Bug 4747 – Make each unittest block a separate function that runs independently of the others

At the moment, each module has a single unittest function made up of all of the unittest blocks in the module. This causes 2 problems: 1. A failure in one unittest block results in none of the others getting run, even though they are independent in the code. 2. Stack traces from unit tests are useless because you have no way of knowing _which_ unittest block failed. By making each unittest block its own function, they can be run independently. It also will help with stack traces, because each one would have its own name - even it differs only by a number (since presumably, that's how unittest blocks at the same scope would be given names by the compiler). At least then you could count the unittest blocks to find it. Right now, there's no way to know. Now, if you add that with named unit tests, e.g. unittest (mytest) { } then you have a function name for the unit test and the stack trace is properly specific (though presumably unnamed unit tests would still use a numbered naming scheme of some kind). I know that an enhancement request which includes named unit tests for D1 was reported a while back ( bug 2749 ) - though it also includes stuff like nesting unit tests (which I think is pointless) - and I know that it has been previously discussed that unit test blocks should be independent, but I'm unaware of a bug report on it. So, here's an enhancement request to make unittest blocks independent, and as part of that, I would _strongly_ encourage that named unit tests be added as well. That way stack traces are clear and useful. Right now, if you want to do that, you have to create a function that contains your unit test and have the unittest block call it, which is certainly cumbersome, and the potential benefits of named unit tests go far beyond stack traces, so we really should have them at some point. Both would help a lot with unit tests.

I'm not sure but isn't your second point more related to stack traces? I mean if the stack trace would provide proper line numbers it was easy to find out which unittest failed. Further this only applies if all you have is a stack trace. That means if something was thrown you get the line number information. If there is a segfault or similar (division by zero) you currently have to run it under a debugger. I think this should be improved by providing better stack traces. Anyway the first point remains valid. unittests are often independent and it should be possible to execute them independently.

It has to do with naming unittest blacks, which is related both to stack traces and to calling unittest blocks independently. Recently, it _was_ fixed so that the functions generated for unittest blocks are named after the line number (and possibly file? - I don't know the exact naming scheme), which reduces the problem with regards to stack traces, but if you wanted to be able to call specific unittest blocks (say from a unit testing tool), then having actual names for each unittest block makes that work much better. The main thing though is to fix the issue where it's all or nothing for running a module's unit tests. When I created this request, I was under the incorrect understanding that there was one function for all of a module's unittest blocks, which isn't true, but it _is_ true that they can't be run independently at present, and if I understand correctly, some dmd changes are required to fix that.

But since unittests now have the line number in its name this issue is only concerned having the ability to call individual unittests. Right? Because now you can tell from the stack trace which unittest failed.

It makes it so that having named unittest blocks is less critical for stack traces, but having actual names for them would still me quite a bit better (it would also help for code organization, because it could document what the test was for). Regardless, the main point of this issue was always that the tests be indivdually callable. Names for them just makes that a _lot_ cleaner, but they'd have to be optional names regardless, because anything else would break tons and tons of code.

Pull request: https://github.com/D-Programming-Language/dmd/pull/1131 (Allows to run unit tests individually. Also prepares name support, but full support needs changes in lexer/parser)

Has this been resolved?

Yes this has been resolved.