>>>>> On Tue, 23 Feb 2010 01:02:00 -0800, Michael G Schwern <schw...@pobox.com> >>>>> said:
>> None of the above is a reason to tell people to drop a test. While I >> don't know what the bug is that the OP asked about it is no solution to >> suggest dropping a test. Next time you suggest to drop all tests, right? >> Oh, yes, maybe this is the solution to all testing problems. > Don't get so excited. Reread what's being discussed. This is a > "test" that's running a check of the SIGNATURE file. I'm sorry, I always get excited when people are told to drop a valid, interesting, and valuable test. > The signature test isn't really a test. Its not testing that the code > does its job, its testing that it passes its signature. Its not a > functionality test, its a security measure, and doesn't really belong > in the test suite. At best its an author check of the integrity of > the distribution not to be distributed. I can only disagree. It's certainly of value to know that integrity checks work somewhere else and not only on author's home box before packaging. > But even that's of low value, > because its just checking that a 3rd party tool did its job... but its > doing it wrong and creating a false bug. I'm not sure if I can follow you here. False bugs need fixing just as real bugs. If a test doesn't test what it wants to test it is worth fixing. > Don't get dogmatic about testing. Tests exist to catch bugs. So shall it be. > If a > test isn't doing that then at best its dead weight. At worst its a > false failure and a maintenance hassle. You only have so much time to > spend working on tests, choose where you're going to spend it > wisely. We're arguing on same ground: testing whether the signature can be verified saves a lot of time in the case that somewhere else the signature cannot be verified. It's good to catch such a bug better sooner than later. > BTW Eric, I didn't notice that Module::Signature was complaining about > MYMETA.yml not being in the MANIFEST. I puzzled out why its doing > that. Module::Signature appears to assume that anything in the > directory at the point of "build test" is going to be either A) in the > MANIFEST or B) refuted by the MANIFEST.SKIP. Kind of a dodgy > assumption, given the build step throws files all over, but it totally > breaks down if you don't ship a MANIFEST.SKIP and your code is doing > something not expected by the default MANIFEST.SKIP... like making a > MYMETA.yml file. Thank you for digging. I'm looking forward to the bug report that comes out of it. -- andreas