[PATCH] D14227: Add a new attribute: norecurse

[Aaron Ballman via llvm-commits]

On Mon, Nov 2, 2015 at 3:32 PM, James Molloy <james at jamesmolloy.co.uk> wrote:
> Understood.
>
> My expectation would be that a frontend would only generate these when it
> has some knowledge which is difficult to infer. For example, "main" in C++.
> I have a change in phab that causes the attribute to be inferred from the
> callgraph where possible, and I'd expect this to be the general usecase, as
> with LTO you have the whole callgraph.

That makes me much more comfortable. :-) Thank you for the clarification.

~Aaron

>
> James
>
> On Mon, 2 Nov 2015 at 20:19 Aaron Ballman <aaron.ballman at gmail.com> wrote:
>>
>> I am less concerned about the llvm attribute than I am the clang side of
>> things, if it is intended to be exposed or generated there. This strikes me
>> as the sort of attribute that may be easy to overuse (with either polarity).