fix allocation profiling shutdown with USE_ZEND_ALLOC=0 crashing (#2294)

Author: realFlowControl

Diff for: profiling/src/allocation.rs

@@ -176,6 +176,16 @@ pub fn allocation_profiling_rinit() {
 }
 
 pub fn allocation_profiling_rshutdown() {
+    let allocation_profiling = REQUEST_LOCALS.with(|cell| {
+        cell.try_borrow()
+            .map(|locals| locals.profiling_allocation_enabled)
+            .unwrap_or(false)
+    });
+
+    if !allocation_profiling {
+        return;
+    }
+
     // If `is_zend_mm()` is true, the custom handlers have been reset to `None`
     // already. This is unexpected, therefore we will not touch the ZendMM handlers
     // anymore as resetting to prev handlers might result in segfaults and other

Diff for: profiling/tests/phpt/allocation_jit_01.phpt

@@ -0,0 +1,33 @@
+--TEST--
+[profiling] Allocation profiling should not crash when run with `USE_ZEND_ALLOC=0`
+--DESCRIPTION--
+When running PHP with `USE_ZEND_ALLOC=0` and disabled allocation profiling
+(either explicit disabled or due to the JIT bug), PHP will install the system
+allocator in https://heap.space/xref/PHP-8.2/Zend/zend_alloc.c?r=4553258d#2895-2918
+and set the `AG(heap)->use_custom_heap` to `ZEND_MM_CUSTOM_HEAP_STD` (which is 1),
+so future calls to `is_zend_mm()` will return false which might lead to a situation
+where we assume we are hooked into, while we are not.
+--SKIPIF--
+<?php
+if (!extension_loaded('datadog-profiling'))
+    echo "skip: test requires datadog-profiling", PHP_EOL;
+?>
+--ENV--
+USE_ZEND_ALLOC=0
+--INI--
+datadog.profiling.enabled=yes
+datadog.profiling.log_level=debug
+datadog.profiling.allocation_enabled=yes
+datadog.profiling.experimental_cpu_time_enabled=no
+zend_extension=opcache
+opcache.enable_cli=1
+opcache.jit=tracing
+opcache.jit_buffer_size=4M
+--FILE--
+<?php
+echo "Done.", PHP_EOL;
+?>
+--EXPECTREGEX--
+.*Done.
+.*Stopping profiler.
+.*