[PHP 8.3] constants have their namespace lowercased

[ju1ius]

Description

The following code:

const char* name = "Foo\\Bar\\BAZ";
zend_register_long_constant(name, strlen(name), 42, CONST_CS|CONST_PERSISTENT, module_number);

$re = new \ReflectionExtension('my_extension');
var_dump($re->getConstants());

Resulted in this output:

array(1) {
  ["foo\bar\BAZ"]=>
  int(42)
}

But I expected this output instead:

array(1) {
  ["Foo\Bar\BAZ"]=>
  int(42)
}

This also affects user-defined constants, see #11423 (comment).

This behaviour was introduced in php 8.3 by the removal of the name field of the zend_constant struct in f42992f (#10954).

PHP Version

PHP 8.3-dev

Operating System

irrelevant

[kocsismate]

Uhh, thanks for raising this issue! I'll definitely look at it ASAP, possibly tomorrow.

[kocsismate]

The behaviour which causes the issue was introduced in e2ad4e01780bf1e6cba162750946b9adfc40526b for fixing a bug. I missed to remove it since the bug is not possible anymore (as constants are always case sensitive now).

UPDATE: case-insensitive constants are irrelevant here, since this bug is about the namespace.

[kocsismate]

I added the invalid tag for now because surprisingly, the new behavior is still consistent with userland constants defined via define().

If you have a look at this test, you'll notice that the namespace of userland constants defined by define() is also converted to lowercase characters.

So I'm not sure if we have to/can follow any strict rule regarding how the namespace of constants should be stored.

UPDATE: I'm a bit tired I guess... Of course, this behavior is the result of my changes in PHP 8.3.. And previously, the constant names were displayed with the correct casing.

So do you guys think that my changes should be reverted? Or do you see any reasonable fix for the problem? @Girgias @iluuu1994

[iluuu1994]

This is not limited to internal constants.

--TEST--
get_defined_constants() namespace casing
--FILE--
<?php
namespace Foo {
    const BAR = 'BAR';
}
namespace {
    foreach (get_defined_constants() as $name => $value) {
        if ($value === 'BAR') {
            var_dump($name);
        }
    }
}
?>
--EXPECT--
string(7) "foo\BAR"

Tbh I don't believe shrinking the size of constants is particularly important. We're saving 8 bytes per constant. It doesn't matter too much for space or cache locality. We do try to preserve casing in most cases, including error messages. There is already an edge case where the name casing is not preserved: php/doc-en#2383.

Casing consistency is not a huge deal to me, but I'm not sure if we should intentionally break unless there is a clear win.