re #34: reproduce in CI

[biojppm]

re #34

[lemire]

Thanks. We deliberately want C++ to treat the objects as trivial, hence the default constructor.

I have added inline initializer which should hopefully do away with the warning. Please sync your branch.

[biojppm]

@lemire The previous changes do reproduce the problem in the CI.

I think it is a genuine one: in absence of ctor initialization, this statement will return mantissa uninitialized:

[biojppm]

I see you've added a fix. Let me try it.

[biojppm]

It's curious that later gcc versions fail to point to this issue.

[lemire]

It's curious that later gcc versions fail to point to this issue.

I submit to you that it is because it is not an issue.

I think it is a genuine one: in absence of ctor initialization, this statement will return mantissa uninitialized:

I disagree. Here is the code...

  adjusted_mantissa am = pns.too_many_digits ? parse_long_mantissa<binary_format<T>>(first,last) : compute_float<binary_format<T>>(pns.exponent, pns.mantissa);
  if(am.power2 < 0) { am = parse_long_mantissa<binary_format<T>>(first,last); }

See how if am.power2 is negative, then mantissa is never used?

It is not needed that mantissa ever be initialized if it is never being read.

If the compiler is not smart enough, its analysis will end up with a false positive.

[lemire]

Great PR. Merging.

[biojppm]

Thanks for clarifying.