On Mon, 4 May 2026 14:52:37 GMT, Anton Artemov <[email protected]> wrote:
> Hi, please consider the following changes:
>
> This PR re-enables stringop-overflow warnings.
>
> It was assumed that the assert in `Thread::current()`, which calls the
> `ATTRIBUTE_NORETURN` report function on failure, would be sufficient to
> inform the GCC compiler that the current thread is always non-null. However,
> it turns out that the assert is weakened by the `DebuggingContext`:
>
> Instead of having
> `assert(current != nullptr, "Thread::current() called on detached thread");`
>
> in fact one gets this:
> `if ( !(DebuggingContext::is_enabled() || current != null) ) {
> report_vm_error(...); }`
>
> If `DebuggingContext` is enabled then the error will not be triggered
> regardless of what `current` is. This makes GCC consider the virtually
> impossible path and emit false positive warnings about atomics.
>
> The same problem is found in the Shenandoah code.
>
> The fix is to strengthen the assertion by adding a new kind of assert called
> `invariant_assume()`
> which removes the assertion completely in product builds, but for debug
> builds makes it clear for the compiler that the operand cannot be nullptr.
>
> With this findings I conclude that it is not a compiler bug.
>
> Tested in tiers 1-5 and GHA.
>
> Separate builds with Shenandoah done.
>
> ---------
> - [x] I confirm that I make this contribution in accordance with the [OpenJDK
> Interim AI Policy](https://openjdk.org/legal/ai).
This pull request has been closed without being integrated.
-------------
PR: https://git.openjdk.org/jdk/pull/31025
