Re: netbsd-11 gcc bug

[Mouse <mouse%Rodents-Montreal.ORG@localhost>]

> The reason for the "belief" that UB is (often) related to optimizers
> is squarely resting on the fact that when the optimizer is turned off
> then the UB goes away.

No, it doesn't.  It is still UB.  All that changes is what the compiler
chooses to do in that case where its behaviour is undefined by the
spec.

A more accurate phrasing would be that the compiler's choice of
behaviour changes depending on optimization.

The compiler must do *something* when faced with UB, whether it's
compiling code that does what the source author intended, segfaulting,
making usefulness-impairing inferences, or (in non-runtime cases)
printing all primes from 2 to 4294967291 at compile time.  UB just
means that the spec does not constrain what the compiler does.

> As I said, efforts to eliminate UB from C are actually happening at
> the standards committee level, so....

That will be difficult, unless some of them are handled by converting
UB to IDB by fiat.  Indirecting through a garbage pointer comes to mind
as an example.  (You're not going to get rid of garbage pointers
without crippling the language for use cases such as implementing VM
subsystems.)

/~\ The ASCII				  Mouse
\ / Ribbon Campaign
 X  Against HTML		mouse%rodents-montreal.org@localhost
/ \ Email!	     7D C8 61 52 5D E7 2D 39  4E F1 31 3E E8 B3 27 4B