silence deprecation warning for enum with custom __eq__

[davidhewitt]

Fixes #4687

This adjusts the deprecation warning so that it's conditionally emitted using good old auto ref specialization, so that if the user provides __eq__ (or any other __richcmp__ component) then the deprecation warning is not emitted.

[Icxolu]

Ah, very clever! This seems much nicer than an opt-out option. (But I'm still happy once we can remove all of this again 🙃)

It doesn't work however with a manual __richcmp__ implementation directly. From a quick test we could do something like to following to catch that as well:

diff --git a/pyo3-macros-backend/src/pymethod.rs b/pyo3-macros-backend/src/pymethod.rs
index 1254a8d51..a5993c981 100644
--- a/pyo3-macros-backend/src/pymethod.rs
+++ b/pyo3-macros-backend/src/pymethod.rs
@@ -1348,6 +1348,13 @@ impl SlotDef {
         )?;
         let name = spec.name;
         let holders = holders.init_holders(ctx);
+
+        let dep = if method_name == "__richcmp__" {
+            quote! { impl #pyo3_path::impl_::pyclass::HasCustomRichCmp for #cls {} }
+        } else {
+            TokenStream::default()
+        };
+
         let associated_method = quote! {
             #[allow(non_snake_case)]
             unsafe fn #wrapper_ident(
@@ -1355,6 +1362,7 @@ impl SlotDef {
                 _raw_slf: *mut #pyo3_path::ffi::PyObject,
                 #(#arg_idents: #arg_types),*
             ) -> #pyo3_path::PyResult<#ret_ty> {
+                #dep
                 let function = #cls::#name; // Shadow the method name to avoid #3017
                 let _slf = _raw_slf;
                 #holders

[davidhewitt]

Ah, good catch, thanks👍. Pushed a commit with that patch.

(But I'm still happy once we can remove all of this again 🙃)

Haha yes indeed, we can probably delete this next week after the 0.23 branch is cut and released!

[Icxolu]

Just pushed a commit to fix ci (allow non_local_definitions and bless ui tests)

[davidhewitt]

Thanks 🙏