php.net |  support |  documentation |  report a bug |  advanced search |  search howto |  statistics |  random bug |  login
Request #61147 No conflict resolution for properties in traits
Submitted: 2012-02-20 18:59 UTC Modified: 2012-02-26 15:32 UTC
Votes:3
Avg. Score:4.3 ± 0.9
Reproduced:2 of 2 (100.0%)
Same Version:1 (50.0%)
Same OS:0 (0.0%)
From: greywire at gmail dot com Assigned:
Status: Suspended Package: Class/Object related
PHP Version: 5.4.0RC8 OS: MacOS X 10.7
Private report: No CVE-ID: None
Have you experienced this issue?
Rate the importance of this bug to you:

 [2012-02-20 18:59 UTC] greywire at gmail dot com
Description:
------------
There is no way to resolve a conflict in two used traits with the same property.  The resolution method used to resolve conflicting method names should be able to be used similarly on properties.

I realize this may be the intended behavior to discourage use of properties in traits as properties were not originally intended in traits, but since properties are allowed, there should be a way to resolve the conflict, and the solution is simple and fits with existing syntax.



Test script:
---------------
<?php
trait testone {
    public $test;

}

trait testtwo {
    public $test;

}

//** this should work but doesn't
class traittest {
    use testtwo, testone {
        testone::$test insteadof testtwo;
        testtwo::$test as $testalso;
    }
}
?>

Expected result:
----------------
No errors or warnings.

Actual result:
--------------
Parse error: syntax error, unexpected '$test' (T_VARIABLE), expecting identifier (T_STRING) in index.php on line 14

Patches

Add a Patch

Pull Requests

Add a Pull Request

History

AllCommentsChangesGit/SVN commitsRelated reports
 [2012-02-21 02:53 UTC] aharvey@php.net
-Package: Compile Warning +Package: Class/Object related
 [2012-02-21 09:26 UTC] gron@php.net
-Status: Open +Status: Suspended
 [2012-02-21 09:26 UTC] gron@php.net
Unfortunately, the solution is not easy at all.

One reason is the dynamic nature of PHP. You can easily access properties based 
on their string name:

$foo = 'bar';
$this->$foo = 'foobar';

For exactly the same reason, we do NOT provide 'renaming'. The conflict 
resolution for methods allows to introduce an alias.
Nothing more. An alias is a new name pointing to an unchanged method-body.
(If you think about what that means, note that it breaks recursion in typical 
cases.)

Thus, your proposal does not work with the current semantics. It would only 
result in a new property in the class that is never actually used.

So far, we refrained from adding any of the constructs proposed in literature, 
since they would add considerable complexity.

From time to time there have been proposals for related features that could make 
it into a future version of PHP, but for the moment being, I do not see how we 
can make this work without changing the nature of PHP.

Suggestions are welcome of course.
Thanks
Stefan
 [2012-02-21 15:24 UTC] greywire at gmail dot com
I'm trying to understand here.  When you say alias, do you mean generically like "reference" or "pointer" to a method?  Or is there some aliasing construct I'm not aware of in the language?

Traits are supposed to be flattened in the class, so its more like a language assisted cut and paste, correct?  Which means two conflicting method names have to be renamed uniquely (or not named at all?) and then a reference/alias/pointer to the desired one is put in place with the original name.  I don't know how the traits are actually implemented internally...

Attempting to do this by hand I see that you could, using runkit, remove the two conflicting methods and put in one "aliased" method pointing to one that was removed.  But doing this with a property as you said results in a new property that does not affect the other two.  I thought maybe you could set that new property with a "&$obj->renamedprop" but that doesnt work.

Running through my options (thinking about if runkit could be used to fix it.  I've used runkit before to sorta implement my own traits functionality) I see its a tricky problem to solve indeed.

The only thing I came up with is that if at runtime PHP just removed the conflicting properties altogether so the code compiles and runs, and then the two properties (and the needed alias) could just be caught with __get/__set and handled by the user.  You'd get no warning or error about the conflict, but, you would immediately know about it when you accessed one of the properties.  This may seem like a hack but at least it would give people an option...

Is that a palatable compromise that wouldn't be hard to implement?  :)
 [2012-02-26 15:32 UTC] gron@php.net
When I write 'alias', I refer to the 'as' keyword.
https://wiki.php.net/rfc/horizontalreuse#conflict_resolution
Verbosely, the example in the RFC (`use A, B {B::bigTalk as talk;}`) can be read 
as: use the method body of B::bigTalk and refer to it as 'talk' in the composed 
class. Important is here, that bigTalk ends up still in the class and has not 
been changed in any way by the addition of a new alias. (In the RFC's example 
there is even a conflict that needs resolution.)

Classes just have hash-tables of methods, i.e., the method name is mapped to the 
method body.
Method bodies are not changed at all by the traits implementation, and all 
references to methods are basically by name, i.e., a string. Thus, a method call 
will look up the method body in the hash table based on a name/string.
The same is true for properties.

I do not see how your idea with __get/__set works. It is still the same name of 
the property that would be used. And it is not clear to me how they could be 
distinguished.
Something like that would require name-mangling like for private properties, in 
the trait's method.
But then again, you have quite a number of possible semantics of how to handle 
properties. Sometimes, they are supposed to access the same, sometimes they are 
supposed to be distinct.

That would require quite a bit of syntax/complexity. And, as I said, the 
academic literature on that topic brings way to much complexity for my taste.
There was a proposal on the list or on the wiki to introduce a 'local' keyword.
We might want to consider that for a future version.

The current solution, fail if it smells like something funny is going on, seems 
at least to me as a compromise that avoids nasty surprises when maintaining 
code, and allows us to weaken the restrictions later on.

Best regards
Stefan
 
PHP Copyright © 2001-2020 The PHP Group
All rights reserved.
Last updated: Sun Oct 25 17:01:24 2020 UTC