As you can likely see, I’ve freshened up the site a bit. The versions of Jekyll and Jekyll Bootstrap I had been using since 2013 were a bit long in the tooth, so it was time for a clean(ish) start. I won’t bore you with the details, but I hope you find it more pleasant. If you notice any problems, feel free to file an issue.

Back in January I announced version 1.5 of UnitTest++, and with that the adoption of Semantic Versioning. Since then, there have been three more releases, and while this falls a bit short of the “once per month” rate I had planned, I’m happy that we’ve had four releases in 2016, soon to be five. This fifth release, however, introduces our first potentially breaking changes, so I wanted to post a bit more about this in advance – especially since the breakage is rather subtle.

CHECK 1, 2

The CHECK macro is the most basic of checks, asserting simple “truthiness” of the provided expression. Ever since the migration to GitHub, however, there has been an open issue to change the signature of this method from pass-by-value to pass-by-reference. While the obvious impetus – avoiding unnecessary copies – has always been there, this was not a strong motivation for change. However, a recent issue pointed out the real problem with this – copies with side-effects can trigger an unexpected failure.

Now, I for one do my best to avoid side-effects in copy constructors, and in the case of this particular code the object probably shouldn’t even be copyable. However, a test library should not be making assumptions of, or attempting to enforce particular practices in, the code under test. Rather, this was an excellent driver to finally make the breaking change.

The simplest example I could come up with to expose the issue looks like this:

    class TruthyUnlessCopied
       : truthy_(true) { }

       TruthyUnlessCopied(const TruthyUnlessCopied&)
       : truthy_(false) { }

       operator bool() const { return truthy_; }

       bool truthy_;

       TruthyUnlessCopied objectThatShouldBeTruthy;


The prior implementation of CHECK would fail here because the default-constructed, truthy object passed to UnitTest::Check would be copied; the new implementation does not exhibit this behavior. However, if anybody out there has tests that are reliant on this behavior – intentionally or not – this could cause failures where none existed.

The changes are available in master for you to test against today.

Updates to macro names

Pull 114 attempts to introduce some consistency and flexibility to the macro names that make up the majority of the UnitTest++ public interface. However, since it was never clearly stated which macros make up the public interface, I consider this a breaking change – although it should not break most users.

Starting with version 2.0, all macros that are considered part of the public interface are of the form UNITTEST_<SHORTNAME>, and those that are not are of the form UNITTEST_IMPL_<NAME>. The <SHORTNAME> forms of the public interface of course remain, but they can now be shut off to hopefully avoid name collisions with other macros in your project. See the pull request for full details of the changes.

These changes are also available in master for you to test against.

~-- pj --~