Re: 5.4 regression: non-existent sub-sub keys now have values

Em Wed, 23 Nov 2011 14:19:42 -0000, Ferenc Kovacs <tyra3l@gmail.com> escreveu:

On Wed, Nov 23, 2011 at 3:14 PM, Daniel Convissor < danielc@analysisandsolutions.com> wrote: It's actually very simple. Take variable $a which is a string ("foo"). Now it you do $a[0] that would produce first letter - "f". Now here's a tricky part - if you do $a['blah'] it would convert 'blah' to number, get 0 and return the same letter "f".

To me, this is the bug. $a['blah'] does not exist. An undefined index notice should be raised. The key "blah" should not be converted to 0. The following two things should behave the same: $b = array('exists' => 'foo'); echo $b['blah'] . "\n"; $a = 'foo'; echo $a['blah'] . "\n"; But that second one echos out "f". This is a huge WTF.

and the following should also behave the same: $a = 'foo'; echo $a[2]; echo $a['2']; echo $a['2 cats']; because this is how the type juggling works in php: http://php.net/manual/en/language.types.type-juggling.php

But this is *not* (and afaik has never been) how indexes in arrays work. The conversion of array keys has always different rules: <?php $a['2 cats'] = null; var_dump($a); ^D array(1) { ["2 cats"]=> NULL } The mapping between strings and integers used for arrays is 1 to 1, and strings that can be mapped to integers are always mapped (so you cannot have a '1' (string) key and a 1 (int) key). These different rules are what makes possible to store any kind of string in an array, and changing this would be a huge error. I see, however, that it's not the case for indexing strings, where the lax rules are followed: <?php $a = 'foo'; var_dump($a['2 cats']); ^D string(1) "o" If there's any change to be made, it would be to change the way string indexing works, not the other way around. -- Gustavo Lopes