[flang-commits] [flang] [flang][semantics] Allow forward-typed PARAMETER constants under IMPLICIT NONE (PR #203398)

[Eugene Epshteyn via flang-commits]

================
@@ -0,0 +1,41 @@
+! RUN: not %flang_fc1 -pedantic %s 2>&1 | FileCheck %s
+! Test extension: a named constant defined by a PARAMETER statement may appear
+! before its explicit type declaration in a scope with IMPLICIT NONE(TYPE).
+! It acquires the type it would have had under implicit typing rules, which a
+! later explicit declaration must match.
+
+! Forward reference accepted, matching INTEGER declaration appears later.
+!CHECK: warning: 'n' was used without (or before) being explicitly typed
+subroutine s1
+  implicit none
+  parameter(n=4096)
+  integer n
+end
+
+! The would-be-implicit type need not be INTEGER; a matching REAL declaration
+! is accepted.
+!CHECK: warning: 'x' was used without (or before) being explicitly typed
+subroutine s2
+  implicit none
+  parameter(x=1.5)
+  real x
+end
+
+! A later declaration whose type differs from the would-be-implicit type is
+! rejected.
+!CHECK: warning: 'm' was used without (or before) being explicitly typed
+subroutine s3
+  implicit none
+  parameter(m=4096)
+!CHECK: error: The type of 'm' has already been implicitly declared as INTEGER(4)
----------------
eugeneepshteyn wrote:

> When you say 'm' has been implicitly declared as INTEGER(4), is that based off of the value assigned to 'm' or is that based on Fortran's implicit type rules which state that variables that begin with 'i' through 'n' are integer?

Based on Fortran's implicit type rules.

> Also, is 'm' technically declared implicitly with default integer and not necessarily integer(4)?

Fair point, I'll update the comment.

> In other words, should you also have tests that do not have the implicit none statement?

I definitely should!

https://github.com/llvm/llvm-project/pull/203398