rfc:case_insensitive_constant_deprecation

PHP RFC: Deprecate and Remove Case-Insensitive Constants

Introduction

PHP currently supports both case-sensitive and case-insensitive constants. Case-insensitive constants see very little practical use, are subject to various inconsistencies in functionality and cause undue implementational complexity. This RFC proposes to deprecate and remove case-insensitive constants.

The current state of the matter is:

  • Class constants are always case-sensitive.
  • Global constants declared with const are always case-sensitive. It should be noted that this applies only to the shortname of the constant, while namespaces in PHP are always case-insensitive.
  • Constants declared with define() are case-sensitive by default.
  • It is possible to declare case-insensitive constants by passing true as the third parameter of define().

This RFC proposes to:

  • In PHP 7.3: Deprecate calling define() with third parameter true.
  • In PHP 7.3: Deprecate accessing a case-insensitive constant with a casing that differs from the declaration-site. The constants true, false and null are exempt from this.
  • In PHP 8.0: Remove the possibility of declaring case-insensitive constants.
  • In PHP 8.0: true, false and null are converted from special-cased constants into reserved keywords.

Motivation

Symbols in PHP don't use consistent casing rules. Some symbols such as function names are case-insensitive, while others such as variable names are case-sensitive. The problem with constants is that they can be both. Apart from unnecessarily complicating the language, this also causes various issues outlined in the following.

Aliasing is case-sensitive

PHP supports the use const NS\FOO syntax for importing constants in namespaced code. Because aliasing (and namespaces in general) is purely compile-time functionality, this syntax cannot depend on whether the constant NS\FOO is case-sensitive or case-insensitive, as this is only decided at run-time. As such, the syntax has to choose one way or the other.

Because constants in PHP are predominantly case-sensitive, the use const syntax assumes case-sensitivity. Of course, this means that the functionality will not work correctly with case-insensitive constants:

namespace {
    define('NS\FOO', 42, true); // Case-insensitive constant
}
 
namespace Test {
    use const NS\FOO;
    var_dump(FOO); // Works
    var_dump(foo); // Warning: Use of undefined constant foo
}

This is a fundamental issue that cannot be resolved.

Constant redeclarations

Constants are supposed to be constant, i.e., their value may not change after initial declaration. For this reason PHP makes sure that constants are not redeclared:

define('FOO', 42);
var_dump(FOO); // int(42)
define('FOO', 24); // Notice: Constant FOO already defined
var_dump(FOO); // int(42)

These checks are not performed (or only in some very narrow cases) if case-sensitive and case-insensitive constants are mixed:

define('foo', 42, true);
var_dump(FOO); // int(42);
define('FOO', 24);
var_dump(FOO); // int(24)

Not only was the declaration of a clashing constant permitted, but it effectively changed the value of the FOO constant.

This problem is further confounded by assumptions (such as: constants are constant) in the PHP engine and opcache optimizations, resulting in additional issues like bug #74450, where the value of a constant changes retroactively.

This is an issue that can in principle be resolved, however it would come with significant additional implementation complexity and a hit to performance and memory usage. At the least, it would require storing lower-cased variants of all constants and checking against them on new constant declarations.

Implementation complexity and overhead

Support for case-insensitive constants makes the implementation more complex and slower. Constant lookups are implemented by first looking up the constant name directly, and then looking up a lowercased variant.

A particularly extreme case are access to unqualified constants inside namespaces. For example, if constant FOO is accessed inside namespace NS, the FETCH_CONST opcode is created with a record-breaking five literals. In order, ns\FOO, ns\foo, FOO and foo need to be looked up, and finally NS\FOO is used for error-reporting. For the common case where the intended constant was FOO this results in three lookups. Thankfully the impact is mitigated by runtime caching (which is actually incorrect due to the previous point).

Proposal

In PHP 7.3 both the declaration of case-insensitive constants, as well as their access (with a name different from the declared one) will result in a deprecation warning:

define('FOO', 42, true); // Deprecated: define(): Declaration of case-insensitive constants is deprecated
var_dump(FOO); // Ok!
var_dump(foo); // Deprecated: Case-insensitive constants are deprecated. The correct casing for this constant is "FOO"

The defined() function is not affected. It will continue to return true for case-insensitive constants, without generating a deprecation warning.

Declaration of case-insensitive constants by extensions will not generate a deprecation warning (though their access will). The reason behind this is that the end-user will not be able to do anything about this deprecation warning, while the extension maintainer may not be able to change the declaration for BC reasons at this point.

The reason why both declaration and access generate deprecation warnings is that both may generally fall into the responsibility of different maintainers. While accesses are performed by library users and can always be trivially fixed, definitions may be part of libraries that cannot immediately switch to case-sensitive constants due to backwards compatibility guarantees.

In PHP 8 the ability to declare case-insensitive constants will be removed.

Handling of true, false and null

true, false and null in PHP are originally “ordinary” case-insensitive constants, though in practice they are subject to various special casing. For example, these constants are not subject to namespace fallback, as we must be able to resolve their values at compile-time.

true, false and null are the only case-insensitive constants which are commonly used with a casing different from their declaration (which is TRUE, FALSE and NULL). In the mind of programmers not familiar with the details of the PHP implementation, these resemble keywords more than constants.

This RFC proposes to convert true, false and null into proper reserved keywords in PHP 8 (reserved keywords are always case-insensitive). This has two implications with regard to backwards compatibility:

  • As true, false and null are no longer constants, they will not be accessible through constant(“true”), defined(“true”) etc. The names can also no longer be treated as namespaced names, such that \true and namespace\true would become invalid.
  • true, false and null can no longer be used as identifiers. They will remain legal method and class constant names, as these are not subject to reserved keyword restrictions. It should be noted that these symbols are already forbidden as class names and as (namespaced) global constant names, so effectively this means that only the use as a function name is additionally forbidden.

Backward Incompatible Changes

Additional deprecation warnings are thrown in PHP 7.3. Case-insensitive constants are removed in PHP 8.

true, false and null become reserved keywords in PHP 8. See the end of the previous section for the BC implications this has.

Unaffected PHP Functionality

Magic constants are not affected. These are already reserved keywords (always case-insensitive), not accessible via constant(), etc.

Class constants are not affected, they are already case-sensitive.

Vote

Since this is a language change, a 2/3 majority is required. The vote ends 2018-07-16.

Deprecate (and later remove) case-insensitive constants?
Real name Yes No
ab (ab)  
ajf (ajf)  
ashnazg (ashnazg)  
bwoebi (bwoebi)  
carusogabriel (carusogabriel)  
cmb (cmb)  
colinodell (colinodell)  
danack (danack)  
derick (derick)  
dmitry (dmitry)  
galvao (galvao)  
gasolwu (gasolwu)  
hywan (hywan)  
jhdxr (jhdxr)  
jwage (jwage)  
kalle (kalle)  
kelunik (kelunik)  
kguest (kguest)  
lbarnaud (lbarnaud)  
levim (levim)  
lex (lex)  
marcio (marcio)  
mcmic (mcmic)  
narf (narf)  
nikic (nikic)  
ocramius (ocramius)  
peehaa (peehaa)  
pmjones (pmjones)  
pmmaga (pmmaga)  
pollita (pollita)  
rtheunissen (rtheunissen)  
sobak (sobak)  
stas (stas)  
svpernova09 (svpernova09)  
trowski (trowski)  
weierophinney (weierophinney)  
yunosh (yunosh)  
zeev (zeev)  
zimt (zimt)  
Final result: 39 0
This poll has been closed.
rfc/case_insensitive_constant_deprecation.txt · Last modified: 2018/07/16 17:18 by nikic