Walter knowingly, silently unfixed a bug

[ag0aep6g]

I filed issue 17635 [1] in 2017. A low-quality fix was merged later that 
year.

In 2018, Walter moved the test case for issue 17635 from /compilable/ to 
/fail_compilation/ [2]. I.e., he flipped it upside down. He provided no 
explanation. He did not update the issue in Bugzilla. I only found out 
about it now, because I'm working on a related issue and that test 
failed when I tried my fix.

The pull request was approved by Jacob Carlborg. He also had nothing to 
say about the flipping of the test case. But the branch was force-pushed 
eight times after the approval, so it's possible that the problematic 
change wasn't part of the version that Jacob reviewed. In which case, 
the approval should have been considered outdated.

If I were unreasonably amicable, I might say that Walter simply forgot 
to go back to Bugzilla and update the issue. But even then he should 
have mentioned in the pull request that it affects an older issue. And 
having dealt with Walter before, I'm not willing to find excuses 
anymore. I'd sooner believe that he snuck that change past Jacob with a 
force-push.

Changes to pre-existing tests must be done very carefully. When a pull 
request replaces a test with its exact opposite, that should be a big 
red flag that must be well justified.

Sadly, I can't be surprised that this happened. But I am pissed.


[1] https://issues.dlang.org/show_bug.cgi?id=17635
[2] https://github.com/dlang/dmd/pull/8048