flat_hash_map fails to compile when 'Warnings as Errors' is turned on in VS2017

[ChewyGumball]

When I include absl/container/flat_hash_set.h or absl/container/flat_hash_map.h I get the following warnings:

absl\container\internal\raw_hash_set.h(206): warning C4244: 'argument': conversion from 'T' to 'uint32_t', possible loss of data
        with
        [
            T=uint64_t
        ]
absl\container\internal\raw_hash_set.h(431): note: see reference to function template instantiation 'int absl::container_internal::TrailingZeros<uint64_t>(T)' being compiled
        with
        [
            T=uint64_t
        ]
absl\container\internal\raw_hash_set.h(212): warning C4267: 'argument': conversion from 'size_t' to 'uint32_t', possible loss of data
absl\container\internal\raw_hash_set.h(480): note: see reference to function template instantiation 'int absl::container_internal::LeadingZeros<size_t>(T)' being compiled
        with
        [
            T=size_t
        ]

Looking at the TrailingZeros and LeadingZeros functions we see:

template <typename T>
int TrailingZeros(T x) {
  return sizeof(T) == 8 ? base_internal::CountTrailingZerosNonZero64(x)
                        : base_internal::CountTrailingZerosNonZero32(x);
}

template <typename T>
int LeadingZeros(T x) {
  return sizeof(T) == 8 ? base_internal::CountLeadingZeros64(x)
                        : base_internal::CountLeadingZeros32(x);
}

The warning is spurious because the base_internal call that won't lose data will be used, but the other is still compiled, and thus warned about.

In my case, I have turned on 'Warnings as Errors', so I can't even compile without static casting x to the type of the argument of CountLeadingZerosXX or CountTrailingZerosNonZero64.

[derekmauro]

I wasn't able to reproduce this (maybe I have a different version of MSVC?), but I added some casts that should fix this for you. Let us know if it doesn't.

[Bu11etmagnet]

This should be a compile-time decision, not a run-time decision, e.g. if constexpr (...).
Is there a constexpr ? : yet?

[derekmauro]

Since we are supporting C++11, we can't use if constexpr, but I'm pretty sure all modern compilers are smart enough to optimize away the constant branch in this case.