====== PHP RFC: Allow Object Property Writes on Objects Referenced by Constants ======
* Version: 0.4
* Date: 2026-06-28
* Author: Khaled Alam, khaledalam.net@gmail.com
* Status: Implemented in PHP 8.6
* Implementation: https://github.com/php/php-src/pull/20903
===== Introduction =====
This RFC allows writing to **object properties** (and isset() / unset() on properties) when the object instance is referenced by a constant.
PHP constants are immutable in the sense that the constant binding cannot be changed. However, objects in PHP are mutable by design: mutating an object property does not rebind the constant, it only updates the state of the referenced object.
Today, PHP rejects CONST_OBJ->prop = ... with a fatal error ("Cannot use temporary expression in write context").
This RFC proposes permitting property write operations in this specific scenario, aligning behavior with user expectations while keeping the constant binding immutable.
value = 123;
var_dump(OBJ->value); // int(123)
unset(OBJ->value);
var_dump(isset(OBJ->value)); // bool(false)
?>
This applies both to **global constants** (const / define()) and to **class constants** (Cls::CONST).
===== Proposal =====
This RFC proposes to allow property write operations where the base expression is a constant that evaluates to an object.
The change is intentionally narrow:
* It permits **object property** writes (->) and related operations (isset, unset, compound assignment, increment/decrement, string concatenation) on objects referenced by constants.
* It permits **passing constant object properties by reference** to functions (including the "BP_VAR_FUNC_ARG" path when the function is declared after the call site).
* It does **not** make constants mutable (rebinding remains illegal).
* It does **not** change array/dimension write semantics ([]), nor does it change "write-to-temporary" behavior for dims.
* It applies to both **global constants** (CONST->prop = ...) and **class constants** (Cls::CONST->prop = ... and $obj::CONST->prop = ...).
Specifically, the following operations become valid:
* Assignment to a property:
prop = "value";
?>
prop = 42; // via class name
var_dump(C::O->prop); // int(42)
$c = new C();
$c::O->prop = 'updated'; // via instance
var_dump(C::O->prop); // string(7) "updated"
?>
* Compound operations on a property:
counter = 0;
OBJ->counter++;
OBJ->counter--;
OBJ->counter += 10;
OBJ->str = 'hello';
OBJ->str .= ' world';
?>
* isset() / unset() on a property:
prop = 1;
var_dump(isset(OBJ->prop)); // true
unset(OBJ->prop);
var_dump(isset(OBJ->prop)); // false
?>
* Nested property operations:
a = new stdClass();
OBJ->a->b = 7;
var_dump(OBJ->a->b); // int(7)
?>
* Passing properties by reference:
val = 10;
function modify(&$v) { $v = 42; }
modify(OBJ->val);
var_dump(OBJ->val); // int(42)
?>
==== Semantics ====
* The constant binding remains immutable.
* The referenced object may be mutated, consistent with PHP's object behavior.
* If the constant does not evaluate to an object at runtime, existing behavior (type error/fatal in property access) applies as usual.
* Class constants holding objects behave identically to global constants; only the object state can change, never the constant binding.
==== Note on Array Constants ====
A related but distinct scenario exists with array constants:
Because arrays use copy-on-write semantics, modifying the copy has no effect on the original constant. Similarly:
These dim-write cases involve assignment to temporaries that silently have no effect. Whether such no-effect writes should produce a warning or error is a valid concern but is **out of scope** for this RFC and may be addressed separately (e.g. warning/throwing on "ASSIGN_DIM" when OP1 is not a pointer — i.e. not an indirect, reference, or object — as the operation can never have an effect). See the Future Scope section.
==== Non-goals / Explicitly Unchanged ====
This RFC does NOT change:
* Rebinding constants:
* Dimension/array writes or other write-to-temporary semantics:
==== Implementation Details ====
The compiler marks the constant AST node rather than emitting the fetch directly. zend_delayed_compile_prop() in Zend/zend_compile.c calls zend_mark_const_object_fetch(), which walks down through any ZEND_AST_DIM nodes and sets a ZEND_CONST_OBJECT_FETCH attribute on a ZEND_AST_CONST or ZEND_AST_CLASS_CONST base. zend_delayed_compile_var() then handles those two AST kinds when the attribute is set, so the constant is fetched at runtime and the property write targets the real object.
zend_emit_fetch_constant() and zend_compile_class_const() take the fetch type: in read context (BP_VAR_R, BP_VAR_IS) the result stays IS_TMP_VAR as before, and in write context it is IS_VAR. Because a constant may still be substituted at compile time (true, null, persistent constants, same-file class constants), a compile-time-evaluated constant in write context is wrapped in a ZEND_QM_ASSIGN to materialise it into an IS_VAR; the property opcodes have no handler for an IS_CONST container operand.
Two supporting changes were needed. zend_optimizer_update_op1_const() now refuses to substitute a constant into op1 of the object-write opcodes (ZEND_ASSIGN_OBJ, ZEND_FETCH_OBJ_W, ZEND_UNSET_OBJ, ZEND_PRE_INC_OBJ and friends), which would otherwise reintroduce the invalid operand during optimisation. And zend_can_write_to_variable() accepts a constant at the base of a PROP/DIM chain when the chain contains at least one property fetch, so that destructuring assignment and by-reference foreach can target a constant object property; a write to the constant itself, or to a dimension of it, still targets a temporary and remains illegal.
==== Why This Brings Value ====
This feature addresses a common surprise: PHP already permits objects in constants (since PHP 8.1), and developers naturally expect to mutate the object's state through the constant reference. The current fatal error is not about constant immutability per se, but rather about an overly broad "temporary write context" restriction in the compiler. Narrowly enabling property writes reduces friction, improves ergonomics, and matches the runtime model of objects without increasing language complexity for other write contexts.
===== Backward Incompatible Changes =====
Enum cases are class constants, so this change affects them. Writing to or unsetting a readonly case property — for example unset(Enum::Case->value) or Enum::Case->value = $x — now raises the standard runtime readonly-property "Error" ("Cannot unset/modify readonly property") instead of the previous compile-time fatal ("Cannot use temporary expression in write context").
This makes enum-case property access consistent with plain variable access ($x = Enum::Case;
unset($x->value);). Enum case properties remain immutable — no value that was previously read-only becomes writable, and dynamic properties on enum cases continue to be rejected.
Apart from this, code that previously terminated with a fatal error for CONST_OBJ->prop = ... will now execute successfully. This is a compatibility improvement — no previously-working code changes behavior.
===== Proposed PHP Version(s) =====
PHP 8.6
===== RFC Impact =====
==== To the Ecosystem ====
* IDEs / LSPs: Mostly positive; previously-invalid syntax becomes valid.
* Static analyzers: May need minor updates to remove a false-positive for property writes via constants. Semantics remain straightforward: constant binding is unchanged, object state may change.
* Formatters / linters: No expected changes beyond accepting the syntax.
==== To Existing Extensions ====
No expected impact. This is a core compilation change around property write access, not an extension API change.
==== To SAPIs ====
No expected impact.
===== Open Issues =====
None currently.
===== Future Scope =====
* **No-effect dim writes**: A future RFC could explore whether certain no-effect write cases should produce a warning or error. For example, "ASSIGN_DIM" when OP1 is not a "pointer" (indirect, reference, or object) can never have an effect, since the write is to a temporary that is immediately discarded. This is a distinct concern from the property-write issue addressed here.
* **Other write-to-temporary semantics**: Cases like foo()[0] = 1 where the return value is a temporary are similarly out of scope but could be revisited.
===== Voting Choices =====
Primary Vote requiring a 2/3 majority to accept the RFC:
* Yes
* No
* Abstain
===== Patches and Tests =====
Proof of concept implementation:
* https://github.com/php/php-src/pull/20903
===== Implementation =====
- Merged into: PHP 8.6
- Git commit: https://github.com/php/php-src/commit/b16cab7da8e5748b173542da42fc8b602ae59dea
- PR: https://github.com/php/php-src/pull/20903
===== References =====
* GH-10497: https://github.com/php/php-src/issues/10497
* PR #20903: https://github.com/php/php-src/pull/20903
* Internals discussion thread: https://news-web.php.net/php.internals/129157
* Internals thread (language change notes, incl. class-constant feedback): https://news-web.php.net/php.internals/130606
===== Rejected Features =====
* Extending the change to dimension/array writes (out of scope for this RFC; involves distinct copy-on-write and temporary semantics)
* Changing "write-to-temporary" warning/error policy for dims (out of scope; may be addressed in a future RFC)
===== Changelog =====
* 2026-02-24: Initial draft created.
* 2026-04-04: v0.2 — Updated after PR review feedback. Added by-ref passing examples, implementation details section, note on array constant behavior per internals feedback, explicit test inventory, updated voting timeline.
* 2026-06-28: v0.3 — Extended scope to class constants (Cls::CONST->prop and $obj::CONST->prop) per Claude Pache's feedback. Documented the enum readonly-property backward-incompatible change. Restored the Abstain vote option (added by Tim Düsterhus). Restarted the discussion period for the widened scope.
* 2026-08-13 (v0.4): Updated Implementation Details to match the merged implementation. Marked implemented in PHP 8.6.
Key changes from v0.1:
- Added by-ref passing examples (confirmed working in tests)
- Added implementation details section explaining the compiler change
- Added "Note on Array Constants" section addressing Ilija's and Tim's feedback about silent no-op dim writes
- Expanded Future Scope with Tim's suggestion about warning on ASSIGN_DIM with non-pointer OP1
- Voting options were revised during later RFC updates; the current vote includes Yes, No, and Abstain
- Removed the placeholder voting close date until voting opens
- Targeted PHP 8.6 explicitly
- Listed all test files in the Patches and Tests section
- Removed Open Issues (the by-ref question is answered, isset/unset is confirmed)
- Added full URLs to References
- Added changelog entry for v0.2