This isn't actually a regression. I won't argue about whether or not this is a Critical bug, but it's not a regression. It's not that we've broken past functional behaviour, but rather that we've released something that doesn't quite do what it promises to do. We have always been using TestResult._exc_info_to_string from unittest for rendering tracebacks. That hides stack levels with __unittest set, and it has always hidden the unittest levels of the stack where __unittest is set. To show the *full* stack is right, but it's a new behaviour for us.
To enable this, and to fix bug 854769, I am currently leaning toward moving the implementation of _exc_info_to_string over to our code. This might also simplify some of our unicode handling.
PS. We have a facility for unhiding *our* levels of the stack (run_tests_with = FullStackRunTest). Bug 881780 points out how this could be better.
This isn't actually a regression. I won't argue about whether or not this is a Critical bug, but it's not a regression. It's not that we've broken past functional behaviour, but rather that we've released something that doesn't quite do what it promises to do. We have always been using TestResult. _exc_info_ to_string from unittest for rendering tracebacks. That hides stack levels with __unittest set, and it has always hidden the unittest levels of the stack where __unittest is set. To show the *full* stack is right, but it's a new behaviour for us.
To enable this, and to fix bug 854769, I am currently leaning toward moving the implementation of _exc_info_to_string over to our code. This might also simplify some of our unicode handling.
PS. We have a facility for unhiding *our* levels of the stack (run_tests_with = FullStackRunTest). Bug 881780 points out how this could be better.