MySQLi abstract layer dont works with hhvm, PHP without hhvm its works #4505
Comments
|
Line Also, proper formatting would be nice 😜 |
|
This code makes me cry. Every time you write a class misusing Anyway, you say HHVM crashes, can you provide a stack trace? Install hhvm-dbg or hhvm-nightly-dbg and make it crash once again. The stack trace should then be in /tmp. |
|
I've update de orginal issue, and the error is at line 49. with hhvm-dbg, the stack trace display this: Host: maquina ThreadType: Web Request 0 bt_handler at /tmp/tmp.JlCliioODY/hphp/runtime/base/crash-reporter.cpp:71 PHP Stacktrace: |
|
When I execute directly in a terminal in terminal
Display this: |
|
Part 1 The function definitions of bind-param() and bind-result() in http://php.net/manual/en/mysqli-stmt.bind-param.php and http://php.net/manual/en/mysqli-stmt.bind-result.php expect variable references (mixed &$var1). You are passing number 0, which is not a reference. So you can try: Both hhvm 3.3 and 3.4 work. Part 2 If I only correct $data but not $fields, for example However, hhvm shouldn't crash. Adding some checks about if the variable has the correct type in bytecode.cpp to stop executing the function from reflection or in ext_mysqli.cpp (line 129) can prevent from toInt64Helper() crashing and corruption. Part 3 If I change the lines My understanding is that bind_param_helper() in ext_mysqli.cpp stores all the arguments in a vector and calls arReturn() at the end. The reference variables are freed in arReturn() due to this commit, so update_result() in mysql_common.cpp can't set the result later ( Part 4: Memory leak Since that commit was about fixing memory leak, I reverted that commit and did a stress test. I found that the memory leaked so fast that somewhere else must be leaking. jemalloc profiling dumps showed MySQLStmt::bind_result() leaked. It turned out that ~MySQLStmt() and ~MySQLStmtVariables() were never called. The buffer will also leak if the result has strings. Every time hhvm runs your scripts above, it will be gradually leaking memory (about 1gb/5000 requests in my testing). |
…structs. Part 4 of facebook#4505.
|
@wjywbs: nice debugging, thanks! |
This mysqli abstract layer work very well with php-fpm and nignx. But when Im trying to implement it with HHVM + nginx there isn't way to got working correctly.
I think the problem is around the Reflection class to pass the arguments to be bind (bind params, bind results) as arrays.
connect_errno) { print("Página no disponible. Intente de nuevo más tarde."); exit(); } } protected static function prepare() { self::$stmt = self::$conn->prepare(self::$sql); self::$reflection = new ReflectionClass('mysqli_stmt'); } protected static function setParams() { $method = self::$reflection->getMethod('bind_param'); $method->invokeArgs(self::$stmt, self::$data); } protected static function getData($fields) { self::$results = array(); $method = self::$reflection->getMethod('bind_result'); $method->invokeArgs(self::$stmt, $fields); while(self::$stmt->fetch()) { self::$results[] = unserialize(serialize($fields)); } } protected static function close() { self::$stmt->close(); self::$conn->close(); } public static function execute($sql, $data, $fields=False) { self::$sql = strtolower($sql); self::$data = $data; self::open(); self::$conn->set_charset("utf8"); self::prepare(); self::setParams(); self::$stmt->execute(); if($fields) { self::getData($fields); return self::$results; } else { if(strpos(self::$sql, strtolower('INSERT')) === 0) { return self::$conn->insert_id; } } self::close(); } ``` } ?>example :
$sql = "SELECT user_id, name FROM user WHERE user_id > ?";
$data = array("i",0);
$fields = array('ID'=>'', 'NAME=>'');
$fields = DBModel::execute($sql, $data, $fields);
var_dump($fields);
The error log display:
\nWarning: Argument 1 not passed as reference
\nNotice: Array to string conversion in /var/www/core/DBModel.php on line 49
\nNotice: Array to string conversion in /var/www/core/DBModel.php on line 49
Core dumped: Aborted
Stack trace in /tmp/stacktrace.11929.log
And then HHVM crashes!
I'm not the author of this layer. more infor about http://pastebin.com/SHwDkQzL
The text was updated successfully, but these errors were encountered: