Understanding PHP Garbage Collection and Reference Counting

1. PHP Reference Counting

1.1 How Garbage Is Determined

Each object has a reference counter that increments when a new reference is created and decrements when a reference is destroyed. When the counter reaches zero, the value can be released and is not considered garbage; otherwise it may become garbage and is later examined by the collector.

If refcount becomes 0 after decrement, the value is released.

If refcount remains >0, the value is a potential garbage.

Potential garbage is accumulated until a threshold triggers the garbage identification program , which finally releases true garbage.

Drawback: Maintaining the counter adds overhead and struggles with cyclic references.

1.2 PHP Variable Internals

Every PHP variable is stored in a zval container, which holds the type, value, and two extra bytes: is_ref (indicates if the variable belongs to a reference set) and refcount (the number of symbols pointing to this zval).

1.3 Types Using Reference Counting

The following types are reference‑counted: string, array, object, resource, reference . Only variables whose type_flag is among the eight specified types and have IS_TYPE_REFCOUNTED=true use this mechanism.

2. Collection Principles

2.1 Collection Timing

Automatic collection: When a zval's refcount drops to 0, the value is immediately released.

Manual collection: Calling unset() (similar to Java's System.gc() ) triggers explicit cleanup.

2.2 Garbage Identification

The collector gathers possible garbage until a threshold is met, then runs a multi‑step identification process:

Mark values gray, decrement members' refcount to white.

If a value's refcount is 0, it is garbage (black); otherwise restore the count.

Remove non‑white nodes from the buffer.

Release the remaining garbage values.

3. Source Code Overview

3.1 Garbage Caretaker

The _zend_gc_globals structure acts as the garbage caretaker, storing possible garbage in a buffer ( buf ).

3.2 Initialization

During PHP startup, gc_init() parses php.ini , allocates the buffer, and calls gc_reset() to set up internal variables.

3.3 Determining Need for Collection

When a variable is destroyed, i_zval_ptr_dtor() checks the refcount . If it becomes 0, the value is released; if >0, the value is marked as possible garbage for later collection.

3.4 Collection Process

The gc_possible_root routine scans the buffer, extracts unused nodes, and when the buffer is full, triggers the identification and reclamation phases.

3.5 Releasing Garbage

The core function zend_gc_collect_cycles() performs the following steps:

Scan root nodes

Collect root nodes

Invoke the collector

Clean up variables

Finish collection

Example Code

$x = array();
$y = $x;
$z = $y;
unset($y);

4. Conclusion

PHP uses reference counting, unlike Java's reachability analysis.

PHP adds a custom algorithm to handle cyclic references by temporarily reducing reference counts and using color marking (white, black, purple) to identify true garbage.

The three colors represent: white – garbage, black – live data, purple – prevents duplicate insertion.

Is PHP the best language? The discussion is open.