[Patch, fortran] PR89462 - [11/12/13/14 Regression] gfortran loops in code generation

[Patch, fortran] PR89462 - [11/12/13/14 Regression] gfortran loops in code generation

[Paul Richard Thomas]

[-- Attachment #1.1: Type: text/plain, Size: 787 bytes --]

Hi All,

Jakub pinpointed the source of this bug in comment 6 of the PR. The rest
was 'obvious' :-)

I plan to push the patch to mainline in the next 24 hours unless there are
opinions to the contrary. Backporting is proposed to occur a couple of
weeks later.

Best regards

Paul

Fortran: Generate new charlens for shared symbol typespecs [PR89462]

2024-04-23  Paul Thomas  <pault@gcc.gnu.org>
   Jakub Jelinek  <jakub@gcc.gnu.org>

gcc/fortran
PR fortran/89462
* decl.cc (build_sym): Add an extra argument 'elem'. If 'elem'
is greater than 1, gfc_new_charlen is called to generate a new
charlen, registered in the symbol namespace.
(variable_decl, enumerator_decl): Set the new argument in the
calls to build_sym.

gcc/testsuite/
PR fortran/89462
* gfortran.dg/pr89462.f90: New test.

[-- Attachment #2: submit.diff --]
[-- Type: text/x-patch, Size: 2179 bytes --]

diff --git a/gcc/fortran/decl.cc b/gcc/fortran/decl.cc
index a7576f4bc40..b8308aeee55 100644
--- a/gcc/fortran/decl.cc
+++ b/gcc/fortran/decl.cc
@@ -1713,7 +1713,7 @@ gfc_verify_c_interop_param (gfc_symbol *sym)
 /* Function called by variable_decl() that adds a name to the symbol table.  */
 
 static bool
-build_sym (const char *name, gfc_charlen *cl, bool cl_deferred,
+build_sym (const char *name, int elem, gfc_charlen *cl, bool cl_deferred,
 	   gfc_array_spec **as, locus *var_locus)
 {
   symbol_attribute attr;
@@ -1778,7 +1778,10 @@ build_sym (const char *name, gfc_charlen *cl, bool cl_deferred,
 
   if (sym->ts.type == BT_CHARACTER)
     {
-      sym->ts.u.cl = cl;
+      if (elem > 1)
+	sym->ts.u.cl = gfc_new_charlen (sym->ns, cl);
+      else
+	sym->ts.u.cl = cl;
       sym->ts.deferred = cl_deferred;
     }
 
@@ -2960,7 +2963,7 @@ variable_decl (int elem)
      create a symbol for those yet.  If we fail to create the symbol,
      bail out.  */
   if (!gfc_comp_struct (gfc_current_state ())
-      && !build_sym (name, cl, cl_deferred, &as, &var_locus))
+      && !build_sym (name, elem, cl, cl_deferred, &as, &var_locus))
     {
       m = MATCH_ERROR;
       goto cleanup;
@@ -10938,7 +10941,7 @@ enumerator_decl (void)
   /* OK, we've successfully matched the declaration.  Now put the
      symbol in the current namespace. If we fail to create the symbol,
      bail out.  */
-  if (!build_sym (name, NULL, false, &as, &var_locus))
+  if (!build_sym (name, 1, NULL, false, &as, &var_locus))
     {
       m = MATCH_ERROR;
       goto cleanup;
diff --git a/gcc/testsuite/gfortran.dg/pr89462.f90 b/gcc/testsuite/gfortran.dg/pr89462.f90
new file mode 100644
index 00000000000..9efdb1adbc7
--- /dev/null
+++ b/gcc/testsuite/gfortran.dg/pr89462.f90
@@ -0,0 +1,12 @@
+! { dg-do compile }
+! Test the fix for PR89462 in which the shared 'cl' field of the typespec
+! shared between 'test', 'TR' and 'aTP' caused the compiler to go into an
+! infinite loop.
+! Contributed by Sergei Trofimovich  <slyich@gmail.com>
+  CHARACTER*1 FUNCTION test(H)
+     CHARACTER*1 test2,TR,aTP
+     ENTRY test2(L)
+     CALL ttest3(aTP)
+     test = TR
+     RETURN
+  END

Re: [Patch, fortran] PR89462 - [11/12/13/14 Regression] gfortran loops in code generation

[Paul Richard Thomas]

[-- Attachment #1.1: Type: text/plain, Size: 1543 bytes --]

Hi,

The linaro pre-commit error testing picked up errors for arm and aarch
since they set the option -pedantic-errors.
/home/tcwg-build/workspace/tcwg_gnu_4/abe/snapshots/gcc.git~master/gcc/testsuite/gfortran.dg/pr89462.f90:6:14:
Warning: Obsolescent feature: Old-style character length at (1)
/home/tcwg-build/workspace/tcwg_gnu_4/abe/snapshots/gcc.git~master/gcc/testsuite/gfortran.dg/pr89462.f90:7:17:
Warning: Obsolescent feature: Old-style character length at (1)

I have added the option to the testcase together with the corresponding
warnings as in the attached.

I will wait for 24 hours more.

Paul

On Tue, 23 Apr 2024 at 16:25, Paul Richard Thomas <
paul.richard.thomas@gmail.com> wrote:

> Hi All,
>
> Jakub pinpointed the source of this bug in comment 6 of the PR. The rest
> was 'obvious' :-)
>
> I plan to push the patch to mainline in the next 24 hours unless there are
> opinions to the contrary. Backporting is proposed to occur a couple of
> weeks later.
>
> Best regards
>
> Paul
>
> Fortran: Generate new charlens for shared symbol typespecs [PR89462]
>
> 2024-04-23  Paul Thomas  <pault@gcc.gnu.org>
>    Jakub Jelinek  <jakub@gcc.gnu.org>
>
> gcc/fortran
> PR fortran/89462
> * decl.cc (build_sym): Add an extra argument 'elem'. If 'elem'
> is greater than 1, gfc_new_charlen is called to generate a new
> charlen, registered in the symbol namespace.
> (variable_decl, enumerator_decl): Set the new argument in the
> calls to build_sym.
>
> gcc/testsuite/
> PR fortran/89462
> * gfortran.dg/pr89462.f90: New test.
>
>

[-- Attachment #2: submit.diff --]
[-- Type: text/x-patch, Size: 2984 bytes --]

diff --git a/gcc/fortran/decl.cc b/gcc/fortran/decl.cc
index a7576f4bc40..b8308aeee55 100644
--- a/gcc/fortran/decl.cc
+++ b/gcc/fortran/decl.cc
@@ -1713,7 +1713,7 @@ gfc_verify_c_interop_param (gfc_symbol *sym)
 /* Function called by variable_decl() that adds a name to the symbol table.  */
 
 static bool
-build_sym (const char *name, gfc_charlen *cl, bool cl_deferred,
+build_sym (const char *name, int elem, gfc_charlen *cl, bool cl_deferred,
 	   gfc_array_spec **as, locus *var_locus)
 {
   symbol_attribute attr;
@@ -1778,7 +1778,10 @@ build_sym (const char *name, gfc_charlen *cl, bool cl_deferred,
 
   if (sym->ts.type == BT_CHARACTER)
     {
-      sym->ts.u.cl = cl;
+      if (elem > 1)
+	sym->ts.u.cl = gfc_new_charlen (sym->ns, cl);
+      else
+	sym->ts.u.cl = cl;
       sym->ts.deferred = cl_deferred;
     }
 
@@ -2960,7 +2963,7 @@ variable_decl (int elem)
      create a symbol for those yet.  If we fail to create the symbol,
      bail out.  */
   if (!gfc_comp_struct (gfc_current_state ())
-      && !build_sym (name, cl, cl_deferred, &as, &var_locus))
+      && !build_sym (name, elem, cl, cl_deferred, &as, &var_locus))
     {
       m = MATCH_ERROR;
       goto cleanup;
@@ -10938,7 +10941,7 @@ enumerator_decl (void)
   /* OK, we've successfully matched the declaration.  Now put the
      symbol in the current namespace. If we fail to create the symbol,
      bail out.  */
-  if (!build_sym (name, NULL, false, &as, &var_locus))
+  if (!build_sym (name, 1, NULL, false, &as, &var_locus))
     {
       m = MATCH_ERROR;
       goto cleanup;
diff --git a/gcc/fortran/trans-array.cc b/gcc/fortran/trans-array.cc
index 30b84762346..322ff552813 100644
--- a/gcc/fortran/trans-array.cc
+++ b/gcc/fortran/trans-array.cc
@@ -11756,8 +11756,16 @@ gfc_walk_variable_expr (gfc_ss * ss, gfc_expr * expr)
   gfc_fix_class_refs (expr);
 
   for (ref = expr->ref; ref; ref = ref->next)
-    if (ref->type == REF_ARRAY && ref->u.ar.type != AR_ELEMENT)
-      break;
+    {
+      if (ref->type == REF_COMPONENT
+	  && ref->u.c.component->attr.function)
+	{
+	  ref = NULL;
+	  break;
+	}
+      if (ref->type == REF_ARRAY && ref->u.ar.type != AR_ELEMENT)
+          break;
+    }
 
   return gfc_walk_array_ref (ss, expr, ref);
 }
diff --git a/gcc/testsuite/gfortran.dg/pr89462.f90 b/gcc/testsuite/gfortran.dg/pr89462.f90
new file mode 100644
index 00000000000..b2a4912fcc8
--- /dev/null
+++ b/gcc/testsuite/gfortran.dg/pr89462.f90
@@ -0,0 +1,13 @@
+! { dg-do compile }
+! { dg-options "-pedantic-errors" }
+! Test the fix for PR89462 in which the shared 'cl' field of the typespec
+! shared between 'test', 'TR' and 'aTP' caused the compiler to go into an
+! infinite loop.
+! Contributed by Sergei Trofimovich  <slyich@gmail.com>
+  CHARACTER*1 FUNCTION test(H) ! { dg-warning "Old-style character length" }
+     CHARACTER*1 test2,TR,aTP  ! { dg-warning "Old-style character length" }
+     ENTRY test2(L)
+     CALL ttest3(aTP)
+     test = TR
+     RETURN
+  END

Re: [Patch, fortran] PR89462 - [11/12/13/14 Regression] gfortran loops in code generation

[Paul Richard Thomas]

[-- Attachment #1: Type: text/plain, Size: 1803 bytes --]

PS ignore the chunk in trans-array.cc. It is an attempt to fix PR93678 that
literally did nothing.

Paul

On Wed, 24 Apr 2024 at 07:05, Paul Richard Thomas <
paul.richard.thomas@gmail.com> wrote:

> Hi,
>
> The linaro pre-commit error testing picked up errors for arm and aarch
> since they set the option -pedantic-errors.
> /home/tcwg-build/workspace/tcwg_gnu_4/abe/snapshots/gcc.git~master/gcc/testsuite/gfortran.dg/pr89462.f90:6:14:
> Warning: Obsolescent feature: Old-style character length at (1)
> /home/tcwg-build/workspace/tcwg_gnu_4/abe/snapshots/gcc.git~master/gcc/testsuite/gfortran.dg/pr89462.f90:7:17:
> Warning: Obsolescent feature: Old-style character length at (1)
>
> I have added the option to the testcase together with the corresponding
> warnings as in the attached.
>
> I will wait for 24 hours more.
>
> Paul
>
> On Tue, 23 Apr 2024 at 16:25, Paul Richard Thomas <
> paul.richard.thomas@gmail.com> wrote:
>
>> Hi All,
>>
>> Jakub pinpointed the source of this bug in comment 6 of the PR. The rest
>> was 'obvious' :-)
>>
>> I plan to push the patch to mainline in the next 24 hours unless there
>> are opinions to the contrary. Backporting is proposed to occur a couple of
>> weeks later.
>>
>> Best regards
>>
>> Paul
>>
>> Fortran: Generate new charlens for shared symbol typespecs [PR89462]
>>
>> 2024-04-23  Paul Thomas  <pault@gcc.gnu.org>
>>    Jakub Jelinek  <jakub@gcc.gnu.org>
>>
>> gcc/fortran
>> PR fortran/89462
>> * decl.cc (build_sym): Add an extra argument 'elem'. If 'elem'
>> is greater than 1, gfc_new_charlen is called to generate a new
>> charlen, registered in the symbol namespace.
>> (variable_decl, enumerator_decl): Set the new argument in the
>> calls to build_sym.
>>
>> gcc/testsuite/
>> PR fortran/89462
>> * gfortran.dg/pr89462.f90: New test.
>>
>>