[patch, fortran] PR27698 - Add diagnostic message for invalid names

[patch, fortran] PR27698 - Add diagnostic message for invalid names

[Bernhard Fischer]

[-- Attachment #1: Type: text/plain, Size: 864 bytes --]

:ADDPATCH fortran:

According to 3.2.1, a name is
letter [alphanumeric-character]...

While we did reject invalid names, this PR wanted to have a diagnostic
message to be printed along rejecting the invalid name.

The attached patch adds this diagnostic message, adds one testcase to
expect the newly added error message and adjusts testcases that trigger
the new error.

Bootstrapped and regtested with check-gfortran on i686-linux-gnu. Ok for
trunk?

PS:
A vaguely related issue in the area of invalid names is whether or not
-fdollar-ok should allow for the '$' as the very first character in a
name ¹). Accepting it would mean that implicit types may need to be
updated for dollar as first char ²)
Attached patch does not address this issue.

¹)
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=27698#c6
²)
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=27698#c11

[-- Attachment #2: gcc.gfortran.pr27698.01.txt --]
[-- Type: text/plain, Size: 410 bytes --]

fortran/ChangeLog

2007-01-03  Bernhard Fischer  <aldot@gcc.gnu.org>

	PR fortran/27698
	* match.c (gfc_match_name): Print diagnostics for invalid character
	in names.


testsuite/ChangeLog

2007-01-03  Bernhard Fischer  <aldot@gcc.gnu.org>

	PR fortran/27698
	* gfortran.dg/invalid_name.f90: New test.
	* gfortran.dg/do_iterator.f90: Adjust pattern for expected error.
	* gfortran.dg/gomp/pr29759.f90: Ditto.

[-- Attachment #3: gcc.gfortran.pr27698.01.diff --]
[-- Type: text/plain, Size: 2737 bytes --]

Index: gcc-4.3/gcc/fortran/match.c
===================================================================
--- gcc-4.3/gcc/fortran/match.c	(revision 120338)
+++ gcc-4.3/gcc/fortran/match.c	(working copy)
@@ -396,6 +396,8 @@ gfc_match_name (char *buffer)
   c = gfc_next_char ();
   if (!ISALPHA (c))
     {
+      if (gfc_error_flag_test() == 0)
+        gfc_error ("Invalid character in name at %C");
       gfc_current_locus = old_loc;
       return MATCH_NO;
     }
Index: gcc-4.3/gcc/testsuite/gfortran.dg/invalid_name.f90
===================================================================
--- gcc-4.3/gcc/testsuite/gfortran.dg/invalid_name.f90	(revision 0)
+++ gcc-4.3/gcc/testsuite/gfortran.dg/invalid_name.f90	(revision 0)
@@ -0,0 +1,6 @@
+! { dg-do compile }
+! Tests the fix for PR27698, where names not starting with a letter were
+! rejected but not diagnosed with a proper message.
+SUBROUTINE _foo ! { dg-error "Invalid character in name" }
+END
+
Index: gcc-4.3/gcc/testsuite/gfortran.dg/do_iterator.f90
===================================================================
--- gcc-4.3/gcc/testsuite/gfortran.dg/do_iterator.f90	(revision 120338)
+++ gcc-4.3/gcc/testsuite/gfortran.dg/do_iterator.f90	(working copy)
@@ -10,4 +10,4 @@ DO I=1,5       ! { dg-error "cannot be r
    READ(5,*,iostat=i) j ! { dg-error "cannot be redefined" "changing do-iterator 3" }
 ENDDO
 END
-
+! { dg-error "Invalid character" "character" { target *-*-* } 7 }
Index: gcc-4.3/gcc/testsuite/gfortran.dg/gomp/pr29759.f90
===================================================================
--- gcc-4.3/gcc/testsuite/gfortran.dg/gomp/pr29759.f90	(revision 120338)
+++ gcc-4.3/gcc/testsuite/gfortran.dg/gomp/pr29759.f90	(working copy)
@@ -22,21 +22,21 @@ PROGRAM test_omp
 
 
 !$OMP PARALLEL &		! { dg-error "Unclassifiable OpenMP" }
-!$    NUM_THREADS(2)		! { dg-error "Unclassifiable" }
+!$    NUM_THREADS(2)		! { dg-error "Unclassifiable|Invalid character" }
 !$OMP END PARALLEL		! { dg-error "Unexpected" }
 
 !$OMP PARALLEL &		! { dg-error "Unclassifiable OpenMP" }
-!$    & NUM_THREADS(2)		! { dg-error "Unclassifiable" }
+!$    & NUM_THREADS(2)		! { dg-error "Unclassifiable|Invalid character" }
 !$OMP END PARALLEL		! { dg-error "Unexpected" }
 
 !$OMP PARALLEL &		! { dg-error "Unclassifiable OpenMP" }
 !
-!$    NUM_THREADS(2)		! { dg-error "Unclassifiable" }
+!$    NUM_THREADS(2)		! { dg-error "Unclassifiable|Invalid character" }
 !$OMP END PARALLEL		! { dg-error "Unexpected" }
 
 !$OMP PARALLEL &		! { dg-error "Unclassifiable OpenMP" }
 !
-!$    & NUM_THREADS(2)		! { dg-error "Unclassifiable" }
+!$    & NUM_THREADS(2)		! { dg-error "Unclassifiable|Invalid character" }
 !$OMP END PARALLEL		! { dg-error "Unexpected" }
 
 END PROGRAM

Re: [patch, fortran] PR27698 - Add diagnostic message for invalid names

[Steve Kargl]

On Wed, Jan 03, 2007 at 05:50:49PM +0100, Bernhard Fischer wrote:
> :ADDPATCH fortran:
> 
> According to 3.2.1, a name is
> letter [alphanumeric-character]...
> 
> While we did reject invalid names, this PR wanted to have a diagnostic
> message to be printed along rejecting the invalid name.
> 
> The attached patch adds this diagnostic message, adds one testcase to
> expect the newly added error message and adjusts testcases that trigger
> the new error.
> 

Why is do_iterator.f90 trigger the patch?  Is this a illegal code triggering
the wrong error message?

> A vaguely related issue in the area of invalid names is whether or not
> -fdollar-ok should allow for the '$' as the very first character in a
> name ?). Accepting it would mean that implicit types may need to be
> updated for dollar as first char ?)
> Attached patch does not address this issue.

IMHO, $ should not be allowed in the 1st position.  But, I'll concede
that I don't know what the form of this extension is with respect to
dusty deck code.  

-- 
Steve

Re: [patch, fortran] PR27698 - Add diagnostic message for invalid names

[Bernhard Fischer]

On Wed, Jan 03, 2007 at 09:02:50PM -0800, Steve Kargl wrote:
>On Wed, Jan 03, 2007 at 05:50:49PM +0100, Bernhard Fischer wrote:
>> :ADDPATCH fortran:
>> 
>> According to 3.2.1, a name is
>> letter [alphanumeric-character]...
>> 
>> While we did reject invalid names, this PR wanted to have a diagnostic
>> message to be printed along rejecting the invalid name.
>> 
>> The attached patch adds this diagnostic message, adds one testcase to
>> expect the newly added error message and adjusts testcases that trigger
>> the new error.
>> 
>
>Why is do_iterator.f90 trigger the patch?  Is this a illegal code triggering
>the wrong error message?

do_iterator.f90 is illegal code for which an error is thrown ("cannot be
redefined inside loop beginning at") and then a different matcher
(trying 'READ(5,*) I' as a name) is tried.
>
>> A vaguely related issue in the area of invalid names is whether or not
>> -fdollar-ok should allow for the '$' as the very first character in a
>> name ?). Accepting it would mean that implicit types may need to be
>> updated for dollar as first char ?)
>> Attached patch does not address this issue.
>
>IMHO, $ should not be allowed in the 1st position.  But, I'll concede
>that I don't know what the form of this extension is with respect to
>dusty deck code.  

Let's see if somebody needs $ as first character in a name. I'll
follow-up in a separate thread with a patch to the documentation
that clarifies this constraint. Sounds good?

Re: [patch, fortran] PR27698 - Add diagnostic message for invalid names

[Steve Kargl]

On Thu, Jan 04, 2007 at 11:12:01AM +0100, Bernhard Fischer wrote:
> On Wed, Jan 03, 2007 at 09:02:50PM -0800, Steve Kargl wrote:
> >On Wed, Jan 03, 2007 at 05:50:49PM +0100, Bernhard Fischer wrote:
> >> :ADDPATCH fortran:
> >> 
> >> According to 3.2.1, a name is
> >> letter [alphanumeric-character]...
> >> 
> >> While we did reject invalid names, this PR wanted to have a diagnostic
> >> message to be printed along rejecting the invalid name.
> >> 
> >> The attached patch adds this diagnostic message, adds one testcase to
> >> expect the newly added error message and adjusts testcases that trigger
> >> the new error.
> >> 
> >
> >Why is do_iterator.f90 trigger the patch?  Is this a illegal code triggering
> >the wrong error message?
> 
> do_iterator.f90 is illegal code for which an error is thrown ("cannot be
> redefined inside loop beginning at") and then a different matcher
> (trying 'READ(5,*) I' as a name) is tried.

OK, I understand the change now.  Please, commit the patch.

-- 
Steve

Re: [patch, fortran] PR27698 - Add diagnostic message for invalid names

[Bernhard Fischer]

On Thu, Jan 04, 2007 at 10:13:32PM -0800, Steve Kargl wrote:
>On Thu, Jan 04, 2007 at 11:12:01AM +0100, Bernhard Fischer wrote:
>> On Wed, Jan 03, 2007 at 09:02:50PM -0800, Steve Kargl wrote:
>> >On Wed, Jan 03, 2007 at 05:50:49PM +0100, Bernhard Fischer wrote:
>> >> :ADDPATCH fortran:
>> >> 
>> >> According to 3.2.1, a name is
>> >> letter [alphanumeric-character]...
>> >> 
>> >> While we did reject invalid names, this PR wanted to have a diagnostic
>> >> message to be printed along rejecting the invalid name.
>> >> 
>> >> The attached patch adds this diagnostic message, adds one testcase to
>> >> expect the newly added error message and adjusts testcases that trigger
>> >> the new error.
>> >> 
>> >
>> >Why is do_iterator.f90 trigger the patch?  Is this a illegal code triggering
>> >the wrong error message?
>> 
>> do_iterator.f90 is illegal code for which an error is thrown ("cannot be
>> redefined inside loop beginning at") and then a different matcher
>> (trying 'READ(5,*) I' as a name) is tried.
>
>OK, I understand the change now.  Please, commit the patch.

Applied to trunk.

Thanks!