On 19 December 2012 22:15, Chris Muller <ma.chris.m at gmail.com> wrote:
>>> Why do you say that's breaking the tests?  If you "Do It" on one of
>>> the failed MC tests it opens the debugger showing the failed assertion
>>> even though initials have already been set.
>>
>> If the author initials aren't set, you get a FITBM asking for it. The
>> proper cleanup and restoration of MC's test state will thus not
>> happen. In particular, MCMockClassA does not get its #one method back,
>> which later tests expect.
>>
>>> These tests weren't failing in 4.3 anyone know what happened?
>>
>> People usually don't run tests without author initials? Certainly
>> these tests have ALWAYS passed on CI, but CI sets the author initials.
>>
>> I have yet to see evidence to contradict my hypothesis.
>
> Eh, well I just opened the 4.3 release image and ran tests.  It asked
> for my initials at the beginning and then ran all tests.  Those MC
> tests didn't fail.  I don't think we can release without understanding
> what's going on with them in 4.4.

Yes, but did you _fill in_ the initials?

At any rate, theories aside, it _is_ the case that the
MCMethodDefinitionTest >> #tearDown doesn't restore state correctly.
Note that this is #testLoadAndUnload test is the test that pops up the
prompt. Subsequent tests fail because, for at least some of them,
MCMockClassA >> #one no longer exists. (MCMethodDefinitionTest >>
#testLoadAndUnload removes it.)

> Would someone volunteer?  I'm currently looking at Dave's SqueakMap
> patch is needed now but not before and making sure we can, in fact,
> deploy packages for 4.4 that will show up in SqueakMaps 4.4 list.

Yes, please. I would like some countervailing evidence and/or
alternate hypotheses.

Nevertheless, these tests definitely run on CI, and definitely pass.
Have always passed, in fact. It was only glenpaling's recent findings
that brought these tests to my/our attention.

frank