Compare collections of floating point values with tolerance

[Flamefire]

Currently you can use either:

• BOOST_TEST_REQUIRE( output == outputDesired, boost::test_tools::per_element() );
• BOOST_TEST_REQUIRE( output == outputDesired, boost::test_tools::tolerance( 1e-12f ) );

But you cannot use both.

What is the preferred way to compare collections (e.g. vectors) of floating point values?

This is a common use case for testing numeric algorithms: Check results of optimized version vs. naive implementation.

[raffienficiaud]

What about using per-unit tolerance with a decorator?
https://www.boost.org/doc/libs/1_68_0/libs/test/doc/html/boost_test/utf_reference/testing_tool_ref/decorator_tolerance.html

[Flamefire]

This is our current work-around but limits the abilities. Imagine a test where you test matrix operations and first have to assert that some intermediate value (e.g. input, transposed matrix, ...) is exactly equal and then test the result of some calculation. Yes those can be put into separate unit tests, but maybe they are so closely related that one does not want to do this.

[raffienficiaud]

@Flamefire there is a WIP branch for this: topic/PR-187-BOOST_TEST-with-tolerance-message: it would be nice if you can give a try until I polish it up. It should address this particular issue together with #242 and #187.

I'll try to make that for 1.73.

[raffienficiaud]

Merged to master.

[Char-Aznable]

@raffienficiaud I struggle to find documentation on the OP's proposal. What exactly is the solution to this problem that get merged?

[raffienficiaud]

@Char-Aznable sorry for the late answer: now it is possible to combine both boost::test_tools::per_element() and boost::test_tools::tolerance( 1e-12f ). The documentation is only in the change log of 1.73.

If we need more, let me know I'll open a ticket.