If I add in the unit tests that we have for another product we're shipping with the current project, we have well over 5000 tests. That, if anything, should inspire confidence. However, this isn't a Web app that I'm building. This is an app that's going to be installed on Linux servers and shipped all over the country. I can't control what happens when it gets to the client site. I can't stop them from playing with the code or using it some way that I never anticipated, much less tested for. I have a vision of hundreds of clients calling us repeatedly for some little bug that we all missed.
It would be pretty easy to write a test that checked MD5 sums of your distributed files. Then to get diagnostic output, you just ask people to run the entire test suite, dumping the results to a file. If that test fails, you know they've messed with things.
Re:t/000-no-touchie.t
inkdroid on 2003-06-03T17:44:28
Damn, that's a cool idea.Re:t/000-no-touchie.t
dlc on 2003-06-03T19:54:58
Isn't this what part of what Test::Signature does? When tests start failing, and one of the failing tests is t/00signature.t, you know they started touching things.
Re:t/000-no-touchie.t
chromatic on 2003-06-03T20:16:24
True, but it's such a pain to install I'm not sure about asking a customer to install it. Math::Pari is the biggest culprit I've found.
Re:t/000-no-touchie.t
koschei on 2003-06-03T23:32:47
I guess that's one use for it. Never thought of that.
It won't help if they modify the installed versions of files though.Re:t/000-no-touchie.t
dws on 2003-06-04T04:51:26
We once considered shipping a shadow copy of our sources, along with a diagnositic tool to generate a diff, zip it up, and email it to us.
