MSVC: shift count negative or too big

Question

I am trying to wrap my head around a warning I get from MSVC. From what I can tell, the warning seems to be bogus, but I would like to be sure.

I am trying to convert an off_t to an offset in an OVERLAPPED. Given an off_t named offset and an OVERLAPPED named overlapped, I am trying the following:

overlapped.Offset = static_cast<DWORD>(offset);
if constexpr(sizeof(offset) > 4) {
    overlapped.OffsetHigh = static_cast<DWORD>(offset >> 32);
}

MSVC complains about the bitshift, pretending the shift count is either negative or too big. Since it's clearly not negative - and even MSVC should be able to tell that - it must think it's too big.

How could it be too big? The code in question is only compiled if the size of an off_t is greater than 4. It must therefore be at least 5 bytes (but probably 8), and given 8 bits to the byte meaning a minimum of 40 bits, which is more than 32.

What is going on here?

Is your `if constexpr` inside a template? Because it won't work otherwise. — Quentin, Aug 03 '18 at 11:31
Assuming your *description* of the diagnostic is correct (and why did you not quote it?), MSVC is clearly wrong, so file a bug. — Cheers and hth. - Alf, Aug 03 '18 at 11:33
@Quentin can you tell me where in the standard it specifies that if constexpr only works when used inside a template? Because I can't find it. — Martijn Otto, Aug 03 '18 at 11:38
No, but please have a look [over there](https://stackoverflow.com/q/50051473/3233393) :) — Quentin, Aug 03 '18 at 11:46
@Quentin: `if constexpr` works in non template. The statement is discarded, (but it still requires to be a "valid" statement). — Jarod42, Aug 03 '18 at 12:55
@Jarod42 So, [in this case](https://wandbox.org/permlink/RekhKs5VScrw9R1s) is clang wrong or the fact that the statement only rises a warning (UB, I guess) is not enough to consider it "invalid"? — Bob__, Aug 03 '18 at 13:51
@Bob__: Bitshifting integers is never ill-formed. It may invoke UB, and compilers can *warn* about that. But the compiler shouldn't refuse to compile it, even if it can determine at compile time that the code will always invoke UB. — Nicol Bolas, Aug 03 '18 at 14:04

score 1 · Answer 1 · answered Jan 19 '19 at 15:26

Could it be the assignment into overlapped.OffsetHigh, and not your explicit shift, that is causing the warning? The following code generates the same warning on VS2015 compiling for x86 32-bit:

struct Clump
{
    unsigned int a : 32;
    unsigned int b : 32;
    unsigned int c : 32;
};

unsigned int x = 0;
unsigned int y = 0;
unsigned int z = 0;

Clump clump = { x, y, z }; // This is line 1121.

1>E.cpp(1121): warning C4293: '<<': shift count negative or too big, undefined behavior

But removing the bit-fields, there is no warning:

struct Clump
{
    unsigned int a;
    unsigned int b;
    unsigned int c;
};

unsigned int x = 0;
unsigned int y = 0;
unsigned int z = 0;

Clump clump = { x, y, z }; // No warning.

MSVC: shift count negative or too big

1 Answers1