Re: [PATCH] C undefined behavior fix

[Joe Buck <jbuck at synopsys dot COM>]

Richard Henderson writes:
> You will _not_ like the code you'll get out of gcc if we turn off
> looking through PLUS.  Which I don't even know how to do -- that
> sort of thing is scattered all over the compiler.

Agreed.  We must not turn off looking through

	pointer + expression

as it will kill us.

However, I think we have to turn off looking through the + in
	(typeof pointer)((unsigned long)(pointer) + expression)

to be correct, or, rather, remove any assumption that we know where
the resulting pointer points.  We cannot assume that the above
expression is equivalent to

	pointer + expression / sizeof *pointer 

(meaning that it is known to be a valid pointer somewhere into the same
array, string literal etc) unless we can get bounds on the range of
expression.

Such operations will appear, typically, only in kernel and embedded
code, or allocators (to get things to line up with page boundaries),
and doing worse on some cases isn't going to hurt us significantly.

> I will at this point give up on this disagreement.  On a practical
> level, even if I were to concede your point (which I don't) there
> is absolutely nothing that can be done without starting over from
> scratch with another compiler.

That's too pessimistic, I think.