On Mon, 22 Apr 2013 11:44:58 +0000 "Warlich, Christof" <christof.warlich@xxxxxxxxxxx> wrote: > But imho, anything, even "undefined (but fixed)", would be better than > "varying depending on the level of optimization". "undefined" behavior can't be "fixed". if it is fixed, it is defined. the only way one can get "fixed" behavior is to change the standard, so that standard will clearly specify what to do in such cases. until than compiler authors are free to change results of undefined operations. it's really a bad practice to depend on undefined behavior. this will tie the software to the given version of the given compiler and architecture. even recompiling the software with the exact same compiler version, but for ARM instead of x86, for example, can break everything. so in practice one should assume that result of "undefined behavior" can be *anything*, including creating new Black Hole or Sun becomes Supernova.
Attachment:
signature.asc
Description: PGP signature
