php.net |  support |  documentation |  report a bug |  advanced search |  search howto |  statistics |  random bug |  login
Bug #65317 spl autoload won't work inside an autoload
Submitted: 2013-07-23 16:26 UTC Modified: 2013-07-24 08:57 UTC
From: tyrael@php.net Assigned:
Status: Duplicate Package: Scripting Engine problem
PHP Version: 5.4.17 OS: linux
Private report: No CVE-ID: None
 [2013-07-23 16:26 UTC] tyrael@php.net
Description:
------------
I just bumped into an issue, where we had a Strict error triggered about an 
abstract class, which caused our autoloader to fail.
It seems that the autoloader won't be called when you try to reference a not-yet-
loaded class from an error handler triggered by an autoloader function.
I think that the attached test script makes the scenario more clear(sorry for the 
eval part, I didn't wanted to upload multiple scripts), but feel free to ask me if 
you need more info.

ps: it seems that this never worked http://3v4l.org/AQCue but I couldn't find any 
previous reports or notes/docs on the topic.

Test script:
---------------
<?php
spl_autoload_register(function($class){
        if ($class == 'MyConcrete') {
                eval('
                        class MyConcrete extends MyAbstract {
                                public static function createInstance() {}
                        }
                ');
        } elseif ($class == 'MyAbstract') {
                eval('
                        abstract class MyAbstract {
                                public abstract static function createInstance();
                        }
                ');
        } else {
                eval('class '.$class.'{}');
        }
});

set_error_handler(function($errno, $errstr, $errfile, $errline){
        $myclass = new MyClass;
});

$concrete = new MyConcrete;
$myclass = new MyClass;

Expected result:
----------------
trigger the autoloader from the error handler and load the class

Actual result:
--------------
Fatal error: Class 'MyClass' not found in index.php(14) : eval()'d code on line 3

Patches

Add a Patch

Pull Requests

Add a Pull Request

History

AllCommentsChangesGit/SVN commitsRelated reports
 [2013-07-23 16:51 UTC] laruence@php.net
Hey, it should be a intention behavior:


zend_class_entry *zend_fetch_class_by_name(const char *class_name, uint 
class_name_len, const zend_literal *key, int fetch_type TSRMLS_DC) /* {{{ */
{
    zend_class_entry **pce;
    int use_autoload = (fetch_type & ZEND_FETCH_CLASS_NO_AUTOLOAD) == 0;


it seems try to prevent some recursive autoload, then stack overflow 

so.... won't fix IMO (it's really a edge case :))

maybe document it instead?

thanks
 [2013-07-23 17:05 UTC] tyrael@php.net
it was introduced by Dmitry:
http://git.php.net/?p=php-
src.git;a=commitdiff;h=9305339d94c701e8eda93dbf0231d70918ef4a6a
"Changed "instanceof" and "catch" operators, is_a() and is_subclass_of() 
functions to not call __autoload()."

so nothing about generic recursion protection or explicitly talking about calls 
from another autloader method.
I would guess this specific scenario wasn't really intentional, but let's ask 
Dmitry about his thoughts.

Personally I think that it is rather surprising and unintuitive and it was a 
little bit tricky to figure it out why the code didn't worked as I've expected.
 [2013-07-23 17:09 UTC] laruence@php.net
my previous comment is not right.

it real reason is here, in (zend_lookup_class_ex):


    /* The compiler is not-reentrant. Make sure we __autoload() only during run-
time
     * (doesn't impact fuctionality of __autoload()
    */
    if (!use_autoload || zend_is_compiling(TSRMLS_C)) {
        if (!key) {
            free_alloca(lc_free, use_heap);
        }
        return FAILURE;
    }


so, the error is trigged in compiling MyAbstract... 

so is in compile time...

so,,
 [2013-07-23 17:13 UTC] tyrael@php.net
hm, that explains why nobody really reported it.
as far as I can remember most compile time errors are fatal which won't trigger 
the error handler anyways, and the others are E_STRICT, which wasn't really used 
before we included it to E_ALL in 5.4.
maybe we should reconsider adding new compile time non-fatal errors until we can 
properly support it?
 [2013-07-23 17:29 UTC] tyrael@php.net
Btw. isn't the fact that the error handler is called proves that the compiler is eiher reentrants or 
that the compilation is done?
 [2013-07-24 06:21 UTC] tyrael@php.net
-Package: SPL related +Package: Scripting Engine problem
 [2013-07-24 08:57 UTC] tyrael@php.net
-Status: Open +Status: Duplicate
 [2013-07-24 08:57 UTC] tyrael@php.net
I've created a new report with the updated info and testcase to avoid confusion, 
see https://bugs.php.net/bug.php?id=65322
 
PHP Copyright © 2001-2019 The PHP Group
All rights reserved.
Last updated: Thu Oct 17 02:01:27 2019 UTC