https://gcc.gnu.org/bugzilla/show_bug.cgi?id=124762
Bug ID: 124762
Summary: GCC should reject (not just warn) narrowing
conversions from pointer to bool
Product: gcc
Version: 16.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: c++
Assignee: unassigned at gcc dot gnu.org
Reporter: peppe at gcc dot gnu.org
Target Milestone: ---
I understand that GCC took the stance that, in general, narrowing conversions
in list initializations from non-constants are accepted, because otherise they
would break too much code:
char c1 { 42 }; // 1: OK (and legal)
char c2 { 10'000 }; // 2: ERROR (ill-formed)
int i = 123;
char c3 { i }; // 3: WARNING
GCC rejects #2 (narrowing from a constant), but for #3 it limits itself to emit
a diagnostic and accepts the result. (For comparison, Clang rejects the last
one.) This is fully conforming: the code is ill-formed, a diagnostic is issued,
the code is still accepted as a compiler extension.
---
However since C++20 pointer to bool conversions are considered narrowing. I
think that for *those* GCC shouldn't be emitting warnings and accepting, but
unconditionally reject.
Consider:
bool b1 { (int *)0 }; // 4: ERROR, narrowing from constant expression
bool b2 { "test" }; // 5: WARNING¹
int *ptr = nullptr;
bool b3 { ptr }; // 6: WARNING
In my opinion all of these should be rejected. There's no actual concrete use
case for allowing pointer->bool narrowing anywhere (unlike, say, int->char).
For pointer->bool the Standard does not even carve out certain constant
expressions, exempting them from being narrowing conversions. They always are.
And GCC already rejects (some of) them.
---
Separate issue: why is #5 just a warning? Is there an issue with constant
folding?
constexpr const char* p = "test";
bool b { p }; // WARNING, should be ERROR
static int x;
bool c { &x }; // WARNING, should be ERROR
constexpr int* q = nullptr;
bool d { q }; // ERROR