https://gcc.gnu.org/bugzilla/show_bug.cgi?id=128829

            Bug ID: 128829
           Summary: Tail-call failure due to IPA-induced return type
                    mismatch
           Product: gcc
           Version: 15.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: ipa
          Assignee: unassigned at gcc dot gnu.org
          Reporter: tonalan at ebay dot com
  Target Milestone: ---

When a split has no return value or does not return, the outlined
function becomes `void`-returning.  However, if the enclosing function
is non-`void` and has a `musttail` call, the call to the outlined split
is also annotated with `musttail`, leading to a return type mismatch.

```
// Compile with `-O3`, GCC >=15.
// Sandbox: <https://godbolt.org/z/bzMxM5Eo6>.

template <int N>
static int f(int *);
[[noreturn]] void fail(int *, int *, int *);

int (*fs[])(int *) = { f<1>, f<2>, f<3> };

template <int N>
static int f(int *p)
{
  if (*p < 0)
    fail(p, p, p);
  [[gnu::musttail]] return fs[*p](p + N);
}

int go(int *p)
{
  return f<1>(p);
}
```

I submitted a patch fixing this[^1], but was told to also open a bug
report since the fix might be backported.  I was not able to find any
reports demonstrating this exact failure mode among the open
IPA+musttail bugs.

[^1]: https://gcc.gnu.org/pipermail/gcc-patches/2026-November/737370.html

Reply via email to