Re: [C++] PR19531 NRV is performed on volatile temporary

[Christian BRUEL <christian dot bruel at st dot com>]

Mark Mitchell wrote:
<< snap>>

> Is the check for volatility on both the return type of the function --
> TREE_TYPE (TREE_TYPE (fndecl)) -- and on the RESULT_DECL -- TREE_TYPE
> (r)) -- necessary?
>
> I would expect that checking the volatility of the function return type
> is sufficient; that's the language issue.  If, for whatever reason, the
> compiler were to make the RESULT_DECL volatile, for a function with a
> non-volatile return, I believe that the NRV would still be valid.
>
> Would you please test removing the check on the RESULT_DECL from your patch?

it is the other way: with the nrv8.C (attached in the patch) test, only 
the check on the RESULT_DECL is necessary, testing the volatility of the 
*volatile A l;* object.  Without it the check doesn't pass.
I think this is what 12.8 15 was originally about, before Core DR20 
extended it to the function's return type.

I also though that there would be a language issue with the volatility 
of the function's return type, but the volatility flag is not set on the 
return type of the cpctor. Should it be ? I'm not sure since with:
A (volatile const A & o) { d = o.d + 2; }
the cpctor return type is not volatile, only the temporary object is, 
but I'd prefer to let that to the experts's hands.

So I conservatively added the check on the function's return type to 
match core dr20 extension (quoted in the bugzilla entry but I don't have 
it) although I was not able to make up a test case.

Best Regards
Christian