Safer casting from double to long in value.hpp

[kevinbackhouse]

Fixes: #1830

I think it's easiest to require the value to be in the INT_MIN to INT_MAX range. It means that we're not using the full range of long on platforms with 64-bit long, but it means that the cast won't produce different results on platforms where long is 32 bits.

I have been trying to avoid changing the header files on 0.27-maintenance, but I can't see another solution for this bug. Is this change going to cause any problems?

[hassec]

Approving this, since I also don't know a better way to fix this without introducing these specializations in the header.

👍

[Floessie]

@kevinbackhouse Hi Kev, is there a reason (except for brevity) to use INT_(MIN|MAX) instead of the more C++ish std::numeric_limits<int>::(min|max)()?

[kevinbackhouse]

@Floessie: I think brevity was my only reason. To be honest though, I was also just trying to fix the UBSAN issue in the simplest way possible, without changing existing behavior. Limiting the range to INT_(MIN|MAX) is a quick fix, not a proper solution. On the main branch, I would like to investigate why we are casting from double to long, and whether we can restrict it to much tighter bounds. For example, are we using this to calculate a percentage, where the expected range is 0..100?