bug?: possible surprising results for inclusive on integer axes?

[henryiii]

Compiling the following program:

#include <iostream>
#include <boost/histogram.hpp>

namespace bh = boost::histogram;

int main() {

    bh::axis::integer<> x1(0, 4);
    std::cout << "integer default: " <<  (bh::axis::traits::inclusive(x1) ? "True" : "False") << std::endl;

    bh::axis::integer<int, bh::use_default, bh::axis::option::overflow_t> x2(0, 4);
    std::cout << "integer overflow only: " <<  (bh::axis::traits::inclusive(x2) ? "True" : "False") << std::endl;

    bh::axis::integer<int, bh::use_default, bh::axis::option::underflow_t> x3(0, 4);
    std::cout << "integer underflow only: " <<  (bh::axis::traits::inclusive(x3) ? "True" : "False") << std::endl;

    bh::axis::integer<int, bh::use_default, bh::axis::option::none_t> x4(0, 4);
    std::cout << "integer noflow: " <<  (bh::axis::traits::inclusive(x4) ? "True" : "False") << std::endl;

    bh::axis::regular<> x5(4, 0, 4);
    std::cout << "regular default: " << (bh::axis::traits::inclusive(x5) ? "True" : "False") << std::endl;

    bh::axis::regular<double, bh::use_default, bh::use_default, bh::axis::option::overflow_t> x6(4, 0, 4);
    std::cout << "regular overflow only: " << (bh::axis::traits::inclusive(x6) ? "True" : "False") << std::endl;

    bh::axis::regular<double, bh::use_default, bh::use_default, bh::axis::option::underflow_t> x7(4, 0, 4);
    std::cout << "regular underflow only: " <<  (bh::axis::traits::inclusive(x7) ? "True" : "False") << std::endl;

    bh::axis::regular<double, bh::use_default, bh::use_default, bh::axis::option::none_t> x8(4, 0, 4);
    std::cout << "regular noflow: " <<  (bh::axis::traits::inclusive(x8) ? "True" : "False") << std::endl;
}

Gives this result:

integer default: True
integer overflow only: True
integer underflow only: True
integer noflow: False
regular default: True
regular overflow only: False
regular underflow only: False
regular noflow: False

It seems integer axis with only one flow bin is still "inclusive", unlike Regular (and Variable, though not shown above). Is this expected?

PS: If you start with integer, and then try regular, sadly this compiles correctly: bh::axis::regular<double, bh::use_default, bh::axis::option::overflow_t> - but is obviously wrong - the options are in the metadata slot. Don't know if there's a way to protect or print a warning if this happens, as anything is supported for metadata.

[henryiii]

This seems to say that the default is to only be inclusive if both underflow and overflow is set:

https://github.com/boostorg/histogram/blob/ff188c61f66844ad4907d128197feb3c8d0e5fe9/include/boost/histogram/axis/traits.hpp#L52

But, when customized for Integer,

https://github.com/boostorg/histogram/blob/ff188c61f66844ad4907d128197feb3c8d0e5fe9/include/boost/histogram/axis/integer.hpp#L156-L158

Then options() & option::underflow || options() & option::overflow is true if either one is set. I think it should be options() & option::underflow && options() & option::overflow, perhaps?

[HDembinski]

Right, looks like a bug, thank you for reporting.

[HDembinski]

It looks like I did this on purpose and not by mistake, though, hm. I should have left a comment.

[HDembinski]

PS: If you start with integer, and then try regular, sadly this compiles correctly: bh::axis::regular<double, bh::use_default, bh::axis::option::overflow_t> - but is obviously wrong - the options are in the metadata slot. Don't know if there's a way to protect or print a warning if this happens, as anything is supported for metadata.

Yes, a safer order would have been value_type, metadata, options, transform for the template arguments, but I chose to put the transform as a second argument instead, which I thought is more user-friendly. I will try to add some static_assert to check the passed types. This will be much nicer with concepts in the future.

Edit: On closer inspection, I think there is nothing I can do here to avoid this. Since metadata is allowed to be anything, I cannot really catch this.

[HDembinski]

@henryiii Did this bug affect you downstream?