improving the quality of the image

Brad Fuller brad at bradfuller.com
Mon Jan 29 15:25:10 UTC 2007


Ralph Johnson wrote:
> Failing tests do not belong in the image.  They belong in Mantis.  If
> you have a test that ought to run but does not then post it to Mantis.
> 
> We should not release an image with failing tests.  In general, the
> 3.10 release team will not accept changes that break tests.  If the
> image contains failing tests then it is hard for people to tell that
> their proposed changed breaks them.  It is crucial that there be NO
> failing tests in the image so that anyone can easily check whether
> their change caused any of the tests to fail.  Go to the test runner,
> select all tests, run them, and make sure everything is green.
> 
> There is no need to make some new way of classifying tests.  The way
> we classify tests is that tests that do not work are in Mantis as bug
> reports.  If you want to work on them then you can easily file them
> in.

Is there not a difference between an image under development and a
released image? I think so. A released image should not have failing
tests but it should also not contain the code that the test failed on.
So, defect #5527 should read:

"Some of the tests in the final released 3.9 image fail. You should be
able to run all tests in a *RELEASED* image and get a green bar in the
TestRunner."

Also, what do you mean by this:

"... and I deleted some of them and moved the rest of them to subclasses
in a category FailingTests, which I then deleted. I have attached a
change file that makes all the tests green in the final released image."

Did you delete the tests AND the code that it tests? Just removing the
tests doesn't really help much if the code that it tests is still there.
I'm sure you agree with this, so that is why I'm a bit stumped.

-- 
brad fuller
www.bradfuller.com



More information about the Squeak-dev mailing list