changelog for #2705

[rwgk]

#2705 could be seen as adding a feature: const/non-const propagation that wasn't supported before. I think a changelog is useful, but I'd be happy to drop this, up to you.

[YannickJadoul]

I'm not a great fan of advertising const-correctness in the casters until we've checked/fixed this in general. (Though I'm very much up for actually having this in pybind11!)

But it's also not wrong, so ... eh. Tending to "I wouldn't", but not strongly.

If this is considered a new feature, strictly speaking it should be 2.7.0 instead of 2.6.2, though, according to Semantic Versioning (as I tagged #2705, a handful of minutes ago).

[henryiii]

If we don't have proper tests for it and/or it's hard to use so far, though I'm normally a fan of a detailed changelog, I could go either way in this case. It's a pretty specific statement, not like it's advertising complete const correctness.

[YannickJadoul]

If we don't have proper tests for it and/or it's hard to use so far, though I'm normally a fan of a detailed changelog, I could go either way in this case. It's a pretty specific statement, not like it's advertising complete const correctness.

That's also true, yes.
If I had written it, I would be slightly more careful, phrasing it similar to the original "make std::reference_wrapper act more like an ordinary C++ reference" (or, as I said in #2705, if it's considered a bugfix, I wouldn't even bother trying to explain the change). But if @rwgk prefers this, sure, why not.

We should only be a bit careful with SemVer on whether this is a bugfix or feature, cfr. which future version of pybind11 is this part of (2.6.2 or 2.7.0).

[rwgk]

If we don't have proper tests for it

We do! I haven't tried it out myself, but it is my understanding that some of @laramiel's test additions fail without his change, and work with his change.

In any case, I agree that this isn't very important to have in the changelog. I created it assuming Henry would ask for it. I'll close this PR.

[YannickJadoul]

Up to @henryiii or @bstaletic, if they still want a changelog entry?

I'm not strongly against, but I have my reservations (as phrased above :-) ).

We do! I haven't tried it out myself, but it is my understanding that some of @laramiel's test additions fail without his change, and work with his change.

That's great to know, though! What I meant is mostly that I don't know how these kinds of const-preserving casters interfere with other built-in casters (like tuple etc), though; so that's why I'm a bit cautious about proclaiming "const-correctness".
If we start a thorough review of the casters and everything does work (not just tests, though; I'd like to see this fit in to the global architecture), that would be really nice, though! If it's easy enough, I could even envision some future (pybind11 2.8?) built-in functionality that will preserve const?