> You may have to run the test by first flushing all the code to the
> left margin before reindenting

That would certainly make the tests better. Only testing preindented
buffers is good but insufficient. What you describe is a good way to
test non preindented buffers.

> Oh, you're referring to known indentation problems, where you want to
> have the test but have it be marked as "expected".
>
> Yes, that's currently not really supported: I try to put "FIXME" on
> the corresponding lines, but very often an incorrect indentation on
> one line impact subsequent lines as well, so not only does the
> "diff" not care about those FIXME markers, but you can get false
> positives or false negatives to boot.

Sounds more complex than using ERT :expected-result.

If I indicate :expected-result is :failed on an invocation of a
test/indent test from ERT, it would suppress all the expected
successes because of one expected failure. Alternatively, the test
could filter out lines with "FIXME", but that's ugly.

I think finer grained ERT tests that can be marked as :expected-result
of :passed or :failed (and don't need to parse for FIXMEs) are the
better approach for test/automated 'make check' tests.