTS));
+var_dump(json_validate("/* \xFF */{}", 512, JSON_ALLOW_COMMENTS));
 
 ?>
 --EXPECT--
@@ -51,4 +63,22 @@ string(16) "// not a comment"
 string(19) "/* not a comment */"
 array(0) {
 }
+array(0) {
+}
+array(0) {
+}
+array(0) {
+}
+array(0) {
+}
+array(2) {
+  [0]=>
+  int(1)
+  [1]=>
+  int(2)
+}
+int(1)
+int(1)
+bool(true)
+bool(true)
 bool(true)

ext/json/tests/json_decode_comments_errors.phpt

@@ -27,6 +27,8 @@ decode_dump("// x\n");
 // comments separate tokens but do not join values
 decode_dump("1 2", 0);
 decode_dump("1/*c*/2");
+// a comment does not replace a required colon
+decode_dump("{\"a\" /*c*/ 1}");
 
 try {
     json_decode("/*", true, 512, JSON_ALLOW_COMMENTS | JSON_THROW_ON_ERROR);
@@ -62,4 +64,6 @@ NULL
 4: Syntax error near location 1:3
 NULL
 4: Syntax error near location 1:7
+NULL
+4: Syntax error near location 1:12
 JsonException: 4 Syntax error near location 1:1

ext/json/tests/json_decode_comments_trailing_flags_off.phpt

@@ -20,6 +20,11 @@ decode_dump("[1, 2,]");
 // one flag does not enable the other relaxation
 decode_dump("[1,]", JSON_ALLOW_COMMENTS);
 decode_dump("[1, /*c*/ 2]", JSON_ALLOW_TRAILING_COMMAS);
+// scanner errors right after a comma keep their specific error codes
+decode_dump("[1,\x01]");
+decode_dump("[1,\x80]");
+decode_dump("[1,\"\\ud834\"]");
+decode_dump("{\"a\":1,\x01}");
 // validate behaves identically
 var_dump(json_validate("// x\n1", 512, JSON_ALLOW_TRAILING_COMMAS));
 var_dump(json_validate("[1,]", 512, JSON_ALLOW_COMMENTS));
@@ -44,5 +49,13 @@ NULL
 4: Syntax error near location 1:4
 NULL
 4: Syntax error near location 1:5
+NULL
+3: Control character error, possibly incorrectly encoded near location 1:4
+NULL
+5: Malformed UTF-8 characters, possibly incorrectly encoded near location 1:4
+NULL
+10: Single unpaired UTF-16 surrogate in unicode escape near location 1:4
+NULL
+3: Control character error, possibly incorrectly encoded near location 1:8
 bool(false)
 bool(false)

ext/json/tests/json_decode_trailing_commas.phpt

@@ -26,6 +26,20 @@ decode_dump("[1,,2]");
 decode_dump("{\"a\":1,,}");
 // mismatched closer after a trailing comma
 decode_dump("{\"a\":1,]");
+decode_dump("[1,}");
+// a comment between the opener and a lone comma does not make it valid
+decode_dump("[/*c*/,]", JSON_ALLOW_COMMENTS | JSON_ALLOW_TRAILING_COMMAS);
+// trailing commas and JSON_THROW_ON_ERROR
+try {
+    var_dump(json_decode("[1,]", true, 512, JSON_ALLOW_TRAILING_COMMAS | JSON_THROW_ON_ERROR));
+} catch (JsonException $e) {
+    echo "JsonException: ", $e->getCode(), " ", $e->getMessage(), "\n";
+}
+try {
+    json_decode("[1,,]", true, 512, JSON_ALLOW_TRAILING_COMMAS | JSON_THROW_ON_ERROR);
+} catch (JsonException $e) {
+    echo "JsonException: ", $e->getCode(), " ", $e->getMessage(), "\n";
+}
 // validate mirrors decode
 var_dump(json_validate("[1,]", 512, JSON_ALLOW_TRAILING_COMMAS));
 var_dump(json_validate("{\"a\":1,}", 512, JSON_ALLOW_TRAILING_COMMAS));
@@ -95,6 +109,15 @@ NULL
 4: Syntax error near location 1:8
 NULL
 2: State mismatch (invalid or malformed JSON) near location 1:8
+NULL
+2: State mismatch (invalid or malformed JSON) near location 1:4
+NULL
+4: Syntax error near location 1:7
+array(1) {
+  [0]=>
+  int(1)
+}
+JsonException: 4 Syntax error near location 1:4
 bool(true)
 bool(true)
 bool(false)

ext/json/tests/json_last_error_msg_error_location_011.phpt

@@ -41,6 +41,10 @@ json_validate_trycatchdump($json, 512, JSON_ALLOW_COMMENTS);
 echo "\nError on line 1, column 13:\n";
 json_validate_trycatchdump("[1, /* x */ ,]", 512, JSON_ALLOW_COMMENTS | JSON_ALLOW_TRAILING_COMMAS);
 
+// CRLF line endings inside a block comment count as one line each.
+echo "\nError on line 3, column 8:\n";
+json_validate_trycatchdump("[\r\n/* x\r\ny */ 1 2]", 512, JSON_ALLOW_COMMENTS);
+
 ?>
 --EXPECT--
 Error on line 4, column 20:
@@ -72,3 +76,8 @@ Error on line 1, column 13:
 bool(false)
 int(4)
 string(31) "Syntax error near location 1:13"
+
+Error on line 3, column 8:
+bool(false)
+int(4)
+string(30) "Syntax error near location 3:8"