Modernize ranges::move

[miscco]

Note this is already written forward looking to what comes with the other PRs namely #959 and #930, so it will only compile after they have been merged. That said I plan to squeeze one of those into my lunch break and rebasing them should be faster than writing them some time in the future.

[miscco]

So this is really strange. For whatever reason it tries to copy the int_wrapper class rather than moving it.

Any lead on why?

[CaseyCarter]

So this is really strange. For whatever reason it tries to copy the int_wrapper class rather than moving it.

Any lead on why?

[miscco]

So it seems that the test machinery requires the elements to be copy constructible and copy assignable.

Is that intentional?

[StephanTLavavej]

If you want to trigger a test run, you can push commits changing and undoing some comment; you might also be able to intentionally form an empty commit and push that (I know those can be created and pushed; not sure if Azure Pipelines will trigger based on them).

[CaseyCarter]

So it seems that the test machinery requires the elements to be copy constructible and copy assignable.

Is that intentional?

Nope: it's a bug in test::proxy_reference. We need test::proxy_reference<Cat, Elem> and T to model std::common_reference_with whenever Elem& and T do. Fixing it causes many of the tests to fail in MSVC's permissive mode, so they'll effectively need to be constrained to only test ranges with non-proxy-reference-types when compiling in permissive mode.

Biggish patch incoming, which I assume will conflict with many of the "Modernize meow" PRs.

EDIT: I've managed to greatly limit the impact of the changes after reverting a change that made proxy_reference's explicit constructor implicit. I made that change investigating a different formulation that tried to make proxy_reference itself the common_reference of proxy_reference and T before I landed on creating a new type. This may not be as bad/impactful/destabilizing to the ongoing work as I feared.

[miscco]

@CaseyCarter thanks a lot for the fix, this was way over my head. Am I correct that this also solves the problems with the meow_copy algorithms that seem to fail for strange reasons?

I will remove the copy costructor/assignment from int_wrapper to be sure

[CaseyCarter]

@CaseyCarter thanks a lot for the fix, this was way over my head. Am I correct that this also solves the problems with the meow_copy algorithms that seem to fail for strange reasons?

I don't know, but I hope so ;)

I will remove the copy constructor/assignment from int_wrapper to be sure

I did so in this PR already, IIRC.

FWIW, there are still a couple (or I should say were a couple when the test runs weren't all failing spuriously) of test failures that repro on Azure but not on my local machine - possibly some issue that was fixed in 16.7p3. I'm not sure if we should try to localize and workaround or just update the test bots to the newer Preview since they're not exactly healthy right now.

[StephanTLavavej]

Seeing all these cleanups truly moves me to tears. 😹