Posted by Stephen Leaf on 09/16/05 19:19

Can't expect everything to be done right on the first try. But it sure was a
big annoyance. Great to hear that work is being done to improve on this.
I personally would rather it just be silently ignored as to me an error line
at the top is the same as it not working at all.
Would love to find a tool that could go over code and output errors like this
tho. Provide a better solution or whatever along with all the dumb mistakes
that PHP is ignoring for you.

With my experiences I have never used pointers or references in PHP. I'm sure
they are powerful and all. but I just don't want to mess with them or have
worry about them. That's just 1 thing in programming I don't have much
experience with. I give a function a variable and I think I'm passing the
value. if the function takes the reference and modifies that.. oh well.. I
got my result back that's all I care about :)

as for the sort thought. If the programmer does that and can't figure out
what's going on.. they obviously didn't read the manual or just don't know
basic programming ;)

bool sort ( array &array [, int sort_flags] )

Returns bool.. that's not an array ;) If it takes an array but doesn't return
an array.. it must operate on the passed in array if it didn't what a useless
function. Lets sort the array and say yah we sorted it but you can't have
it :)

So yes a bug. But for those that want to be able to grab only what we need. in
my case array_pop's returned element. I don't wanna be hassled with the "are
you sure you wanted to ignore part of what we did?"
It's almost like every program asking "are you sure you wanted to close me?"
everytime I get asked that I always think... "I did just click the X.. so....
Yah......."

Just my 2 cents. I'm sure you guys will come up with something.. either way
PHP is the only language I'll use for a website. Thank you very much for the
detailed reason behind this.

On Friday 16 September 2005 09:28 am, Rasmus Lerdorf wrote:
> Stephen Leaf wrote:
> > $this->urlArr[0] = array_pop($arr = explode("&",$this->urlArr[0]));
> >
> > I still have to scratch my head as to why I *need* that "$arr = "
> > prior to 5.0.5 this was not needed.
> > $this->urlArr[0] = array_pop(explode("&",$this->urlArr[0]));
>
> This is a much misunderstood issue. And we are still contemplating the
> best way to handle this. Let's take a step back to a really simple
> example:
>
> function foo() {
> return 3;
> }
> function bar(&$arg) {
> $arg = "banana";
> }
> bar(foo());
>
> What do we do with code like this? If you follow it through it is
> essentially doing:
>
> 3 = "banana";
>
> which makes very little sense and is probably something the developer
> would want to know about. PHP 4.3.x happily let you do this, and in
> some circumstances this could even corrupt memory. PHP 4.4 and higher
> have fixed this memory corruption problem, and at the same time since we
> are now able to detect this we can throw an error to let you know that
> your code is probably not doing what you intended.
>
> Now, in your example you are doing:
>
> $a = array_pop(explode("&","a&b"));
>
> to get the last element from the explode. The issue with array_pop()
> and other similar functions is that they do 2 things. They modify the
> array (by removing the last element) and they return something (the
> removed element in this case). When you pass in the result of explode()
> you are passing in something that doesn't have any permanent storage
> associated with it. Internally in PHP this is known as a temp var. You
> can't make a reference to a temp var. In our simpler example it is like
> trying to do &3 which doesn't make any sense. So in this case
> array_pop() has no place to store the modification. Your code happens
> to not care about that and only cares about the returned value.
>
> The real question here is whether we should silently ignore cases where
> such an argument modification is thrown away. In the case of
> array_pop() it may make sense. But how about this:
>
> sort(explode("&","a&b"));
>
> sort() has just one purpose. That is to sort the passed array in place.
> The above line of code is a really slow line of code that does
> absolutely nothing because we are passing in something that goes out of
> scope as soon as the call is done. This line of code in a PHP program
> is an obvious bug. Now, if we change this to:
>
> sort($a=explode("&","a&b"));
>
> then it suddenly makes sense. This is equivalent to writing:
>
> $a = explode("&","a&b"));
> sort($a);
>
> and the result is that we have a sorted array in $a after this.
>
> So the idea here is that PHP should be able to detect these sorts of
> obvious errors. In the array_pop() case it is not as black and white as
> the sort() case because of the dual-purpose nature of array_pop(). You
> could make a case for silently ignoring the array modification in
> array_pop().
>
> My personal feeling on this is that we should throw an E_NOTICE on these
> and let the code continue on. Often there is a more efficient way to do
> some of these things simply because you are calling something that does
> more than you need and ignoring some of what the function does and other
> times, like with the sort() example, this is actually a real bug in your
> code.
>
> If it wasn't for the memory corruption associated with this, we wouldn't
> have touched it in PHP 4 at all. But since it caused weird and
> wonderful crashes in complex PHP 4 code we didn't really have the option
> to ignore it. So what we do now is to demote impossible references to
> non-references and throw an E_NOTICE. In PHP 5 this ended up being a
> fatal error which is something we have been discussing over the past
> week. In its current state having this as a fatal error is incorrect
> and something we will fix. We need to either make it an E_NOTICE or
> there has been talk about coming up with a smarter way of handling these
> impossible reference cases, but it is pretty tricky.
>
> We don't always get things right on the first try. Expect to see some
> tweaks to the current implementation to make it more compatible with
> existing code over the next couple of weeks.
>
> -Rasmus

[Back to original message]