On further investigation, it appears that the error is meant to happen only if spl_autoload is called directly, and not via spl_autoload_call.  Unfortunately when spl_autoload is the only autoloader in the stack it gets called directly and spl_autoload_call doesn't get a look in.

You are calling class_exists() with just a class name, which leaves the second parameter ($autoload) set to true, which then invokes SPL and throws the exception, so no bug here

I beg to differ.  As you say, class_exists() attempts to autoload if there second param is true, but if autoloading fails it should simply return false as usual rather than throw an exception.  Otherwise it is rather useless.

The fact that this only occurs when there isn't another autoloader in the stack should make it clear that this is a bug.  For example, the following code does not throw an exception:

spl_autoload_register();
spl_autoload_register(function(){});
class_exists('foo\bar');

I 100% agree with dangerous dot ben. class_exists should return false if the class 
can't be autoloaded. Allowing it to do so would make it much easier to use an 
autoloader in contexts where you're interacting with an existing codebase that may 
not be so spl_autoload friendly. Bump.

Re-opening, as there still exists the conflict of class_exists() generating an 
error when autoloading fails, which is a chicken and the egg sort of issue. If 
one wants to try autoloading if the class doesn't exist, but autoloading fails, 
it should be possible to recover from that failure.

My understanding of the underlying code is that it generates an error in this 
case. Perhaps it should generate an exception, which can be caught an handled.

Although I have not yet migrated to 5.3, I care about this bug because I'm using 
an autoload function (in 5.2) which contains some directory traversal logic and 
uses class_exists() to decide whether or not to keep looking. I need to do this 
because I'm in the scenario described by james (and so cannot assume that all the 
package files in my codebase follow my preferred naming convention), and I would 
much prefer class_exists() to continue to return false rather than having it throw 
a new Exception.

PHP 5.3.8 (cli) (built: Aug 23 2011 11:50:20)
Copyright (c) 1997-2011 The PHP Group
Zend Engine v2.3.0, Copyright (c) 1998-2011 Zend Technologies
<?php
print_r(class_exists('asdfasdf'));//no error, just false
spl_autoload_register();
print_r(class_exists('asdfasdf'));//LogicException.
?>

<?php
class_exists('asdfasdf');//no error, just false.
spl_autoload_register('spl_autoload');
class_exists('asdfasdf');//no error, just false.
?>

spl_autoload is inconsistent with itself.

Even worse. The no arg register is inconsistent.
<?php
print_r(class_exists('asdfasdf'));//no error, just false.
spl_autoload_register();
print_r(class_exists('asdfasdf'));//LogicException
?>

<?php
print_r(class_exists('asdfasdf'));//no error, just false.
spl_autoload_register('spl_autoload');
spl_autoload_unregister('spl_autoload');
spl_autoload_register();
print_r(class_exists('asdfasdf'));//no error, just false.
?>
If the autoload stack is empty by the time the no arg registered is called, how can it function differently than if the autoload stack has been emptied by the time the no arg register is called?

This is nonsensical behavior, and manual for spl_autoload say nothing about it throwing. Let alone LogicException. I would have expected this to be a RuntimeException. As it depends on the specific environment it is run in as to whether or not spl_autoload will find something.

Is this bug still happening in PHP5.4 and greater? I'm using 5.3.8.

If one uses class_exists it should never throw error, only true or false. It makes 
no sense otherwise. This is one of those where setting of some other application 
can totally broke functionality elsewhere. Should everyone using class_exists 
catch try LogicException too? I'd argue not.

Though the workaround (empty function for spl) is kind of nice, as dangerous dot 
ben mentioned.

This is clearly a bug. We're testing if a class exists not asserting it.
There's seam to be a patch for this. Any chance it will be accepted?

Still broke in 5.4 and 5.5 3 years on!

v5.5.3 still broken

v5.5.9 - still broken.

> My understanding of the underlying code is that it generates an
> error in this case. Perhaps it should generate an exception,
> which can be caught an handled.

The code already throws an exception, see <https://3v4l.org/AaURG>.

It shouldn't throw an exception or cause an error. This is exactly why people move to C# because we don't have to put up with any of this bullshit for 5 years. An exception should be thrown when something happens that is not expected to happen, you expect class_exists to return false when the class does not exist, it is bonkers to think that we want to be wasting our lives catching stupid exceptions that should not be thrown in the first place, either that or catch the bloody exception in the body of the class_exists function for crying out loud. How hard is it really?

I echo the sentiments of matthew dot bonner if not his choice of words. If the purpose of class_exists() is to indicate whether or not a given class has been defined in the current scope or loaded into the current scope via include/require directives, then I simply cannot imagine a scenario in which it is acceptable for this function to throw an Exception.