[patch, fortran] PR30371: KILL subroutine accepts array arguments.

[patch, fortran] PR30371: KILL subroutine accepts array arguments.

[Brooks Moses]

[-- Attachment #1: Type: text/plain, Size: 634 bytes --]

The KILL() subroutine accepts arrays for the first two arguments, but 
then appears (to me) to erroneously treat them as scalars.  I presume 
that this is in error and have filed PR30371; the attached patch adds 
appropriate checks to gfc_check_kill_sub() to confirm that all of the 
arguments are scalar.

-----------------------------------------------------------------
2007-01-03  Brooks Moses  <brooks.moses@codesourcery.com>

	* check.c (gfc_check_kill_sub): Add checks for non-scalar
	arguments.

-----------------------------------------------------------------

Tested on trunk on i686-pc-linux-gnu.  Ok to commit?

- Brooks

[-- Attachment #2: killscalar.diff --]
[-- Type: text/x-patch, Size: 485 bytes --]

Index: check.c
===================================================================
--- check.c	(revision 120442)
+++ check.c	(working copy)
@@ -1327,9 +1327,15 @@
   if (type_check (pid, 0, BT_INTEGER) == FAILURE)
     return FAILURE;
 
+  if (scalar_check (pid, 0) == FAILURE)
+    return FAILURE;
+
   if (type_check (sig, 1, BT_INTEGER) == FAILURE)
     return FAILURE;
 
+  if (scalar_check (sig, 1) == FAILURE)
+    return FAILURE;
+
   if (status == NULL)
     return SUCCESS;
 

Re: [patch, fortran] PR30371: KILL subroutine accepts array arguments.

[Steve Kargl]

On Thu, Jan 04, 2007 at 05:12:27PM -0800, Brooks Moses wrote:
> The KILL() subroutine accepts arrays for the first two arguments, but 
> then appears (to me) to erroneously treat them as scalars.  I presume 
> that this is in error and have filed PR30371; the attached patch adds 
> appropriate checks to gfc_check_kill_sub() to confirm that all of the 
> arguments are scalar.
> 
> -----------------------------------------------------------------
> 2007-01-03  Brooks Moses  <brooks.moses@codesourcery.com>
> 
> 	* check.c (gfc_check_kill_sub): Add checks for non-scalar
> 	arguments.
> 

OK.

Do we also need to the scalar checks in gfc_check_kill?
A quick glance at the code suggests that we do.

-- 
Steve

Re: [patch, fortran] PR30371: KILL subroutine accepts array arguments.

[Brooks Moses]

Steve Kargl wrote:
> On Thu, Jan 04, 2007 at 05:12:27PM -0800, Brooks Moses wrote:
>> The KILL() subroutine accepts arrays for the first two arguments, but 
>> then appears (to me) to erroneously treat them as scalars.  I presume 
>> that this is in error and have filed PR30371; the attached patch adds 
>> appropriate checks to gfc_check_kill_sub() to confirm that all of the 
>> arguments are scalar.
>>
>> -----------------------------------------------------------------
>> 2007-01-03  Brooks Moses  <brooks.moses@codesourcery.com>
>>
>> 	* check.c (gfc_check_kill_sub): Add checks for non-scalar
>> 	arguments.
>>
> 
> OK.
> 
> Do we also need to the scalar checks in gfc_check_kill?
> A quick glance at the code suggests that we do.

I'm not sure.  The KILL() function is flagged as ELEMENTAL in 
intrinsic.c, and I'm not familiar enough with how elemental functions 
are handled to know if that will do the right thing or not.

- Brooks

Re: [patch, fortran] PR30371: KILL subroutine accepts array arguments.

[Steve Kargl]

On Thu, Jan 04, 2007 at 05:38:18PM -0800, Brooks Moses wrote:
> Steve Kargl wrote:
> >On Thu, Jan 04, 2007 at 05:12:27PM -0800, Brooks Moses wrote:
> >>The KILL() subroutine accepts arrays for the first two arguments, but 
> >>then appears (to me) to erroneously treat them as scalars.  I presume 
> >>that this is in error and have filed PR30371; the attached patch adds 
> >>appropriate checks to gfc_check_kill_sub() to confirm that all of the 
> >>arguments are scalar.
> >>
> >>-----------------------------------------------------------------
> >>2007-01-03  Brooks Moses  <brooks.moses@codesourcery.com>
> >>
> >>	* check.c (gfc_check_kill_sub): Add checks for non-scalar
> >>	arguments.
> >
> >OK.
> >
> >Do we also need to the scalar checks in gfc_check_kill?
> >A quick glance at the code suggests that we do.
> 
> I'm not sure.  The KILL() function is flagged as ELEMENTAL in 
> intrinsic.c, and I'm not familiar enough with how elemental functions 
> are handled to know if that will do the right thing or not.
> 

I think the ELEMENTAL marking is incorrect because kill() is
only present due to backwards compatibility with g77.  g77.info
says the arguments are scalar.  The implementation in kill.c
suggests that these should also be scalar.

-- 
Steve

Re: [patch, fortran] PR30371: KILL subroutine accepts array arguments.

[Brooks Moses]

Brooks Moses wrote:
> Steve Kargl wrote:
>> Do we also need to the scalar checks in gfc_check_kill?
>> A quick glance at the code suggests that we do.
> 
> I'm not sure.  The KILL() function is flagged as ELEMENTAL in 
> intrinsic.c, and I'm not familiar enough with how elemental functions 
> are handled to know if that will do the right thing or not.

Well, something does the wrong thing, anyway:

debian-gfortran:~/test> cat kill9.f90
program kill9
   integer i1, i2, stat(2)
   stat = kill( (/i1, i2/), (/9, 9/) )
end

debian-gfortran:~/test> ../bin-trunk/bin/gfortran kill9.f90 -o kill9
kill9.f90: In function 'MAIN__':
kill9.f90:1: internal compiler error: in gfc_conv_intrinsic_funcall, at 
fortran/trans-intrinsic.c:1452

Ideally, since KILL is documented as an elemental function, this ought 
to work.  Though we could just make it non-elemental and put in checks, 
as the simple way out.

- Brooks

Re: [patch, fortran] PR30371: KILL subroutine accepts array arguments.

[Steve Kargl]

On Thu, Jan 04, 2007 at 10:13:31PM -0800, Brooks Moses wrote:
> Brooks Moses wrote:
> >Steve Kargl wrote:
> >>Do we also need to the scalar checks in gfc_check_kill?
> >>A quick glance at the code suggests that we do.
> >
> >I'm not sure.  The KILL() function is flagged as ELEMENTAL in 
> >intrinsic.c, and I'm not familiar enough with how elemental functions 
> >are handled to know if that will do the right thing or not.
> 
> Well, something does the wrong thing, anyway:
> 
> debian-gfortran:~/test> cat kill9.f90
> program kill9
>   integer i1, i2, stat(2)
>   stat = kill( (/i1, i2/), (/9, 9/) )
> end
> 
> debian-gfortran:~/test> ../bin-trunk/bin/gfortran kill9.f90 -o kill9
> kill9.f90: In function 'MAIN__':
> kill9.f90:1: internal compiler error: in gfc_conv_intrinsic_funcall, at 
> fortran/trans-intrinsic.c:1452
> 
> Ideally, since KILL is documented as an elemental function, this ought 
> to work.  Though we could just make it non-elemental and put in checks, 
> as the simple way out.
> 

I vote for making nonelemental.  This is what g77 had.  Otherwise, 
we'll need implement better name resolution in iresolve.c and 
implement the array versions of kill().   Seems like a lot of work
for little return.

-- 
Steve

Re: [patch, fortran] PR30371: KILL subroutine accepts array arguments.

[FX Coudert]

> I vote for making nonelemental.

I also do. When I checked the elemental and valid_as_actual_arg  
fields for all intrinsics in intrinsic.c, I must have overlooked this  
one, otherwise I would have corrected it at that time.

FX