Undefined weak symbols should get a value of 0

[hamish]

This has apparently already been fixed for kernel code in #6272 but persists in runtime_loader. Weak symbols that are unresolved should get set to 0.

For example, this program should run and return successfully:

#include <stdlib.h>
extern void func() __attribute__ ((weak));
int main() {
    // Always false but won't be optimized away.
    if (malloc(1)==(void*)1) func();
    return 0;
}

But instead runtime_loader complains about unresolved symbols.

[X512]

Is this related to unsolved symbols of reserved methods like «_ZN5BView15_ReservedView12Ev» when this virtual methods are implemented?

[bonefish]

I could have sworn the feature was implemented in the runtime loader at the same time, but obviously that's not the case.

The implementation is rather easy: In ​resolve_symbol() lookupError simply needs to remain ERROR_UNPATCHED, if sym is weak (i.e. put the assignment in a respective "if").

I can't promise that I get to implement it anytime soon. So, if there are any takers ...

[bonefish]

Mmh, the symbol patcher stuff makes that all a bit more complicated. The interface has to be changed, I'm afraid. We should also fix the incorrect assumption that a value of NULL means the symbol is undefined.

[cian]

This is currently affecting doing a clean build of VLC/libvlc - weak symbol in the core library leads to runtime errors when trying to run the application or even test suites.

If the code is edited to use the non-weak version (intended for non-ELF targets), it works.