[arch-general] gcc: loop do not terminate

[Olivier Langlois]

On Tue, 2013-05-14 at 04:50 -0400, Daniel Micay wrote:
> On Tue, May 14, 2013 at 3:49 AM,  <goodmenzy at gmail.com> wrote:
> >   If this function initArr() is moved to the same cpp file of main(),
> >   all optimization level get the same result.
> 

Good observation. Having a separate TU for init function was intentional
to trigger the problem.

> That's the great thing about undefined behaviour, you never know what
> you'll get. It's really not something anyone should be relying on
> though.... there's a well defined, valid way of declaring a
> flexible-length struct member.

There is no magic or random involved. The compiler interpret your code
and take decisions on what it thinks is good for you. If it sees the
init function, it figure outs that it cannot optimize away the loop.

If you hides the initialization by calling lets say a shared library
function, the compiler will make assumption that it is safe to optimize
the loop.