extension loading is not tested

[Vincent Ladeuil]

Despite bugs #430749 and #406113 there is still a hole:

    >> I'm actually thinking that we should have tests that the extensions are
    >> loaded and *those tests* should fail if they are not.

Which makes life pretty hard for people wanting to run the full
test suite in both configurations :)

Since we want the ability to test:
- with extensions loaded,
- without extensions loaded (not even existing !),
- that the extensions are loaded when they exist

I think we need a test for each extension, all of them depending
on -Echeck-extension-loading which is on by default.

That way we caught early all cases where the extensions are not
built and people who want to run the test suite without these
tests (including us to ensure we support the python fallbacks)
still can.

Thoughts ?

[Vincent Ladeuil (vila)]

The tests can also verify that the extensions are up-to-date to catch people who forget to recompile them...

In the long term we may want to define different test suites but I think using the -E flags is enough in the short term.

[Martin Pool (mbp)]

This is kind of contemplated by the MissingFeature stuff and I'd like
to see it done on top of that.

We intended to distinguish "couldn't test on this platform" from
"could test it if you install (or build) more stuff", and I think
missing extensions should be under the second case, then we should run
tests on those machine insisting on nothing missing.

--
Martin <http://launchpad.net/~mbp/>