force_opt combined with bind_opt results in compilation error

[henri-v]

Creating a lens with force_opt as the innermost operation and then lifting this lens using bind_opt causes a compilation error. For example, adding the following code in test/type_erased_lens.cpp will not compile:

TEST_CASE("lenses, force_opt, type erasure, bind_opt")
{
    struct Foo { int i; };
    //const lager::lens<Foo, std::optional<int>> opt_member_lens =
    const auto opt_member_lens =
        attr(&Foo::i) | force_opt;
    const auto bound_opt_member = bind_opt(opt_member_lens);
    CHECK(view(bound_opt_member, std::optional{Foo{3}}) == std::optional{3});
    CHECK(view(bound_opt_member, std::optional<Foo>{}) == std::nullopt);
}

Error message from GCC 11.4.0

By turning the force_opt lens into a type-erased lens, the code will compile, so there should be nothing wrong with the logic of the code. This can be used as a workaround.

It is possible to get the same result with map_opt. However, in some cases it’s better for code organisation that these steps are in separate functions, which can be combined more freely.

[arximboldi]

Ping @Siapran

[Siapran]

ok, two ways to fix this:

• change auto here to be the actual optional type (cleanest imo)
• explicitly construct type of whole here from the value of part

either way, I really ought to rewrite a lot of this.

[arximboldi]

Whatever you chose @Siapran sounds good, I trust you on this and would very much appreciate a PR if you have the time!

[Siapran]

I'll try to do that today if whatever I caught on the flight back home will let me 👍

[Siapran]

... I really should just rewrite the whole thing sooner than later. there's some really dubious (if not straight up wrong) moves going on in there.

[arximboldi]

Fixed: #216

Thank you so much @Siapran !