add -d:nimLegacyNoHashRef for a transition period which avoids defining hash(ref) (#17858)

changelog.md

@@ -51,6 +51,7 @@
   for previous behavior.
 
 - `hashes.hash` now supports `object` and `ref` (can be overloaded in user code).
+  For a transition period, use `-d:nimLegacyNoHashRef` to avoid defining `hash(ref)`.
 - `hashes.hash(proc|ptr|ref|pointer)` now calls `hash(int)` and honors `-d:nimIntHash1`,
   `hashes.hash(closure)` has also been improved.
 

lib/pure/hashes.nim

@@ -228,12 +228,17 @@ proc hash*(x: pointer): Hash {.inline.} =
     let y = cast[int](x)
   hash(y) # consistent with code expecting scrambled hashes depending on `nimIntHash1`.
 
-proc hash*[T](x: ref[T] | ptr[T]): Hash {.inline.} =
+proc hash*[T](x: ptr[T]): Hash {.inline.} =
   ## Efficient `hash` overload.
   runnableExamples:
     var a: array[10, uint8]
     assert a[0].addr.hash != a[1].addr.hash
     assert cast[pointer](a[0].addr).hash == a[0].addr.hash
+  hash(cast[pointer](x))
+
+when not defined(nimLegacyNoHashRef):
+  proc hash*[T](x: ref[T]): Hash {.inline.} =
+    ## Efficient `hash` overload.
     runnableExamples:
       type A = ref object
         x: int