Does this happen ONLY with PHP_5_3 branch? If not, please use the proper version string.

don't know, that's up to the developers to decide, don't you think?

This is not a bug. It's exactly what's supposed to happen. When you pass crap in you get crap out. The language isn't supposed to hold your hand in every corner.

<?php
var_dump(filter_var(null, FILTER_VALIDATE_INT));
var_dump(filter_var(array(), FILTER_VALIDATE_INT));
var_dump(filter_var('hi', FILTER_VALIDATE_INT));
var_dump(filter_var(1.1, FILTER_VALIDATE_INT));
var_dump(filter_var(fopen('somefile', 'r'), FILTER_VALIDATE_INT));
var_dump(filter_var(1, FILTER_VALIDATE_INT));
var_dump(filter_var('1.0', FILTER_VALIDATE_INT));
?>

all work without error.

<?php
class blah {function __toString(){return '1';}}
var_dump(filter_var(new blah, FILTER_VALIDATE_INT));
?>

works without error.

Only filter_var with an object that doesn't implement __toString fails with an error.  Either filter is incorrectly returning false with no error on all of those other random types, or you're wrong, Jani.

As StdClass has no __toString implementation (obviously), it can't be converted to string.

I tend to agree that it should not raise a fatal (recovable error) here as it defeats the main purpose of this API.

I'm however not sure it is worth the effort to add custom error handler for this edge case. Patch welcome, it could help to get this edge case solved (I can't work on this exact issue any time soon).

the stdClass was just to illustrate the point, this happens with any object that doesn't have a __toString().  I would imagine it would be easier to check for IS_OBJECT and then see if it implements __toString than to do a custom error handler.  The performance should be about the same (dismal).

it is indeed valid for all objects but it is still an edge case.

A check for IS_OBJECT + tostring() existence could be easy enough and with almost no impact(perf impact is null on the existing application). If you have some time to write one, go ahead.

http://pastie.org/586442 should do it. I did not not test more than the test case here.

I will close it as soon as it gets merged in 5.3.1, if the RM agrees.