decoder: Use faster ondemand parser for simdjson_key_count method

Author: JakubOnderka

php_simdjson.cpp

@@ -324,10 +324,6 @@ PHP_FUNCTION(simdjson_key_count) {
         Z_PARAM_BOOL(throw_if_uncountable)
     ZEND_PARSE_PARAMETERS_END();
 
-    if (!simdjson_validate_depth(depth, 3)) {
-        RETURN_THROWS();
-    }
-
     simdjson_php_error_code error;
     if (SIMDJSON_SHOULD_REUSE_PARSER(ZSTR_LEN(json))) {
         error = php_simdjson_key_count(simdjson_get_reused_parser(), json, ZSTR_VAL(key), return_value, depth);

simdjson.stub.php

@@ -286,14 +286,13 @@ function simdjson_key_value(string $json, string $key, bool $associative = false
  *
  * @param string $json The JSON string being decoded
  * @param string $key The JSON pointer being requested
- * @param int $depth The maximum nesting depth of the structure being decoded.
+ * @param int $depth Not used anymore
  * @param bool $throw_if_uncountable If true, then throw SimdJsonException instead of
  *                                   returning 0 for JSON pointers
  *                                   to values that are neither objects nor arrays.
  * @return int
  * @throws SimdJsonDecoderException for invalid JSON or invalid JSON pointer
  *                           (or document over 4GB, or out of range integer/float)
- * @throws ValueError for invalid $depth
  * @see https://www.rfc-editor.org/rfc/rfc6901.html
  */
 function simdjson_key_count(string $json, string $key, int $depth = 512, bool $throw_if_uncountable = false): int {}

simdjson_arginfo.h

@@ -1,5 +1,5 @@
 /* This is a generated file, edit the .stub.php file instead.
- * Stub hash: 21c3b21c513b1ce2b8c72912856cd6d7326dbfe5 */
+ * Stub hash: d30184e8da243c7681539c2f70cb0afbd97ca02d */
 
 ZEND_BEGIN_ARG_WITH_RETURN_TYPE_INFO_EX(arginfo_simdjson_validate, 0, 1, _IS_BOOL, 0)
 	ZEND_ARG_TYPE_INFO(0, json, IS_STRING, 0)

src/simdjson_decoder.cpp

@@ -57,6 +57,14 @@ get_key_with_optional_prefix(simdjson::dom::element &doc, std::string_view json_
     return doc.at_pointer(std_pointer);
 }
 
+static inline simdjson::simdjson_result<simdjson::ondemand::value>
+get_key_with_optional_prefix_ondemand(simdjson::ondemand::document &doc, std::string_view json_pointer) {
+    /* https://www.rfc-editor.org/rfc/rfc6901.html */
+    /* TODO: Deprecate in a subsequent minor release and remove in a major release to comply with the standard. */
+    auto std_pointer = ((!json_pointer.empty() && json_pointer[0] != '/') ? "/" : "") + std::string(json_pointer.begin(), json_pointer.end());
+    return doc.at_pointer(std_pointer);
+}
+
 // Initialize stdClass object and return pointer to propertires HashTable
 static zend_always_inline HashTable* simdjson_init_object(zval *zv, uint32_t size) {
 #if PHP_VERSION_ID >= 80300
@@ -130,7 +138,7 @@ static zend_always_inline HashTable* simdjson_init_mixed_array(zval *zv, uint32_
     return ht;
 }
 
-/** Check if it is necessary to reallocate string to buffer */
+// Check if it is necessary to reallocate string to buffer
 static zend_always_inline bool simdjson_realloc_needed(const zend_string *json) {
     // it is not possible to check allocated size for persistent or permanent string
     bool is_persistent_or_permanent = GC_FLAGS(json) & (IS_STR_PERSISTENT | IS_STR_PERMANENT);
@@ -148,6 +156,14 @@ static zend_always_inline bool simdjson_realloc_needed(const zend_string *json)
     return free_space < simdjson::SIMDJSON_PADDING;
 }
 
+static simdjson::padded_string_view simdjson_padded_string_view(const zend_string *json) {
+    if (simdjson_realloc_needed(json)) {
+        return simdjson::padded_string(ZSTR_VAL(json), ZSTR_LEN(json));
+    } else {
+        return simdjson::padded_string_view(ZSTR_VAL(json), ZSTR_LEN(json), ZSTR_LEN(json) + simdjson::SIMDJSON_PADDING);
+    }
+}
+
 /** Decoded string from JSON must be always UTF-8 valid, so we can provide proper flag to zend_string */
 static zend_always_inline zend_string* simdjson_string_init(const char* buf, size_t len) {
     zend_string *str = zend_string_init(buf, len, 0);
@@ -649,52 +665,29 @@ PHP_SIMDJSON_API simdjson_php_error_code php_simdjson_key_exists(simdjson_php_pa
     return get_key_with_optional_prefix(doc, key).error();
 }
 
-/* }}} */
-
-PHP_SIMDJSON_API simdjson_php_error_code php_simdjson_key_count(simdjson_php_parser* parser, const zend_string *json, const char *key, zval *return_value, size_t depth) /* {{{ */ {
-    simdjson::dom::element doc;
-    simdjson::dom::element element;
-
-    SIMDJSON_PHP_TRY(build_parsed_json_cust(parser, doc, ZSTR_VAL(json), ZSTR_LEN(json), simdjson_realloc_needed(json), depth));
+PHP_SIMDJSON_API simdjson_php_error_code php_simdjson_key_count(simdjson_php_parser* parser, const zend_string *json, const char *key, zval *return_value, size_t depth) {
+    simdjson::ondemand::document doc;
+    simdjson::ondemand::value value;
+    simdjson::ondemand::json_type type;
 
-    SIMDJSON_PHP_TRY(get_key_with_optional_prefix(doc, key).get(element));
+    SIMDJSON_PHP_TRY(parser->ondemand_parser.iterate(simdjson_padded_string_view(json)).get(doc));
+    SIMDJSON_PHP_TRY(get_key_with_optional_prefix_ondemand(doc, key).get(value));
+    SIMDJSON_PHP_TRY(value.type().get(type));
 
-    zend_long key_count;
-    switch (element.type()) {
-        //ASCII sort
-        case simdjson::dom::element_type::ARRAY : {
-            auto json_array = element.get_array().value_unsafe();
-            key_count = zend_long(json_array.size());
-            if (UNEXPECTED(key_count == 0xFFFFFF)) {
-                /* The C simdjson library represents array sizes larger than 0xFFFFFF as 0xFFFFFF. */
-                key_count = 0;
-                for (auto it: json_array)  {
-                    (void)it;
-                    key_count++;
-                }
-                ZEND_ASSERT(key_count >= 0xFFFFFF);
-            }
+    size_t key_count;
+    switch (type) {
+        case simdjson::ondemand::json_type::array:
+            SIMDJSON_PHP_TRY(value.count_elements().get(key_count));
             break;
-        }
-        case simdjson::dom::element_type::OBJECT : {
-            auto json_object = element.get_object().value_unsafe();
-            key_count = zend_long(json_object.size());
-            if (UNEXPECTED(key_count == 0xFFFFFF)) {
-                /* The C simdjson library represents object sizes larger than 0xFFFFFF as 0xFFFFFF. */
-                key_count = 0;
-                for (auto it: json_object) {
-                    (void)it;
-                    key_count++;
-                }
-                ZEND_ASSERT(key_count >= 0xFFFFFF);
-            }
+
+        case simdjson::ondemand::json_type::object:
+            SIMDJSON_PHP_TRY(value.count_fields().get(key_count));
             break;
-        }
+
         default:
             return SIMDJSON_PHP_ERR_KEY_COUNT_NOT_COUNTABLE;
     }
+
     ZVAL_LONG(return_value, key_count);
     return simdjson::SUCCESS;
 }
-
-/* }}} */

src/simdjson_decoder_defs.h

@@ -22,6 +22,7 @@
 struct simdjson_php_parser {
 public:
     simdjson::dom::parser parser;
+    simdjson::ondemand::parser ondemand_parser;
     HashTable dedup_key_strings;
 };
 

tests/decode_max_depth.phpt

@@ -9,37 +9,31 @@ require_once __DIR__ . '/dump.inc';
 
 foreach ([0, PHP_INT_MIN, 1024, PHP_INT_MAX >> 1, PHP_INT_MAX] as $depth) {
     dump(function () use ($depth) { return simdjson_decode('[]', true, $depth); });
-    dump(function () use ($depth) { return simdjson_key_count('{"a":"b"}', 'a', $depth); });
     dump(function () use ($depth) { return simdjson_key_value('{"a":{}}', 'a', true, $depth); });
     dump(function () use ($depth) { return simdjson_key_exists('{"a":{}}', 'a', $depth); });
     dump(function () use ($depth) { return simdjson_is_valid('{}', $depth); });
 }
 ?>
 --EXPECTF--
 Caught ValueError: simdjson_decode(): Argument #3 ($depth) must be greater than zero
-Caught ValueError: simdjson_key_count(): Argument #3 ($depth) must be greater than zero
 Caught ValueError: simdjson_key_value(): Argument #4 ($depth) must be greater than zero
 Caught ValueError: simdjson_key_exists(): Argument #3 ($depth) must be greater than zero
 Caught ValueError: simdjson_is_valid(): Argument #2 ($depth) must be greater than zero
 Caught ValueError: simdjson_decode(): Argument #3 ($depth) must be greater than zero
-Caught ValueError: simdjson_key_count(): Argument #3 ($depth) must be greater than zero
 Caught ValueError: simdjson_key_value(): Argument #4 ($depth) must be greater than zero
 Caught ValueError: simdjson_key_exists(): Argument #3 ($depth) must be greater than zero
 Caught ValueError: simdjson_is_valid(): Argument #2 ($depth) must be greater than zero
 array(0) {
 }
-int(0)
 array(0) {
 }
 bool(true)
 bool(true)
 Caught ValueError: simdjson_decode(): Argument #3 ($depth) exceeds maximum allowed value of %d
-Caught ValueError: simdjson_key_count(): Argument #3 ($depth) exceeds maximum allowed value of %d
 Caught ValueError: simdjson_key_value(): Argument #4 ($depth) exceeds maximum allowed value of %d
 Caught ValueError: simdjson_key_exists(): Argument #3 ($depth) exceeds maximum allowed value of %d
 Caught ValueError: simdjson_is_valid(): Argument #2 ($depth) exceeds maximum allowed value of %d
 Caught ValueError: simdjson_decode(): Argument #3 ($depth) exceeds maximum allowed value of %d
-Caught ValueError: simdjson_key_count(): Argument #3 ($depth) exceeds maximum allowed value of %d
 Caught ValueError: simdjson_key_value(): Argument #4 ($depth) exceeds maximum allowed value of %d
 Caught ValueError: simdjson_key_exists(): Argument #3 ($depth) exceeds maximum allowed value of %d
 Caught ValueError: simdjson_is_valid(): Argument #2 ($depth) exceeds maximum allowed value of %d

tests/decode_max_depth_memory_reduction.phpt

@@ -12,8 +12,6 @@ foreach ([1024, 1 << 27] as $depth) {
     echo "Test depth=$depth:\n";
     $value = simdjson_decode('[]', true, $depth);
     var_dump($value);
-    $value = simdjson_key_count('{"a":"b"}', 'a', $depth);
-    var_dump($value);
     try {
         simdjson_decode(str_repeat('[', 200000) . str_repeat(']', 199999), true, $depth);
         echo "should be invalid\n";
@@ -26,10 +24,8 @@ foreach ([1024, 1 << 27] as $depth) {
 Test depth=1024:
 array(0) {
 }
-int(0)
 Caught SimdJsonDecoderException: The JSON document was too deep (too many nested objects and arrays)
 Test depth=134217728:
 array(0) {
 }
-int(0)
 Caught SimdJsonDecoderException: The JSON document has an improper structure: missing or superfluous commas, braces, missing keys, etc.

tests/depth.phpt

@@ -8,7 +8,6 @@ error_reporting=E_ALL
 require_once __DIR__ . '/dump.inc';
 
 // too small/large depths tested in decode_max_depth
-var_dump(simdjson_key_count('[]', '', 1));
 var_dump(simdjson_decode('[]', true, 1));
 echo "Test '[1]'\n";
 try {
@@ -30,7 +29,6 @@ var_dump(simdjson_decode('[[]]', true, 2));
 var_dump(simdjson_decode('[[1]]', true, 3));
 ?>
 --EXPECTF--
-int(0)
 array(0) {
 }
 Test '[1]'

tests/key_count_exception.phpt

@@ -4,12 +4,6 @@ simdjson_key_count throws exception if key was not found test
 <?php
 $json = file_get_contents(__DIR__ . DIRECTORY_SEPARATOR . '_files' . DIRECTORY_SEPARATOR . 'result.json');
 
-try {
-    \simdjson_key_count($json, "unknown", 1);
-} catch (\RuntimeException $e) {
-    var_dump($e->getMessage());
-}
-
 try {
     \simdjson_key_count($json, "unknown");
 } catch (\RuntimeException $e) {
@@ -18,5 +12,4 @@ try {
 
 ?>
 --EXPECT--
-string(67) "The JSON document was too deep (too many nested objects and arrays)"
 string(56) "The JSON field referenced does not exist in this object."