Re: [PATCH v5 2/5] C++: Support clang compatible [[musttail]] (PR83324)

[Jason Merrill <jason@redhat.com>]

On 5/14/24 19:23, Andi Kleen wrote:
>>>> You need a template testcase; I expect it doesn't work in templates with the
>>>> current patch.  It's probably enough to copy it in tsubst_expr where we
>>>> currently propagate CALL_EXPR_OPERATOR_SYNTAX.
>>>
>>> I tried it with the appended test case, everything seems to work without
>>> changes.
>>>
>>> Does it cover the cases you were concerned about?
>>
>> Not fully; this testcase doesn't seem to check for errors if tail-call
>> fails, only whether the syntax is accepted.  So it would pass if the
>> attribute were simply ignored.
> 
> Okay I'm not clear how I would do that. Pattern match the assembler
> in a target specific test case? From looking at the assembler output
> everything got tail converted.

Write a testcase where the tail-call optimization can't happen, perhaps 
because the caller and callee disagree on return type:

int f();

double h() { [[gnu::musttail]] return f(); } // error

template <class T>
T g() { [[gnu::musttail]] return f(); }

int main()
{
   g<int>();
   g<double>(); // should error, but doesn't with v6 patch set
}

Jason