What is the point of machine-checked annotations in testcases when you have to capture all error output for the
ui tests anyway?
(I'm asking because getting the exact compiler output right for a testcase that cannot yet pass is much more involved than simply saying "I'm expecting this error here")
If you pass
--bless then it updates all the ui test output. It's easy to do this and miss that you've accidentally blessed a regression (a missing error, say) if there are a lot of changed tests. By also using annotations, you need to acknowledge that a test lost an error.
Thanks, that makes sense. So I guess the reason you can't just have an
annotations-cover-all-expected-error-output header command is to make sure one doesn't miss subtle changes in the formatting of the output and instead needs to explicitly
Exactly, without stderr files we’d miss small things like a span changing.