[flang-commits] [flang] [flang][Semantics] Adding check integer kind on STAT= specifier #230463 (PR #231103)

[Eugene Epshteyn via flang-commits]

================
@@ -145,6 +153,15 @@ static void CheckSyncStat(SemanticsContext &context,
               context.Say( // C1172
                   "The stat-variable in a sync-stat-list may not be repeated"_err_en_US);
             }
+            if (const auto *expr{GetExpr(context, stat)})
+              if (auto type{expr->GetType()}; type &&
+                  type->category() == TypeCategory::Integer &&
+                  type->kind() < 2 &&
+                  context.ShouldWarn(
+                      common::LanguageFeature::StatKind)) { // F2023 9.7.4 p1
+                context.Say(
+                    "The stat-variable must have an integer kind of at least 16-bit"_warn_en_US);
+              }
----------------
eugeneepshteyn wrote:

(This comment was generated with the help of AI)

This misses the most common STAT= sites: with `-pedantic`, ALLOCATE, DEALLOCATE and FORM TEAM get no warning.

```fortran
program p
  use iso_fortran_env, only: team_type
  integer(1) :: s
  integer, allocatable :: a(:)
  type(team_type) :: t
  sync all(stat=s)          ! warns
  allocate(a(3), stat=s)    ! no warning
  deallocate(a, stat=s)     ! no warning
  form team(1, t, stat=s)   ! no warning
end program
```

`flang -fsyntax-only -pedantic` warns only on line 6.

F2023 9.7.4 p1 applies to every *stat-variable*, not just the ones in sync-stat-lists, image selectors and EVENT/NOTIFY WAIT. All of these sites except the image selector reach semantics through `parser::StatOrErrmsg`, so doing the check once per `StatOrErrmsg`, plus the image selector, would cover them all and replace the three copies in this file. IOSTAT= is a bare `StatVariable` outside `StatOrErrmsg`, so it would not be affected. The test should then cover ALLOCATE, DEALLOCATE and FORM TEAM too.

https://github.com/llvm/llvm-project/pull/231103