[PATCH] Support Fortran common blocks with addresses >INT_MAX.

[PATCH] Support Fortran common blocks with addresses >INT_MAX.

[Chris January]

The address of a Fortran common block may be >INT_MAX. fix_common_block
should be using CORE_ADDR, not int.

2012-03-08  Chris January  <chris.january@allinea.com>

        * stabsread.c (fix_common_block): Change type of valu argument
        to CORE_ADDR.
---
Index: gdb/stabsread.c
===================================================================
--- gdb/stabsread.c.orig	2011-02-17 11:14:42.713081030 +0000
+++ gdb/stabsread.c	2011-02-17 11:15:14.414327296 +0000
@@ -104,7 +104,7 @@ static void
 patch_block_stabs (struct pending *, struct pending_stabs *,
 		   struct objfile *);
 
-static void fix_common_block (struct symbol *, int);
+static void fix_common_block (struct symbol *, CORE_ADDR);
 
 static int read_type_number (char **, int *);
 
@@ -4805,7 +4805,7 @@ common_block_end (struct objfile *objfil
    the common block name).  */
 
 static void
-fix_common_block (struct symbol *sym, int valu)
+fix_common_block (struct symbol *sym, CORE_ADDR valu)
 {
   struct pending *next = (struct pending *) SYMBOL_TYPE (sym);
 

Re: [PATCH] Support Fortran common blocks with addresses >INT_MAX.

[Joel Brobecker]

Chris,

> The address of a Fortran common block may be >INT_MAX. fix_common_block
> should be using CORE_ADDR, not int.
> 
> 2012-03-08  Chris January  <chris.january@allinea.com>
> 
>         * stabsread.c (fix_common_block): Change type of valu argument
>         to CORE_ADDR.

This change looks correct to me too.  Did you run the testsuite to
validate the change, by any chance?

Do you have copyright assignment papers on file with the FSF?  If not,
this change is sufficiently small that we can accept it, as well as
your other contribution ("Add another way of setting up [...]").

If you are planning on making other contributions in the future,
we should really get you started on the assignment process - contact
me privately if you'd like to do so.

-- 
Joel

Re: [PATCH] Support Fortran common blocks with addresses >INT_MAX.

[Chris January]

On Fri, 2012-03-09 at 11:39 -0800, Joel Brobecker wrote:
> > The address of a Fortran common block may be >INT_MAX. fix_common_block
> > should be using CORE_ADDR, not int.
> > 
> > 2012-03-08  Chris January  <chris.january@allinea.com>
> > 
> >         * stabsread.c (fix_common_block): Change type of valu argument
> >         to CORE_ADDR.
> 
> This change looks correct to me too.  Did you run the testsuite to
> validate the change, by any chance?

No, although there aren't any test cases in testsuite/gdb.fortran that
use a common block anyway.

> Do you have copyright assignment papers on file with the FSF?  If not,
> this change is sufficiently small that we can accept it, as well as
> your other contribution ("Add another way of setting up [...]").

Yes, we have a company-wide assignment.

Chris

Re: [PATCH] Support Fortran common blocks with addresses >INT_MAX.

[Joel Brobecker]

> > > 2012-03-08  Chris January  <chris.january@allinea.com>
> > > 
> > >         * stabsread.c (fix_common_block): Change type of valu argument
> > >         to CORE_ADDR.
> > 
> > This change looks correct to me too.  Did you run the testsuite to
> > validate the change, by any chance?
> 
> No, although there aren't any test cases in testsuite/gdb.fortran that
> use a common block anyway.

The problem is that this potentially affects more than just Fortran
programs using common blocks.

Would you have a GNU/Linux box available? You could test your change
a little differently, by running the testsuite with -gstabs+. I think
the following command should work:

    % make check RUNTESTFLAGS="--target_board % unix/gdb:debug_flags=-gstabs+"

The second issue that this raises is that it would be nice to have
one such testcase to detect future changes that might break this.
But given that this is stabs, and that stabs are on their last breath,
I am not going to insist on it this time around, especially if it is
going to take lots of time from you.

FYI: We, at AdaCore, switch to DWARF on AIX last year. We had to move
to GNU ld for that. One thing I have to say, we're not sorry to be rid
of stabs.

> > Do you have copyright assignment papers on file with the FSF?  If not,
> > this change is sufficiently small that we can accept it, as well as
> > your other contribution ("Add another way of setting up [...]").
> 
> Yes, we have a company-wide assignment.

OK, found it! Thanks...

-- 
Joel

Re: [PATCH] Support Fortran common blocks with addresses >INT_MAX.

[Tom Tromey]

>>>>> "Joel" == Joel Brobecker <brobecker@adacore.com> writes:

Joel> Would you have a GNU/Linux box available?

Anybody working on GDB can get an account on the GCC Compile Farm.
This has a variety of machines, some quite nice.

http://gcc.gnu.org/wiki/CompileFarm

Tom

Re: [PATCH] Support Fortran common blocks with addresses >INT_MAX.

[Chris January]

On Mon, 2012-03-12 at 08:30 -0700, Joel Brobecker wrote:
> > > > 2012-03-08  Chris January  <chris.january@allinea.com>
> > > > 
> > > >         * stabsread.c (fix_common_block): Change type of valu argument
> > > >         to CORE_ADDR.
> > > 
> > > This change looks correct to me too.  Did you run the testsuite to
> > > validate the change, by any chance?
> > 
> > No, although there aren't any test cases in testsuite/gdb.fortran that
> > use a common block anyway.
> 
> The problem is that this potentially affects more than just Fortran
> programs using common blocks.
> 
> Would you have a GNU/Linux box available? You could test your change
> a little differently, by running the testsuite with -gstabs+. I think
> the following command should work:
> 
>     % make check RUNTESTFLAGS="--target_board % unix/gdb:debug_flags=-gstabs+"

I ran the test suite on a GNU/Linux (SLES 10) PPC970 box. Four of the
tests have intermittent failures (before and after the patch), but
ignoring those there were no regressions.

Chris

Re: [PATCH] Support Fortran common blocks with addresses >INT_MAX.

[Joel Brobecker]

> On Mon, 2012-03-12 at 08:30 -0700, Joel Brobecker wrote:
> > > > > 2012-03-08  Chris January  <chris.january@allinea.com>
> > > > > 
> > > > >         * stabsread.c (fix_common_block): Change type of valu argument
> > > > >         to CORE_ADDR.
> I ran the test suite on a GNU/Linux (SLES 10) PPC970 box. Four of the
> tests have intermittent failures (before and after the patch), but
> ignoring those there were no regressions.

Thanks, Chris. This patch is also in, now.

-- 
Joel

Re: [PATCH] Support Fortran common blocks with addresses >INT_MAX.

[Tom Tromey]

>>>>> "Chris" == Chris January <chris.january@allinea.com> writes:

Chris> Yes, we have a company-wide assignment.

Great.
Send me email off-list and I will hook you up with write-after-approval
permissions. 

Tom

[PATCH] Support Fortran common blocks with addresses >INT_MAX.

[Chris January]

The address of a Fortran common block may be >INT_MAX. fix_common_block
should be using CORE_ADDR, not int.

2012-03-08  Chris January  <chris.january@allinea.com>

        * stabsread.c (fix_common_block): Change type of valu argument
        to CORE_ADDR.
---
Index: gdb/stabsread.c
===================================================================
--- gdb/stabsread.c.orig	2011-02-17 11:14:42.713081030 +0000
+++ gdb/stabsread.c	2011-02-17 11:15:14.414327296 +0000
@@ -104,7 +104,7 @@ static void
 patch_block_stabs (struct pending *, struct pending_stabs *,
 		   struct objfile *);
 
-static void fix_common_block (struct symbol *, int);
+static void fix_common_block (struct symbol *, CORE_ADDR);
 
 static int read_type_number (char **, int *);
 
@@ -4805,7 +4805,7 @@ common_block_end (struct objfile *objfil
    the common block name).  */
 
 static void
-fix_common_block (struct symbol *sym, int valu)
+fix_common_block (struct symbol *sym, CORE_ADDR valu)
 {
   struct pending *next = (struct pending *) SYMBOL_TYPE (sym);