Bug 23731 – [REG2.102] Throwing within finalizers causes InvalidMemoryOperationError on Windows

Testcase: https://github.com/dlang/dmd/blob/v2.102.0/druntime/test/allocations/src/alloc_from_assert.d (skipped by DMD on Windows) Analysis starting from: https://github.com/dlang/dmd/pull/14710#issuecomment-1437447865

@kinke updated dlang/dmd pull request #14901 "[stable] druntime: Fix InvalidMemoryOperation error on Windows when throwing exceptions from finalizers" fixing this issue: - Fix Issue 23731 - InvalidMemoryOperation error on Windows when throwing exceptions from finalizers A v2.102 regression, which newly enabled `StackTrace` constructions from within GC finalizers. The construction is `@nogc`, but apparently depends on the `typeid(StackTrace)` monitor having been allocated already, i.e., a previous `StackTrace` construction. So pre-allocate a dedicated `Mutex` in the module constructor to make the `StackTrace` construction really `@nogc`. `druntime/test/allocations/src/alloc_from_assert.d` then passes on Windows again (not tested by DMD, but by LDC). While at it, also prevent `InvalidMemoryOperation` errors when trying to print exceptions within finalizers - by not resolving anything (=> empty printed trace), as resolving currently heavily depends on the GC. https://github.com/dlang/dmd/pull/14901

dlang/dmd pull request #14901 "[stable] druntime: Fix InvalidMemoryOperation error on Windows when throwing exceptions from finalizers" was merged into stable: - 936e7e26bbefeb2172686f5c828655ac9e521376 by Martin Kinkelin: Fix Issue 23731 - InvalidMemoryOperation error on Windows when throwing exceptions from finalizers A v2.102 regression, which newly enabled `StackTrace` constructions from within GC finalizers. The construction is `@nogc`, but apparently depends on the `typeid(StackTrace)` monitor having been allocated already, i.e., a previous `StackTrace` construction. So pre-allocate a dedicated `Mutex` in the module constructor to make the `StackTrace` construction really `@nogc`. `druntime/test/allocations/src/alloc_from_assert.d` then passes on Windows again (not tested by DMD, but by LDC). While at it, also prevent `InvalidMemoryOperation` errors when trying to print exceptions within finalizers - by not resolving anything (=> empty printed trace), as resolving currently heavily depends on the GC. https://github.com/dlang/dmd/pull/14901

dlang/dmd pull request #14938 "merge stable" was merged into master: - a54fbab882b8a4c7fc86086dd1a414e9324dea6c by Martin Kinkelin: [stable] druntime: Fix InvalidMemoryOperation error on Windows when throwing exceptions from finalizers (#14901) * Fix Issue 23731 - InvalidMemoryOperation error on Windows when throwing exceptions from finalizers A v2.102 regression, which newly enabled `StackTrace` constructions from within GC finalizers. The construction is `@nogc`, but apparently depends on the `typeid(StackTrace)` monitor having been allocated already, i.e., a previous `StackTrace` construction. So pre-allocate a dedicated `Mutex` in the module constructor to make the `StackTrace` construction really `@nogc`. `druntime/test/allocations/src/alloc_from_assert.d` then passes on Windows again (not tested by DMD, but by LDC). While at it, also prevent `InvalidMemoryOperation` errors when trying to print exceptions within finalizers - by not resolving anything (=> empty printed trace), as resolving currently heavily depends on the GC. * [work around cyclic modules dependency...] * druntime: Test test/allocations/src/alloc_from_assert.d on Windows too As it regressed with v2.102. https://github.com/dlang/dmd/pull/14938