boost::chrono::duration default constructor doesn't initialize rep_

[florent.castelli@…]

According to the documentation:

constexpr duration();
Effects: Constructs an object of type duration from duration_values<rep>::zero().

The implementation is:

#if  defined   BOOST_NO_CXX11_DEFAULTED_FUNCTIONS
        BOOST_FORCEINLINE BOOST_CONSTEXPR
        duration() : rep_(duration_values<rep>::zero()) { }
#else
        BOOST_CONSTEXPR duration() BOOST_NOEXCEPT {};
#endif

So in one side is does initialise to zero, not in the other side. This is a breaking change triggering tons of MSan issues in my application (and certainly bugs in real applications). Setting rep_ to zero() as it is done above fixes the issue for me.

[viboes]

Hi,

you are right that there is an inconsistency I need to document.

The correct behavior is, following the standard, to don't initialize it. I received a request to be able to initialize it a zero.

I'm really sorry, but I would not initialize it in c++11. I could add a compiler flag to force the initialization.

#if  defined   BOOST_NO_CXX11_DEFAULTED_FUNCTIONS || \
     defined   BOOST_CHRONO_DURATION_DEFAULTS_TO_ZERO
        BOOST_FORCEINLINE BOOST_CONSTEXPR
        duration() : rep_(duration_values<rep>::zero()) { }
#else
        BOOST_CONSTEXPR duration() BOOST_NOEXCEPT {};
#endif

[florent.castelli@…]

I guess that's fair as long as it is documented as a breaking change and the documentation doesn't claim it is initialized.

I was thinking of possibly migrating the code to std::chrono so I'll have to make sure everything is properly initialized anyway. In the mean time, I guess having a local patch to initialize it to zero so I can migrate the rest of my code to 1.58.0 is fine.

[viboes]

So you agree for been forced to define BOOST_CHRONO_DURATION_DEFAULTS_TO_ZERO?

[florent.castelli@…]

Yes

[florent.castelli@…]

Yes

[viboes]

I was on vacation. I will do it asap.

[viboes]

​https://github.com/boostorg/chrono/commit/6c7ff6678eea57e1e4e22c29ae9a7bfc023826ff