Merge for 1.85

[mborland]

@raffienficiaud Any idea how to fix the one new test case failure here: https://github.com/boostorg/test/actions/runs/8076642786/job/22065447368?pr=410#step:8:2024. The build starts failing once this commit is applied to boost::function: https://github.com/boostorg/function/commit/af8e66d03b4366c7039a138306470567c2243130. There's only 4 uses of safe_bool in the library, but I am not entirely sure how they are relevant.

CC: @pdimov

[raffienficiaud]

Hi @mborland . From what I quickly see, it comes from statements like this and this basically checks that the underlying boost::function is initalized. From the changes in boost::function you are referring to, now the cast to bool is explicit, which might explain the error.

I can give you more details tonight.

[pdimov]

The error is

./boost/test/utils/class_properties.hpp:86:15: error: no match for 'operator<<' (operand types are 'std::basic_ostream<char>' and 'const boost::function<void()>')
     return os << p.get();
               ^

which tries to output a boost::function<void> to an ostream.

I don't think this is supposed to happen. It probably worked by chance before by implicitly casting the function to bool and outputting that, but I don't think this was intentional.

[raffienficiaud]

@mborland replacing that faulty line by this seems to work

BOOST_TEST( static_cast<bool>(tc1->p_test_func.get()) );

the diff:

raffi@raffi-macpro test % git diff
diff --git a/test/test-organization-ts/test-tree-management-test.cpp b/test/test-organization-ts/test-tree-management-test.cpp
index 97a22804..37947c9c 100644
--- a/test/test-organization-ts/test-tree-management-test.cpp
+++ b/test/test-organization-ts/test-tree-management-test.cpp
@@ -115,7 +115,7 @@ BOOST_AUTO_TEST_CASE( manual_test_case_creation_test )
     BOOST_TEST( tc1->p_expected_failures == 0U );
     BOOST_TEST( tc1->p_timeout == 0U );
     BOOST_TEST( tc1->p_name == const_string( "empty_" ) );
-    BOOST_TEST( tc1->p_test_func );
+    BOOST_TEST( static_cast<bool>(tc1->p_test_func.get()) );
     BOOST_TEST( tc1->p_default_status == test_unit::RS_INHERIT );
     BOOST_TEST( tc1->p_run_status == test_unit::RS_INVALID );
     BOOST_TEST( !tc1->is_enabled() );

I am actually not seeing any other error on develop. Do you want me to push a fix?

[pdimov]

But that's not the right fix. This BOOST_TEST should work (as it should for any type that has an explicit operator bool and does not support os << x.)

The question is why it tries to print using operator<<, when the type (boost::function in this case) doesn't support it.

[pdimov]

Or maybe BOOST_TEST always requires the values to be printable? I looked at the documentation but can't find this stated one way or another.

I tried

#define BOOST_TEST_MODULE something
#include <boost/test/included/unit_test.hpp>

struct X
{
    explicit operator bool () const noexcept
    {
        return true;
    }
};

BOOST_AUTO_TEST_CASE(something)
{
    X x;
    BOOST_TEST( x );
}

and it causes a static assertion

1>c:\boost-git\develop\boost\test\tools\detail\print_helper.hpp(53,13): error C2338: static_assert failed: 'Type has to implement operator<< to be printable'

so maybe having operator<< is a requirement?

[raffienficiaud]

The question is why it tries to print using operator<<, when the type (boost::function in this case) doesn't support it.

you gave an answer to your question: the underlying type has to be printable.

Note that here, tc1->p_test_func is not the boost::function<void ()> itself, but a wrapper readonly_property which derives from class_property. The class_property::get() in the proposed fix gives access to the underlying boost::function, the cast to bool operates on the boost::function<void ()>.

class_property itself delegates the operator<< to the underlying type here.

so maybe having operator<< is a requirement?

There are other ways (see link above) but this is usually a pain point and should be improved.