[functional/hash] Reason for failure of hash_float_test on Tru64/CXX
As can be seen at http://tinyurl.com/9lded there is a failure for a very specific test of the hash function on Tru64/cxx65-042. I finally took the time to track this down and it turned out to be due to a bogus return value of std::numeric_limits<long double>::denorm_min(). The good news is that it's fixed in the latest compiler release. But should the failure be marked as expected for older versions of the compiler, or should a workaround be added to the test itself? Any preferences? Markus
Markus Schöpflin wrote:
As can be seen at http://tinyurl.com/9lded there is a failure for a very specific test of the hash function on Tru64/cxx65-042.
I finally took the time to track this down and it turned out to be due to a bogus return value of std::numeric_limits<long double>::denorm_min().
The good news is that it's fixed in the latest compiler release. But should the failure be marked as expected for older versions of the compiler, or should a workaround be added to the test itself? Any preferences?
Thanks for looking into this. I think I'd go for the workaround - as the library itself isn't failing. Can you let me know the version number which works? Or feel free to make the change yourself, if you prefer. Daniel
Daniel James wrote:
Markus Schöpflin wrote:
[snip]
The good news is that it's fixed in the latest compiler release. But should the failure be marked as expected for older versions of the compiler, or should a workaround be added to the test itself? Any preferences?
Thanks for looking into this. I think I'd go for the workaround - as the library itself isn't failing. Can you let me know the version number which works? Or feel free to make the change yourself, if you prefer.
I checked in a workaround for the test on the RC branch. Markus
participants (2)
-
Daniel James -
Markus Schöpflin