Just as for builds, if test-runs routinely have warnings in them, no-one will notice when their change causes a new warning, to which they should be paying attention. So we should aim to make test-runs warning-free.

Fortunately QtTestLib provides the tools to

• fail a test if it produces warnings or
• filter out expected messages when a test should produce them, and fail if it doesn't.

We should aim, over time, to clear all our tests of warnings (and preferably other messages) by marking those that should be seen as expected. We would ideally arrange for a Coin Precheck to report any new warnings, even on success; and potentially have Coin reject changes if they add (perhaps: more than some threshold level of) new warnings.

Prompted by a request on the review of making duplicate test-data tags trigger warnings.