[llvm-commits] Please review: adding Function attribute "ThreadSafety"

[Bill Wendling]

On Feb 4, 2013, at 6:00 AM, Kostya Serebryany <kcc at google.com> wrote:

> [resurrecting an old thread... again]
> 
> Chris, Bill, 
> 
> Is the new Attribute implementation good enough to support two more function attributes? 
> We need two attributes do enable/disable tsan and msan with function granularity. 
> Just like with asan, where we have AddressSafety attribute mapped to __attribute__((no_address_safety_analysis)), 
> I propose to add ThreadSafety attribute (will map to the existing __attribute__((no_thread_safety_analysis)))
> and UninitializedChecks attribute (will map to a new __attribute__((no_uninitialized_checks)))
> (naming is hard, I'd love to hear better suggestions). 
> 
> Thoughts? 
> 
I don't see a problem with going ahead and adding the new attributes...unless you can wait for a couple of weeks, which is when I'm hoping to have something more complete to offer the community. :-) The only difference between doing it now and waiting is that you'll be adding your new attributes in the old bit-mask format. That's not a problem on the face of it, but just slightly uglier in general.

As far as naming things, I'm rather bad at it, so I can't help you there...

-bw