On Tuesday, 14 July 2015 at 12:59:34 UTC, Ola Fosheim Grøstad
wrote:
My point is that C UB for overflow on signed int does not imply
not having the same code-gen as D has. So it is essentially not
a language problem per se.
UB implies anything. Yes, it's not a problem, safer languages
based on C are possible, and were done.
The "problem" is cultural. C programmers have this idea that
they should compile everything with the compiler/compiler
setting that gives the absolutely highest performance no matter
what the quality the code.
It's believed that there's no problem with optimized code and
optimizations don't change behavior.
The same thing would happen if LDC added a switch named
"-FAST_AND_RISKY" ;-).
I proposed -Ounsafe, it can actually help with correctness,
because it clearly states the tradeoff and keeps it opt-in
instead of being default, as C compilers do, and it also fits
well into D approach to unsafety.
