rfc:container-offset-behaviour
Differences
This shows you the differences between two versions of the page.
| Next revision | Previous revision | ||
| rfc:container-offset-behaviour [2024/07/04 13:49] – created girgias | rfc:container-offset-behaviour [2025/04/03 13:08] (current) – external edit 127.0.0.1 | ||
|---|---|---|---|
| Line 1: | Line 1: | ||
| ====== PHP RFC: Improve language coherence for the behaviour of offsets and containers ====== | ====== PHP RFC: Improve language coherence for the behaviour of offsets and containers ====== | ||
| - | * Version: 0.1 | + | * Version: 0.3 |
| * Date: 2023-12-07 | * Date: 2023-12-07 | ||
| * Author: Gina Peter Banyard < | * Author: Gina Peter Banyard < | ||
| * Status: Under Discussion | * Status: Under Discussion | ||
| - | * Target Version: PHP 8.4 | + | * Target Version: PHP 8.5 |
| * Implementations: | * Implementations: | ||
| - | | + | |
| - | * < | + | * < |
| - | * Original string offset clean-up PR: https:// | + | * Original string offset clean-up PR: <https:// |
| - | * First Published at: http:// | + | * First Published at: <http:// |
| - | * Markdown source: https:// | + | * Markdown source: |
| ===== Introduction ===== | ===== Introduction ===== | ||
| Line 32: | Line 32: | ||
| ==== Operations ==== | ==== Operations ==== | ||
| - | We consider there to be nine (9) different operations that relate to containers and offsets, | + | We consider there to be ten (10) different operations that relate to containers and offsets, |
| which are the following: | which are the following: | ||
| Line 44: | Line 44: | ||
| * Fetch | * Fetch | ||
| * Fetch-Append | * Fetch-Append | ||
| + | * Incrementing/ | ||
| The read, write, read-write, appending, and unsetting operations are what one would expect. | The read, write, read-write, appending, and unsetting operations are what one would expect. | ||
| We split the existence check operation into two distinct sub-operations as the behaviour between | We split the existence check operation into two distinct sub-operations as the behaviour between | ||
| - | <php>iseet()</ | + | <php>isset()</ |
| A fetch operation occurs when reference to the offset must be acquired, | A fetch operation occurs when reference to the offset must be acquired, | ||
| Line 60: | Line 61: | ||
| In general, a nested operation will perform all the necessary fetch/read operations, | In general, a nested operation will perform all the necessary fetch/read operations, | ||
| interpreting the returned value as a container, until it reaches the final dimension. | interpreting the returned value as a container, until it reaches the final dimension. | ||
| + | |||
| + | Finally, we mention incrementing/ | ||
| + | but as it works on a " | ||
| ==== Container types ==== | ==== Container types ==== | ||
| Line 68: | Line 72: | ||
| * < | * < | ||
| * < | * < | ||
| - | * < | ||
| * < | * < | ||
| * < | * < | ||
| Line 118: | Line 121: | ||
| as the engine treats those container types identically. | as the engine treats those container types identically. | ||
| - | * For read operations, < | + | * For read operations, < |
| - | Warning: Trying to access array offset on TYPE | + | |
| - | </ | + | * For write, read-write, appending, fetch, fetch-append, and increment/ |
| - | * For write, read-write, appending, fetch, | + | |
| - | Cannot use a scalar value as an array | + | * For the unset operation, the following error is thrown:< |
| - | </ | + | |
| - | * For the unset operation, the following error is thrown: < | + | |
| - | Cannot unset offset in a non-array variable | + | |
| - | </ | + | |
| * For existence operations, no warning is emitted and the behaviour is as if the offset did not exist. | * For existence operations, no warning is emitted and the behaviour is as if the offset did not exist. | ||
| Line 143: | Line 143: | ||
| Therefore, the behaviour depending on the operator is as follows: | Therefore, the behaviour depending on the operator is as follows: | ||
| - | * For read operations,< | + | * For read operations, < |
| - | Warning: Trying to access array offset on null | + | |
| - | </ | + | |
| * For write, append, fetch, and fetch-append operations the container is converted to array. And thus behave like an < | * For write, append, fetch, and fetch-append operations the container is converted to array. And thus behave like an < | ||
| - | | + | |
| + | | ||
| * For the unset operation, the container continues to be < | * For the unset operation, the container continues to be < | ||
| + | |||
| * For existence operations, no warning is emitted and the behaviour is as if the offset did not exist. | * For existence operations, no warning is emitted and the behaviour is as if the offset did not exist. | ||
| Line 158: | Line 160: | ||
| Therefore, the behaviour depending on the operator is as follows: | Therefore, the behaviour depending on the operator is as follows: | ||
| - | * For read operations,< | + | * For read operations, < |
| - | Warning: Trying to access array offset on false | + | |
| - | </ | + | |
| - | * For write, append, fetch, and fetch-append operations the container is converted to array, Emitting the following deprecation notice: < | + | * For write, append, fetch, and fetch-append operations the container is converted to array, Emitting the following deprecation notice:< |
| - | Deprecated: Automatic conversion of false to array is deprecated | + | |
| - | </ | + | |
| - | * For read-write operations, the container is converted to array, before the read operation, Emitting the following deprecation notice: < | + | * For read-write |
| - | Deprecated: Automatic conversion of false to array is deprecated | + | |
| - | </ | + | |
| - | * For the unset operation, the container continues to be < | + | * For the unset operation, the container continues to be < |
| - | Deprecated: Automatic conversion of false to array is deprecated | + | |
| - | </ | + | |
| * For existence operations, no warning is emitted and the behaviour is as if the offset did not exist. | * For existence operations, no warning is emitted and the behaviour is as if the offset did not exist. | ||
| Line 247: | Line 242: | ||
| Moreover, some operations are invalid on string offsets: | Moreover, some operations are invalid on string offsets: | ||
| - | * Read-Write operations on a string offset will throw the following error: < | + | * Read-Write operations on a string offset will throw the following error: < |
| - | Cannot use assign-op operators with string offsets | + | |
| - | </ | + | * Unset operations on a string offset will throw the following error: < |
| - | * Unset operations on a string offset will throw the following error: < | + | |
| - | Cannot unset string offsets | + | * The append and fetch-append operations will throw the following error: < |
| - | </ | + | |
| - | * The append and fetch-append operations will throw the following error: < | + | |
| - | [] operator not supported for strings | + | |
| - | </ | + | |
| * Fetch operations will throw different errors depending on the fetch operation, //after// the type of the offset has been checked: | * Fetch operations will throw different errors depending on the fetch operation, //after// the type of the offset has been checked: | ||
| - | | + | |
| - | Cannot create references to/from string offsets | + | * For attempting to use the string offset as a container: < |
| - | </ | + | * For attempting to use the string offset as an object: < |
| - | * For attempting to use the string offset as a container: < | + | * For attempting to use increment or decrement the string offset: < |
| - | Cannot use string offset as an array | + | |
| - | </ | + | |
| - | * For attempting to use the string offset as an object: < | + | |
| - | Cannot use string offset as an object | + | |
| - | </ | + | |
| - | * For attempting to use increment or decrement the string offset: < | + | |
| - | Cannot increment/ | + | |
| - | </ | + | |
| Attempting to read a non initialized string offset emits the following warning: | Attempting to read a non initialized string offset emits the following warning: | ||
| Line 376: | Line 360: | ||
| </ | </ | ||
| results in the following output: | results in the following output: | ||
| - | < | + | < |
| bool(false) | bool(false) | ||
| bool(true) | bool(true) | ||
| Line 434: | Line 418: | ||
| for determining if the value is falsy or not. | for determining if the value is falsy or not. | ||
| This is error-prone, | This is error-prone, | ||
| - | properly. | + | properly. |
| One other requirement of the < | One other requirement of the < | ||
| Line 477: | Line 461: | ||
| The complexity of these requirements for the < | The complexity of these requirements for the < | ||
| and was the source of a bug in < | and was the source of a bug in < | ||
| - | [1:https:// | + | ((https:// |
| The only extension that properly implements all this complexity is SimpleXML | The only extension that properly implements all this complexity is SimpleXML | ||
| Line 496: | Line 480: | ||
| One additional pitfall that is common to all dimension handlers is the need to call < | One additional pitfall that is common to all dimension handlers is the need to call < | ||
| on the offset < | on the offset < | ||
| - | This requirement wasn't followed by < | + | This requirement wasn't followed by < |
| - | < | + | < |
| - | and < | + | and < |
| Moreover, some extensions do dereference the offset, but only indirectly, and it is not know if | Moreover, some extensions do dereference the offset, but only indirectly, and it is not know if | ||
| this was done on purpose or happens to work, for example < | this was done on purpose or happens to work, for example < | ||
| Line 513: | Line 497: | ||
| * Read: the < | * Read: the < | ||
| + | |||
| * Write: the < | * Write: the < | ||
| + | |||
| * Read-Write: the < | * Read-Write: the < | ||
| + | |||
| * Appending: the < | * Appending: the < | ||
| + | |||
| * Unsetting: the < | * Unsetting: the < | ||
| + | |||
| * Existence checks via isset(): the < | * Existence checks via isset(): the < | ||
| + | |||
| * Existence checks via empty(): the < | * Existence checks via empty(): the < | ||
| + | |||
| * Existence checks via the null coalesce operator < | * Existence checks via the null coalesce operator < | ||
| + | |||
| * Fetch: the < | * Fetch: the < | ||
| + | |||
| * Fetch Append: the < | * Fetch Append: the < | ||
| + | |||
| + | * Increment/ | ||
| Because < | Because < | ||
| the following notice is emitted: | the following notice is emitted: | ||
| - | < | + | < |
| Notice: Indirect modification of overloaded element of ClassName has no effect in %s on line %d | Notice: Indirect modification of overloaded element of ClassName has no effect in %s on line %d | ||
| </ | </ | ||
| Line 615: | Line 610: | ||
| $o = new T(); | $o = new T(); | ||
| - | $a = new ArrayObject(); | ||
| - | |||
| $a = new ArrayObject($o); | $a = new ArrayObject($o); | ||
| var_dump(isset($a[' | var_dump(isset($a[' | ||
| Line 623: | Line 616: | ||
| results in the following behaviour: | results in the following behaviour: | ||
| - | < | + | < |
| bool(true) | bool(true) | ||
| Line 687: | Line 680: | ||
| * Call method to verify the offset exists: | * Call method to verify the offset exists: | ||
| - | | + | |
| - | * Otherwise: call method to get value of offset: | + | * Otherwise: call method to get value of offset: |
| - If the value is < | - If the value is < | ||
| - Otherwise: return < | - Otherwise: return < | ||
| Line 737: | Line 730: | ||
| and if it doesn' | and if it doesn' | ||
| These semantics also preventing the widening of the < | These semantics also preventing the widening of the < | ||
| - | accept objects that support accessing offsets, something that has been requested by userland. | + | accept objects that support accessing offsets, something that has been requested by userland. |
| Needing to handle < | Needing to handle < | ||
| Line 809: | Line 802: | ||
| Cross-version compatible code can use DNF types to type their input arguments, e.g: | Cross-version compatible code can use DNF types to type their input arguments, e.g: | ||
| <PHP> | <PHP> | ||
| - | function foo(ArrayAccess|(DimensionReadable& | + | function foo(ArrayAccess|(DimensionFetchable& |
| - | /// Do something useful | + | /* Do something useful |
| } | } | ||
| </ | </ | ||
| Line 821: | Line 814: | ||
| Therefore, we move the handlers out of the < | Therefore, we move the handlers out of the < | ||
| We add new handlers which correspond to the above interfaces which are all defined in a new struct: | We add new handlers which correspond to the above interfaces which are all defined in a new struct: | ||
| - | < | + | < |
| typedef struct _zend_class_dimensions_functions { | typedef struct _zend_class_dimensions_functions { | ||
| /* rv is a slot provided by the callee that is returned */ | /* rv is a slot provided by the callee that is returned */ | ||
| Line 827: | Line 820: | ||
| bool (*has_dimension)(zend_object *object, zval *offset); | bool (*has_dimension)(zend_object *object, zval *offset); | ||
| zval *(*fetch_dimension)(zend_object *object, zval *offset, zval *rv); | zval *(*fetch_dimension)(zend_object *object, zval *offset, zval *rv); | ||
| - | void (*write_dimension)(zend_object *object, zval *offset, zval //value); | + | void (*write_dimension)(zend_object *object, zval *offset, zval *value); |
| void (*append)(zend_object *object, zval *value); | void (*append)(zend_object *object, zval *value); | ||
| zval *(*fetch_append)(zend_object *object, zval *rv); | zval *(*fetch_append)(zend_object *object, zval *rv); | ||
| Line 854: | Line 847: | ||
| Another consequence of using the new algorithm is that some idiosyncratic code that produces side effects | Another consequence of using the new algorithm is that some idiosyncratic code that produces side effects | ||
| - | in the <php>had_dimension</ | + | in the <php>has_dimension</ |
| this also applies to userland classes implementing < | this also applies to userland classes implementing < | ||
| For example, the following code: | For example, the following code: | ||
| Line 886: | Line 879: | ||
| As the < | As the < | ||
| + | |||
| + | == Removal of the zend_class_arrayaccess_funcs struct and CE pointer == | ||
| + | |||
| + | As the < | ||
| + | and it cannot fulfill its role anymore with the new dimension handlers, | ||
| + | the struct is removed and alongside it the pointer to such a struct on the < | ||
| + | |||
| + | == Throw an Error when trying to increment or decrement an object offset == | ||
| + | |||
| + | Incrementing/ | ||
| + | whereas using < | ||
| + | and usually when acting on object properties this is what happens. | ||
| + | |||
| + | The current limitation is because we do not have sufficient specialized VM opcodes for this specific case. | ||
| + | Therefore, we propose to hard error, like we do for string offsets, so that if we do add the relevant opcodes we can | ||
| + | properly support the Read-Write behaviour. | ||
| == Changes to ArrayObject == | == Changes to ArrayObject == | ||
| Line 896: | Line 905: | ||
| * Fix < | * Fix < | ||
| * When using an object as a backing value: | * When using an object as a backing value: | ||
| - | | + | |
| - | * Emit dynamic properties warning when using an object as a backing value that does not allow dynamic properties | + | * Emit dynamic properties warning when using an object as a backing value that does not allow dynamic properties |
| - | * Throw < | + | * Throw < |
| - | * Throw < | + | * Throw < |
| + | * Throw < | ||
| * Continue to ignore any < | * Continue to ignore any < | ||
| Line 910: | Line 920: | ||
| but it also " | but it also " | ||
| - | Our solution is to add legacy dimension handlers to classes that implement | + | The solution |
| - | reproducing the current behaviour | + | < |
| + | but also add legacy dimension handlers | ||
| + | such an object. | ||
| However, if one of the new interfaces is implemented for dedicated support to appending, fetching, | However, if one of the new interfaces is implemented for dedicated support to appending, fetching, | ||
| and fetch-appending, | and fetch-appending, | ||
| Line 931: | Line 943: | ||
| As such we propose to deprecate the aliases in favour of the normal offset methods. | As such we propose to deprecate the aliases in favour of the normal offset methods. | ||
| - | == Changes to MultipleIterator== | + | == Changes to MultipleIterator == |
| The implementation of < | The implementation of < | ||
| Line 957: | Line 969: | ||
| The < | The < | ||
| type removed from the union type for the < | type removed from the union type for the < | ||
| + | |||
| + | Moreover, most code that expects resources as offsets already use an explicit < | ||
| == Emit warnings for invalid offset types on arrays == | == Emit warnings for invalid offset types on arrays == | ||
| Line 990: | Line 1004: | ||
| </ | </ | ||
| - | === Emit warning on read-write operations on <php>null</ | + | === Emit warning on read-write operations on null container === |
| Emit the same warning as a simple read operation when using < | Emit the same warning as a simple read operation when using < | ||
| Line 1012: | Line 1026: | ||
| </ | </ | ||
| - | And if the specific operation is not supported the error would ressemble: | + | And if the specific operation is not supported the error would resemble: |
| - | < | + | < |
| Cannot OPERATION offset of type TYPE on value of type TYPE | Cannot OPERATION offset of type TYPE on value of type TYPE | ||
| </ | </ | ||
| Line 1038: | Line 1052: | ||
| Promote all warnings to < | Promote all warnings to < | ||
| + | |||
| + | ===== Backward Incompatible Changes ===== | ||
| + | |||
| + | A recap of the BC breaking changes being introduced in PHP 8.4: | ||
| + | |||
| + | * New algorithm when calling < | ||
| + | * < | ||
| + | * Trying to read offsets of a < | ||
| + | * Leading numeric string used as an offset for strings would throw an < | ||
| + | * Float numeric strings (i.e. non integer-numeric strings) used as an offset for strings would now throw an < | ||
| + | * < | ||
| + | |||
| + | A recap of the new warnings being introduced in PHP 8.4: | ||
| + | |||
| + | * A warning when checking the existence of an offset on invalid container types (except for < | ||
| + | * A warning prior to reading an undefined offset when it is part of a read-write operation | ||
| + | * A warning is emitted when trying to use a value of type < | ||
| + | * A warning is emitted when trying to check the existence of string offset with invalid offset types | ||
| + | |||
| + | A recap of the deprecations being introduced in PHP 8.4: | ||
| + | |||
| + | * The < | ||
| + | * The < | ||
| + | * The < | ||
| + | |||
| + | For details on each of those, refer to their relevant sections. | ||
| ===== Version ===== | ===== Version ===== | ||
rfc/container-offset-behaviour.1720100986.txt.gz · Last modified: 2025/04/03 13:08 (external edit)