[llvm-dev] Best practices for using update_*_test_checks.py

[Tom Stellard via llvm-dev]

Hi,

I've been playing around with the update_mir_test_checks.py utility lately,
and it seems like a useful tool.  However, I think the convenience of
the tool makes it easy to use it to generate checks without actually
verifying the checks are correct.

This actually happened to me when I first started using it, and I almost
sent out a patch with a bug.  I also think this may have been the culprit
with r332452 which was fixed by r332531.

I think it would be good to develop best practices for using these scripts to
help developers and reviewers to avoid these kinds of mistakes.

I personally don't think these scripts should be used at all unless the 
code has been extensively tested in some other way, like with an external
test suite, but I am not all that familiar with these scripts and their
use cases so maybe that is too extreme.  I'm curious what some of the
more frequent users of these scripts think about this.

-Tom