Issue 4391: rustc should not silently skip tests with erroneous signatur...

[pnkfelix]

...e.

Fixes #4391.

[pnkfelix]

This is getting really tiresome. My local make check runs are not encountering any of these; but they do terminate unsuccessfully due to

failures:
    [run-pass] /Users/fklock/Dev/Mozilla/rust.git/src/test/run-pass/too-much-recursion.rs

I don't want to keep asking bors to do this for me, but make check is not showing me these problems locally. I have some ideas for work-arounds to try to iterate through the testing space locally, but I am clueless as to what to do about the heart of the problem (i.e. that my make check run is not a faithful representative of bor's make check run).

[pnkfelix]

(I guess when I rebased the patch series, I lost @catamorphism original review approval. This is for #4391 .)

[pnkfelix]

Note also that when I lost @catamorphism 's original review approval, I also lost his suggestion that the code switch from using sess.span_fatal to using sess.span_error so that we could accumulate the list of violations rather than dying immediately. I ended up adopting this proposal for my own sanity's sake as I was working through the series of pre-existing violations in the test source tree (I hadn't noticed them before that due to make check failing early on in my local testing earlier).