clang could do a better job at warning for implicit conversion to bool

[llvmbot]

Bugzilla Link	10159
Version	trunk
OS	All
Blocks	llvm/llvm-bugzilla-archive#22259
Reporter	LLVM Bugzilla Contributor
CC	@akyrtzi,@DougGregor,@seanm

Extended Description

MSVC will emit warnings for the following code, clang not:

enum E {
e = 3,
};

int main() {
bool var1 = e; // truncation from 'E' to 'bool'
bool var2 = 3; // truncation from 'int' to 'bool'
}

[akyrtzi]

cloned to rdar://problem/9640246

[llvmbot]

Was about to file a dup, so count me as +1.

The situation is IMHO worse than the previous example shows, since all pointer types are also freely assignable to bool.

bool var1 = "string";
bool var2 = &some_struct
bool var3 = [NSMutableArray array]; // obj-c

I ran into this not with assignment but with return values. I changed an Obj-C method's return type from NSString* to bool, and didn't fix one of the 'return' statements in the method; I didn't figure this out until a test along that code path exhibited the wrong behavior due to true being returned when it should have been false.

[llvmbot]

mentioned in issue llvm/llvm-bugzilla-archive#22259

[wheatman]

Two of the 4 cases seem to be fixed
https://godbolt.org/z/G53cW4e7c

code

enum E {
e = 3
};


struct B{};

int main() {
    // no wanring
    bool var1 = e; // truncation from 'E' to 'bool'
    // no warning
    bool var2 = 3; // truncation from 'int' to 'bool'
    // warning: implicit conversion turns string literal into bool: 'const char[7]' to 'bool' [-Wstring-conversion]
    // from -Weverything
    bool var3 = "string";
    B b;
    // warning: address of 'b' will always evaluate to 'true' [-Wpointer-bool-conversion]
    // default warning
    bool var4 = &b;
    return var1+var2+var3+var4;
}