-data-list-register-names regression?

-data-list-register-names regression?

[Vladimir Prus]

Hello,

I've tried to use Eclipse CDT with GDB 7.0 today, and found that the registers
views is totally broken, as it in does not show any registers. The version
of Eclipse I have used is eclipse-cpp-galileo-SR1, which I believe is the
latest official release, and the problem reproduces with current CVS HEAD.
Everything works more or less fine with 6.8. The root problem is this:

	$ gdb/gdb -i=mi gdb/gdb
	...
	(gdb)
	-data-list-register-names
	^error,msg="No registers."
	(gdb)

It seems that -data-list-register-names does not work until the application
was started. In 6.8, it worked.

Is this change intentional, and if so, do we care about fixing it in
general or for 7.1 in particular?

Thanks,

-- 
Vladimir Prus
CodeSourcery
vladimir@codesourcery.com
(650) 331-3385 x722

Re: -data-list-register-names regression?

[Joel Brobecker]

> I've tried to use Eclipse CDT with GDB 7.0 today, and found that the
> registers views is totally broken, as it in does not show any
> registers.

Just for the record: I've noticed this report and will take into account
the discussion in order to decide what to do for 7.1.

I personally don't have an opinion, right now - I don't know what to think.
On the one hand, I guess we should be able to print the list of register
names even if the program hasn't started yet.  On the other hand, until
we do have an executable file, we don't know what the architecture is.
And even so, doesn't the architecture actually depend on the frame (the
case of the powerpc+spu)?

Given that this was already a problem with 7.0, I propose we pass for
7.1 and consider either 7.1.1 and/or 7.2.

-- 
Joel

Re: -data-list-register-names regression?

[Daniel Jacobowitz]

On Wed, Mar 17, 2010 at 07:26:09AM -0700, Joel Brobecker wrote:
> I personally don't have an opinion, right now - I don't know what to think.
> On the one hand, I guess we should be able to print the list of register
> names even if the program hasn't started yet.  On the other hand, until
> we do have an executable file, we don't know what the architecture is.
> And even so, doesn't the architecture actually depend on the frame (the
> case of the powerpc+spu)?

IOW: get_current_arch.

> Given that this was already a problem with 7.0, I propose we pass for
> 7.1 and consider either 7.1.1 and/or 7.2.

Fine with me.

-- 
Daniel Jacobowitz
CodeSourcery

[patch] Re: -data-list-register-names regression?

[Ulrich Weigand]

Vladimir Prus wrote:

> It seems that -data-list-register-names does not work until the application
> was started. In 6.8, it worked.

Hmm, seems this was my fault:
http://sourceware.org/ml/gdb-patches/2009-06/msg00124.html

I guess we can allow this again by using get_current_arch () instead of
requiring the presence of a selected frame.  The patch below implements
this.

As a side note, I'm wondering why this matters so much to Eclipse: even
if we allow -data-list-register-names before the application started,
Eclipse still will not be able to show register *contents*.  And once
the application *has* started, I'd hope Eclipse re-reads the list of
register names anyway, because it might have changed from before ...

Tested on powerpc-linux.  Thoughts on whether we should this?

Bye,
Ulrich


ChangeLog:

	* mi/mi-main.c (mi_cmd_list_thread_groups): Use get_current_arch
	instead of selected frame architecture.

testsuite/ChangeLog:

	* gdb.mi/gdb680.exp: Revert 2009-06-17 change.


Index: gdb/mi/mi-main.c
===================================================================
RCS file: /cvs/src/src/gdb/mi/mi-main.c,v
retrieving revision 1.169
diff -u -p -r1.169 mi-main.c
--- gdb/mi/mi-main.c	24 Feb 2010 23:11:28 -0000	1.169
+++ gdb/mi/mi-main.c	17 Mar 2010 20:58:38 -0000
@@ -852,7 +852,6 @@ mi_cmd_list_thread_groups (char *command
 void
 mi_cmd_data_list_register_names (char *command, char **argv, int argc)
 {
-  struct frame_info *frame;
   struct gdbarch *gdbarch;
   int regnum, numregs;
   int i;
@@ -864,8 +863,7 @@ mi_cmd_data_list_register_names (char *c
      In this case, some entries of gdbarch_register_name will change depending
      upon the particular processor being debugged.  */
 
-  frame = get_selected_frame (NULL);
-  gdbarch = get_frame_arch (frame);
+  gdbarch = get_current_arch ();
   numregs = gdbarch_num_regs (gdbarch) + gdbarch_num_pseudo_regs (gdbarch);
 
   cleanup = make_cleanup_ui_out_list_begin_end (uiout, "register-names");
Index: gdb/testsuite/gdb.mi/gdb680.exp
===================================================================
RCS file: /cvs/src/src/gdb/testsuite/gdb.mi/gdb680.exp,v
retrieving revision 1.8
diff -u -p -r1.8 gdb680.exp
--- gdb/testsuite/gdb.mi/gdb680.exp	1 Jan 2010 07:32:03 -0000	1.8
+++ gdb/testsuite/gdb.mi/gdb680.exp	17 Mar 2010 20:58:39 -0000
@@ -27,7 +27,7 @@ if [mi_gdb_start] {
 
 proc do_test {count} {
   mi_gdb_test "-data-list-register-names -1" \
-    {\^error,msg=\"No registers.\"} \
+    {\^error,msg=\"bad register number\"} \
     "-data-list-register-names -1, try $count"
 }
 

-- 
  Dr. Ulrich Weigand
  GNU Toolchain for Linux on System z and Cell BE
  Ulrich.Weigand@de.ibm.com

Re: [patch] Re: -data-list-register-names regression?

[Daniel Jacobowitz]

On Wed, Mar 17, 2010 at 10:38:24PM +0100, Ulrich Weigand wrote:
> Tested on powerpc-linux.  Thoughts on whether we should this?

I like this patch, and think we should use it.  The IDE should be
re-fetching the register names, but we don't need to be mean about it :-)

-- 
Daniel Jacobowitz
CodeSourcery

Re: [patch] Re: -data-list-register-names regression?

[Joel Brobecker]

> > Tested on powerpc-linux.  Thoughts on whether we should this?
> 
> I like this patch, and think we should use it.  The IDE should be
> re-fetching the register names, but we don't need to be mean about it :-)

Ditto. Do we want it for 7.1?

-- 
Joel

Re: [patch] Re: -data-list-register-names regression?

[Dave Korn]

On 17/03/2010 22:59, Joel Brobecker wrote:
>>> Tested on powerpc-linux.  Thoughts on whether we should this?
>> I like this patch, and think we should use it.  The IDE should be
>> re-fetching the register names, but we don't need to be mean about it :-)
> 
> Ditto. Do we want it for 7.1?

  It would be helpful, there's a major Eclipse release coming up and there
might not be much time for 7.1.1 (or 7.2) before then.  It would be good if
the 7.1+Eclipse combination could get some solid testing before the launch
date in June.

    cheers,
      DaveK

Re: [patch] Re: -data-list-register-names regression?

[Ulrich Weigand]

Joel Brobecker wrote:
> > > Tested on powerpc-linux.  Thoughts on whether we should this?
> > 
> > I like this patch, and think we should use it.  The IDE should be
> > re-fetching the register names, but we don't need to be mean about it :-)
> 
> Ditto. Do we want it for 7.1?

OK, I've checked this in to mainline now.  It seems reasonable to me
to put this in 7.1 as well, but I think you have final call on that ...
If you'd like me to put in, please let me know.

Bye,
Ulrich

-- 
  Dr. Ulrich Weigand
  GNU Toolchain for Linux on System z and Cell BE
  Ulrich.Weigand@de.ibm.com

Re: [patch] Re: -data-list-register-names regression?

[Joel Brobecker]

> OK, I've checked this in to mainline now.  It seems reasonable to me
> to put this in 7.1 as well, but I think you have final call on that ...
> If you'd like me to put in, please let me know.

OK, if we both agree that it's safe enough to go into 7.1, then let's
apply it on the branch as well.

Thanks!
-- 
Joel

Re: [patch] Re: -data-list-register-names regression?

[Ulrich Weigand]

Joel Brobecker wrote:
> > OK, I've checked this in to mainline now.  It seems reasonable to me
> > to put this in 7.1 as well, but I think you have final call on that ...
> > If you'd like me to put in, please let me know.
> 
> OK, if we both agree that it's safe enough to go into 7.1, then let's
> apply it on the branch as well.

Now checked in to the branch as well.

Thanks,
Ulrich

-- 
  Dr. Ulrich Weigand
  GNU Toolchain for Linux on System z and Cell BE
  Ulrich.Weigand@de.ibm.com