No uninitialized warning for member variable and getter

[fekir]

#include <cstdio>

struct a{
    int i;
    int get_i() const {
        return i;
    }
    a() : i(get_i()){}
};


int main(){
    a instance = a();
    std::printf("%d", instance.i);
}

clang does not generate any warning, even with optimizations enabled.

It generates a warning if a() : i(get_i()){} is changed to a() : i(i){}.

Example on godbolt, with comparison with GCC, which does generate a warning when optimizations are enabled

https://godbolt.org/z/6E43zroWj

[DimitryAndric]

I can at least confirm that clang trunk doesn't warn at all. So it created a nice "random" number generator:

$ ./pr78129
548438528
$ ./pr78129
554314800
$ ./pr78129
543163808
$ ./pr78129
547619032
$ ./pr78129
539220120
$ ./pr78129
544666152

What is also interesting is that UBSan does not warn either:

$ clang++ -O0 -g -fsanitize=undefined pr78129.cpp -o pr78129
$ ./pr78129
2354616

In any case I definitely agree that it should warn. And UBSan should catch this somehow.

[dwblaikie]

Generally clang's warnings are not interprocedural - it's too expensive to do at compile time. So analyzing get_i to warn in the ctor is probably out of scope/infeasible.

UBSan catching this seems like something to look into, at least.