C++: Add getTemplateClass to DeductionGuide

[jketema]

Drive-by fix in this PR: give an older change note a better name.

[sashabu]

Why aren't we finding this? It's the one that should be selected by new C(0.0);, so it's not even unused, which would have been my best guess.

Are we missing the deduction guide itself, or its template class? I.e. can we add a separate query that just does from DeductionGuide d select d?

[jketema]

We are finding it, but the frontend doesn't assign a proper location to it. This is the following line in the expected test results:

| file://:0:0:0:0 | C | test.cpp:4:8:4:8 | C<T> |

[sashabu]

I was assuming that's the copy deduction candidate (which indeed wouldn't have a location)?

In any case, an additional fictional function template derived as above from a hypothetical constructor C(C) is added, called the copy deduction candidate.

[sashabu]

(I guess an easy way to check would be to remove C(const double) -> C<int>; and see if | file://:0:0:0:0 | C | test.cpp:4:8:4:8 | C<T> | is still there.

[jketema]

The copy deduction one will be the unnamed one, which has the location of the class definition.

[sashabu]

🤔 But the unnamed one does have a location, and it's on line 5, where the C(const T); constructor is?

Basically, we have 4 deduction guides here and only 3 results lines (copy deduction guide, the guide for the C(const T); constructor, and the two explicit guides).

[jketema]

Then the copy one is likely missing. Can we side-step the discussion about missing ones for now, and open an internal issue for this? This is pretty much out-of-scope for this PR.

[sashabu]

👍 I agree based on the new test results. Could you please report:

1. The missing locations for non-template deduction guides
2. The missing copy deduction guide?

[sashabu]

I'd like to see some more tests:

1. If we don't use CTAD, are (implicit) deduction guides still emitted?
2. If we have multiple implicit guides, are they all emitted, or only the ones that are used?
3. If we have multiple explicit guides, are they all emitted, or only the ones that are used?

For 2+3, I think we can just extend the existing test to add an unused two-argument constructor and an unused two-argument guide. 1 would need to be a separate test.

[jketema]

1. If we don't use CTAD, are (implicit) deduction guides still emitted?

I don't believe so, because that would have shown up in many other tests already, and it generally doesn't. Note that this is not related to what I'm doing here, as I'm only adding a relation between deduction guides and the class templates. Not adding more deduction guides to the database.

2. If we have multiple implicit guides, are they all emitted, or only the ones that are used?
3. If we have multiple explicit guides, are they all emitted, or only the ones that are used?

Would you be able to spell out some examples? That's probably the fastest way to get this done.

[sashabu]

Take the test added in this PR and add:
2. A constructor C(int, int)
3. A guide C(char, char) -> C<char>

[jketema]

Added 2 & 3.

[sashabu]

Can we change this to:

from DeductionGuide d, string tc
where if exists(d.getTemplateClass()) then tc = d.getTemplateClass().toString() else tc = "<none>"
select d, tc

I think that would help clarify whether various deduction guides are themselves missing, or they don't have a template class associated.

[jketema]

Added a second query instead of changing the existing one.

[sashabu]

👍 I agree based on the new test results. Could you please report:

1. The missing locations for non-template deduction guides
2. The missing copy deduction guide?