Fix #:methods regression at top-level

Please merge to v5.3.4
(cherry picked from commit fa80c53115)

collects/racket/private/define-struct.rkt

@@ -319,7 +319,12 @@
             (raise-syntax-error #f
                                 "not a name for a generics group"
                                 gen:foo gen:foo))
-          (unless (and (identifier? gen:foo) (identifier-binding gen:foo))
+          (unless (and (identifier? gen:foo)
+                       ;; at the top-level, it's not possible to check
+                       ;; if this `gen:foo` is bound, so we give up on the
+                       ;; error message in that case
+                       (or (eq? (syntax-local-context) 'top-level)
+                           (identifier-binding gen:foo)))
             (bad-generics))
           (define gen:foo-val (syntax-local-value gen:foo))
           (unless (and (list? gen:foo-val)

collects/tests/generic/tests.rkt

@@ -15,4 +15,5 @@
          "contract.rkt"
          "from-unstable.rkt"
          "poly-contracts.rkt"
-         "empty-interface.rkt")
+         "empty-interface.rkt"
+         "top-level.rkt")

collects/tests/generic/top-level.rkt

@@ -0,0 +1,20 @@
+#lang racket/base
+
+;; check that generics work at the top-level
+
+(require racket/generic
+         rackunit)
+
+(define ns (make-base-namespace))
+
+(check-not-exn
+ (λ ()
+  (eval '(require racket/generic) ns)
+  (eval '(define-generics foobar [foo foobar a1]) ns)
+  (eval '(struct inst ()
+                 ;; make sure `gen:foobar` doesn't cause an
+                 ;; error here
+                 #:methods gen:foobar
+                 [(define (foo foobar a1) 0)])
+        ns)))
+