[flang-commits] [flang] [flang] Improve error for misplaced statement after CONTAINS in derived type (PR #215886)

[via flang-commits]

================
@@ -17,10 +19,35 @@ module m
 ! CHECK: error: component definition must precede CONTAINS in a derived type
 ! CHECK-NEXT: {{.*}}real, pointer, dimension(:,:,:), public :: gpoint => null()
      real, pointer, dimension(:,:,:), public :: gpoint => null()
+  end type t1
+
+  ! A second CONTAINS is not a component definition.
+  type :: t2
+   contains
+   contains
+! CHECK: error: expected a type-bound procedure binding (PROCEDURE, GENERIC, or FINAL) after CONTAINS
+  end type t2
+
+  ! An IMPORT after CONTAINS is likewise not a binding.
+  type :: t3
+   contains
+     import
+! CHECK: error: expected a type-bound procedure binding (PROCEDURE, GENERIC, or FINAL) after CONTAINS
+  end type t3
+
+  ! A misplaced subprogram after CONTAINS.
+  type :: t4
+   contains
+     subroutine s
+     end subroutine
+! CHECK: error: expected a type-bound procedure binding (PROCEDURE, GENERIC, or FINAL) after CONTAINS
+  end type t4
+
 ! CHECK-NOT: expected 'FINAL'
 ! CHECK-NOT: expected 'GENERIC'
 ! CHECK-NOT: expected 'PROCEDURE'
-  end type t1
+! CHECK-NOT: expected 'COMPLEX'
+! CHECK-NOT: expected 'INTEGER'
 
----------------
mleair wrote:

Were 'COMPLEX' and 'INTEGER' actually the messages emitted before this fix? intrinsicTypeSpec (R704) is a first() chain ending in DOUBLE COMPLEX/BYTE, and Flang's first() keeps the last same-depth failure message — so the pre-fix spurious message was more likely expected 'LOGICAL' or expected 'DOUBLE COMPLEX'. If so, these two CHECK-NOT lines may be passing vacuously. Could you verify by running the test against an unfixed build?

https://github.com/llvm/llvm-project/pull/215886