On 19 March 2014 15:25, Richard Henderson <r...@twiddle.net> wrote: > On 03/18/2014 11:21 PM, Stefan Weil wrote: >> Are there test cases which fail with the old code? > > Only when running under analysis tools looking for technical violations of the > standard. Under normal circumstances it would never crash since, as with the > native backends, the underlying cpu is going to treat this some rational way.
I think the major reason for fixing this kind of thing is that modern C compilers are becoming increasingly adversarial, and if we don't try to clean up our use of undefined behaviours one day gcc is going to decide that it can optimize our code into something that breaks... thanks -- PMM
