Opened 5 years ago
Closed 5 years ago
#13109 closed Bugs (fixed)
Incorrect result when casting from multiprecision::uint128_t to unsigned long long.
| Reported by: | Owned by: | John Maddock | |
|---|---|---|---|
| Milestone: | To Be Determined | Component: | multiprecision |
| Version: | Boost 1.63.0 | Severity: | Problem |
| Keywords: | Cc: |
Description
I found in Boost 1.63.0 on: Ubuntu 14.04.5 with g++ (Ubuntu 4.8.4-2ubuntu1~14.04.3) 4.8.4, 64-bit Windows 10's Windows Subsystem for Linux with g++ (Ubuntu 5.4.0-6ubuntu1~16.04.4) 5.4.0 20160609.
code snippet: uint128_t b128 = 0xFFFFFFFFFFFFFFFFFFFFFFFFC0A800FF_cppui; unsigned long long ull( static_cast<unsigned long long>( b128 ) ); printf( "ull: %016llX\n", ull );
The output is: ull: FFFFFFFFFFFFFFFF
Under Windows 10 and Visual Studio 2015 the output is: ull: FFFFFFFFC0A800FF
The work around I'm using is to mask out the upper 64 bits: ull = static_cast<unsigned long long>(b128 & 0x0FFFFFFFFFFFFFFFF);
Thanks, Gary Visser
Note:
See TracTickets
for help on using tickets.
The intended behaviour is to return the maximum value of the target type when a narrowing conversion overflows (I realise this is different behaviour to most built in integer types, but then you're into undefined behaviour).
This patch: https://github.com/boostorg/multiprecision/commit/b35f1c8f613db66ad80ce260f94b7027814ede71 enforces uniform behaviour across all backend integer types.
In future, if you really want just the low order N bits, then you will need to mask these out - the same is true of built in integers actually if you want to avoid code-sanitizer errors (cland usan etc).