bpo-16510: Using more specific assertions in miscellaneous tests by serhiy-storchaka · Pull Request #792 · python/cpython

I still contend that most of these should not be done. Unless the test is failing, there is no reason to suspect that a different error message is needed (i.e. it is unlikely that we will ever see a benefit from even a tiny fraction of this huge number of changes). When we make changes to the code, we have our tests as a safety net, but when making wholesale changes to the test suite itself, there is no safety net (i.e. if you make a mistake, we silently lose the value of the test). The changes are not being made as Guido suggested (he has a strong preference for holistic refactoring where the work is done while focusing on the needs and use cases of a particular module with full understanding of the ramifications of the changes in that module, rather than sweeping broadly changing many dozens of files, thinking only of the desire to use specific asserts rather than studying one module as time, focusing on its needs). In at least some of the cases, the new error messages add no more information than was done before. I like that simple assertTrue cases have a virtue in making it exactly clear what code is being exercised, which stands in contrast to some of the "specific asserts" do their tests indirectly. This is especially problematic for the standard library where we are testing the language itself. For example, in testing sets, assertTrue(a < b) is trying to actually exercise the "<" operator. Replacing this with assertLessThan(a, b) obfuscates the test and make me wonder whether someone got creative with unittest and was calling "b > a" or "not a >= b" or some such. As the writer of the test and as a reader, I want to know for certain that the operator was exercised (in addition "<" doesn't even mean "less than" for sets). Another issue, is that I have written many tests in a test-driven-development style. I have made a failing test and then made it pass. That makes the test sacred. Once your refactoring is done, I would have far less confidence that the test is working as intended. Likewise, use of "specific asserts" is a stylistic choice and it is unreasonable to inflict your choice on everyone else, altering the tests someone else wrote in a way that made sense to them. Personally, I find some of the specific asserts to be jarring and don't align well with what I was thinking when I wrote the tests. Lastly, if you decide to ignore me yet again, please don't backport these changes. There is no argument to be made that these are bug fixes; in fact, they just destabilize existing releases, creating a risk of introducing a buggy test.