Bug #77100 [Opn]: Inconsistent behavior between constants and magic constants

Edit report at https://bugs.php.net/bug.php?id=77100&edit=1 ID: 77100 Updated by: nikic@php.net Reported by: fabien dot villepinte at gmail dot com Summary: Inconsistent behavior between constants and magic constants Status: Open Type: Bug Package: Scripting Engine problem PHP Version: Irrelevant Block user comment: N Private report: N New Comment: I'm inclined to mark this won't fix. Magic constants are just keywords that have syntactical similarity to constants. I don't think we have any business checking that constant names do not match keywords. If we forbid define('__DIR__'), should we also forbid define('function')? Ultimately everything here works just fine in the sense that you can define that constant and also look it up using constant(). You can't access it with constant syntax because the __DIR__ keyword takes precedence, but there's any number of other reasons why such an access may not be possible. For example, you can also do define('123', 'foo') and then access it using constant('123'), but for obvious reasons it will not change the behavior of directly writing 123 in code. Previous Comments: ------------------------------------------------------------------------ [2018-11-04 11:21:23] cmb@php.net > I'd suggest prepopulating the constants table with them only I > don't know how something like constant('__LINE__') could work. Prepopulating the constants table would require to update it frequently during run time to be consistent with the compile time constants. Besides being inefficient, it would also change “constant” values. It might be preferable to not allow any of the predefined magic constants (or maybe all constant names with two leading and trailing underscores) to be used with define(), defined() and constant(). Either way would break BC, though. ------------------------------------------------------------------------ [2018-11-03 21:27:48] a at b dot c dot de To continue that: Which means that when run time comes around the names are free to be redefined as user constants. But the literal identifiers aren't in the code any more so the only way they can be accessed is through the constant('') construct. The compile-time evaluation of magic constants also means define('__DIR__', 0); is legal, but const __DIR__ = 0; is not, since the latter definition happens at compile time too. Maybe a check (c|sh)ould be made that users aren't trying to define() magic constants themselves? I'd suggest prepopulating the constants table with them only I don't know how something like constant('__LINE__') could work. ------------------------------------------------------------------------ [2018-11-03 16:54:34] cmb@php.net FWIW, the docs[1] say: | All these "magical" constants are resolved at compile time, | unlike regular constants, which are resolved at runtime. [1] <http://php.net/manual/en/language.constants.predefined.php> ------------------------------------------------------------------------ [2018-11-03 16:12:34] fabien dot villepinte at gmail dot com Description: ------------ The functions define(), defined() and constant() don't work the same way whether the name of the constant given as an argument is the name of an already defined constant or the name of a magic constant. For instance : define('PHP_EOL', 0); // returns false and emits a notice define('__DIR__', 0); // returns true __DIR__; // returns the path of the current directory constant('__DIR__'); // returns the value of the user-defined constant, 0 in this case There is nothing in the documentation mentioning this case, so I can't say if the current behavior is correct or not. Test script: --------------- <?php var_dump(defined('__LINE__')); var_dump(define('__LINE__', 'test')); var_dump(constant('__LINE__')); var_dump(__LINE__); Actual result: -------------- bool(false) bool(true) string(4) "test" int(5) ------------------------------------------------------------------------ -- Edit this bug report at https://bugs.php.net/bug.php?id=77100&edit=1