1.4.0

[Jeckerson]

Added

• Added PHP compatible string offsets, so reading, writing, isset(), empty() and unset() on a string behave as they do in PHP, including negative offsets, out of range warnings, dynamically typed sources and for c in s #760
• Added an unassigned-variable warning, on by default, for a local that is read but never assigned #2654

Changed

• Widened the conditional-initialization warning to every local that can be read before it is assigned, instead of only one assigned exactly once in a deeper branch #2679

Fixed

• Fixed Closure::bindTo(), Closure::bind() and Closure::call() dropping a closure's use (...) captures, which read back as null or crashed the process #2667
• Fixed var_dump() and print_r() of a closure crashing the process on PHP 8.4 and later, and reporting the internal capture carrier as $this instead of the enclosing object on every version #2667
• Fixed a compile error being replaced by a deprecation notice on PHP 8.4 and later, by marking six implicitly nullable parameters as nullable and letting a deprecation report as PHP reports it #2663
• Fixed a by-reference call argument reading a copy of the container instead of its storage slot, which lost or aliased a write to a shared array, reported a missing element PHP creates, and freed the property's array when the callee replaced its argument #2691
• Fixed a by-reference subscript argument borrowing the container value, which lost the write whenever the value was shared and leaked the offsetGet() result for an ArrayAccess container #2691
• Fixed a read-only subscript read of an ArrayAccess object releasing the value offsetGet() owns, which leaked 96 bytes per call or corrupted the heap, and limited the read-only shortcut to a container proven to hold a native array #2682
• Fixed explode() returning the limit = 2 result for a limit of 0, 1 or negative, and looping forever on an empty separator where PHP throws ValueError #2674
• Fixed fetch and empty() on an ArrayAccess object releasing the string offset before offsetGet() received it, so the object was silently handed the key offsetget #2656
• Fixed the array isset helpers never releasing the offsetExists() return value, and holding no reference to the container across the two userland calls, as PHP's own zend_std_read_dimension() does #2656
• Fixed unset obj->property leaking the property name string on every call #2656
• Fixed a leaked exception instance when a throw cannot construct its exception, a leaked callable when an internal call names an unknown function, and an uninitialized return slot in the zephir_return_call_*() helpers #2656
• Fixed a declared but never assigned variable reaching userland as an uninitialized zval, printed by var_dump() as UNKNOWN:0, where PHP evaluates an unset variable as null #2654
• Fixed a variable assigned only inside a conditional branch reaching userland as an uninitialized zval on a path that skips the assignment #2679
• Fixed the zephir_array_*_long() helpers truncating a negative or large index to 32 bits on Windows, where unsigned long is narrower than zend_long, so s[-1] read index 4294967295 and a write tried to grow the string to 4 GB #2666
• Fixed Zephir long/ulong mapping to a C long, 32-bit on Windows, so a 64-bit value truncated and Z_PARAM_LONG() stored 8 bytes into a 4-byte local #2666
• Fixed the kernel signatures that carry a PHP integer (zephir_get_intval_ex(), the comparison, division and modulo helpers, zephir_substr(), zephir_fast_explode(), zephir_preg_match() and the multi-dimensional array key) using a C long #2666
• Fixed a multi-dimensional array index reaching zephir_array_update_multi() as a C int where the callee reads a zend_long, so a[0][1] produced keys such as 140733193388033 on Windows #2666
• Fixed a uint/ulong multi-dimensional array index being passed by address, so the pointer value became the array key #2666
• Fixed six zephir_safe_mod_*() helpers being declared and emitted but never defined, so any % with a float operand died at load with undefined symbol #2666
• Fixed % yielding a float instead of an int, which lost every result above 2^53 #2666
• Fixed / 0 and % 0 raising an E_WARNING and returning 0 instead of throwing DivisionByZeroError as PHP 8 does #2666
• Fixed PHP_INT_MIN % -1 crashing the process with SIGFPE #2666
• Fixed a float divisor being silently truncated to an integer by / and % #2666
• Fixed a -Wformat warning in zephir_fclose() by casting the resource handle to zend_long and formatting it with the portable ZEND_LONG_FMT macro instead of %d, preserving compatibility with PHP 8.0's int handle.
• A variable whose only consumer is a closure's use (...) clause now counts as used: it is no longer reported as unused-variable, and it is declared in the generated C. A declared-but-unassigned capture was skipped by both, so the generated code referenced an undeclared identifier and the extension failed to build. Capturing a variable that was never declared now fails with Cannot capture variable 'x' because it wasn't declared instead of a PHP fatal error #2029
• Fixed array defaults on a class (typed array property defaults, trait array property defaults and array class constants), which were one shared table that every instance mutated in place instead of being copy-on-write #2651
• Fixed the unchecked length arithmetic in the zephir_concat_*() helpers, which now throw String size overflow like PHP instead of writing past the allocation #2657
• Fixed let s .= n with an integer operand calling the declared but never defined zephir_concat_self_long(), so the extension built and then died at load with undefined symbol, and made the same append work on a var left-hand side #2660
• Fixed closure use (...) captures being shared by every closure created from the same source line, and implemented use (&x) as a real reference #2652
• Fixed the closure use (...) clause dropping const, which is now a read-only capture, and a by-reference capture of an explicitly typed local #2653
• Fixed a closure declared inside another closure never having its source file generated, so the extension failed to link, and this inside it resolving against the outer closure instead of the enclosing class #2655
• Fixed every closure invocation leaking one reference of the closure object and of each captured value #2652

---

This discussion was created from the release 1.4.0.