Bug 29863 - Segmentation fault vs invalid results, memory models, and control/data dependencies.

[Florian Weimer]

* Carlos O'Donell via Libc-alpha:

> - We should pursue performance given a well formed understanding of the memory
>   models glibc supports e.g. ISO C11, and POSIX Issue 7.

But the C memory model is known to be broken.  For example, it does not
rule out the existence of out-of-thin-air values, which makes relaxed MO
pretty useless.  I don't think we can use it as an argument to reject
requests from application developers.

> - Authors using OCC should explain their memory model in detail, and
>   may need to raise the issue with the broader toolchain authors to gain
>   consensus for the design constraints imposed by OCC on the
>   implementation language being compiled e.g. compiler picks a special
>   memcpy, or does other operations.

I think their requirements for memcmp are pretty clear.  We don't need a
fullly fleshed out memory model for that.

Thanks,
Florian