Your compiler, or rather your architecture definition supplied to the compiler, is broken. NULL should be an invalid pointer value — it need not actually be memory address 0. (The explicit constant 0 in a pointer context should be automatically translated by the compiler into this special invalid pointer value.) There have been several such architectures already, so this should not present an insurmountable problem.
Re:GCC 6.1 Released
by Kristine1975 on Wednesday April 27, 2016 @03:05PM
(#11583548)
Same thing in say the Linux kernel, where gcc removing a NULL pointer check actually created a vulnerability: http://lwn.net/Articles/342330/
Re:GCC 6.1 Released
by brandmeyer on Wednesday April 27, 2016 @04:36PM
(#11584308)
Yep. In fact, it was this very vulnerability that brought my attention to this issue in embedded systems.
AIUI, this code is not affected:
X *foo = ...;
if (foo) foo->bar();
But this code will be: X *foo = ...;
foo->bar();
if (foo) { ... }
Since foo is dereferenced unconditionally, the compiler is allowed to assume that it will not be null in the above code, and therefore the test on the third line to determine whether foo is null or not is useless and can be omitted.
Re:GCC 6.1 Released
by iso-8859-1 on Wednesday April 27, 2016 @04:28PM
(#11584237)
Since this code is broken, wouldn't it be better to emit a warning and let the programmer fix it?
The real case where it matters is where the "if (foo) { ... }" test happens in a function that's been inlined into the caller, but not all the callers do the foo->bar() dereference. So it's only being removed in some cases.
Re:GCC 6.1 Released
by brandmeyer on Wednesday April 27, 2016 @04:40PM
(#11584331)
There used to be a fairly broad warning (-Wunreachable-code), but it caused so many false positives and false negatives that it was removed.
In my particular case, I'm programming in a microcontroller environment where there is ordinary RAM at the NULL address. Dereferencing NULL therefore doesn't halt the program at all.