Don't try to report resumed thread it the thread list is empty.

Don't try to report resumed thread it the thread list is empty.

[Vladimir Prus]

[-- Attachment #1: Type: text/plain, Size: 158 bytes --]


I've checked in the below to make GDB not assert in MI mode on
targets where single-threaded programs have zero threads in the
GDB thread table.

- Volodya


[-- Attachment #2: on_resume.diff --]
[-- Type: text/x-diff, Size: 1332 bytes --]

Index: gdb/ChangeLog
===================================================================
RCS file: /cvs/src/src/gdb/ChangeLog,v
retrieving revision 1.9517
diff -u -p -r1.9517 ChangeLog
--- gdb/ChangeLog	5 Jul 2008 13:48:20 -0000	1.9517
+++ gdb/ChangeLog	5 Jul 2008 17:56:47 -0000
@@ -1,3 +1,8 @@
+2008-07-05  Vladimir Prus  <vladimir@codesourcery.com>
+
+	* mi/mi-interp.c (mi_on_resume): Don't try to report
+	resumed thread it the thread list is empty.
+	
 2008-07-05  Pierre Muller  <muller@ics.u-strasbg.fr>
 
 	* cli/cli-decode.c (add_setshow_optional_filename_cmd): Set
Index: gdb/mi/mi-interp.c
===================================================================
RCS file: /cvs/src/src/gdb/mi/mi-interp.c,v
retrieving revision 1.34
diff -u -p -r1.34 mi-interp.c
--- gdb/mi/mi-interp.c	25 Jun 2008 15:15:42 -0000	1.34
+++ gdb/mi/mi-interp.c	5 Jul 2008 17:56:47 -0000
@@ -338,6 +338,12 @@ mi_on_resume (ptid_t ptid)
 
   if (PIDGET (ptid) == -1)
     fprintf_unfiltered (raw_stdout, "*running,thread-id=\"all\"\n");
+  else if (thread_count () == 0)
+    {
+      /* This is a target where for single-threaded programs the thread
+	 table has zero threads.  Don't print any thread-id field.  */
+      fprintf_unfiltered (raw_stdout, "*running\n");
+    }
   else
     {
       struct thread_info *ti = find_thread_pid (ptid);

Re: Don't try to report resumed thread it the thread list is empty.

[Pedro Alves]

A Saturday 05 July 2008 18:58:29, Vladimir Prus wrote:
> I've checked in the below to make GDB not assert in MI mode on
> targets where single-threaded programs have zero threads in the
> GDB thread table.
>

FYI, running the MI testsuite on a target that doesn't register
the main thread in ST programs, has failures that look like:

-exec-continue
^running
*running
(gdb) 
*stopped,reason="breakpoint-hit",disp="keep",bkptno="2",thread-id="0",frame={addr="0x000087e0",func="do_special_tests",args=[],file="../../../src/gdb/testsuite/gdb.mi/var-cmd.c",fullname="/home/pedro/gdb/monitor_always_a_thread/src/gdb/testsuite/gdb.mi/var-cmd.c",line="319"}
(gdb) 
FAIL: gdb.mi/mi2-var-display.exp: continue to do_special_tests (failed to 
resume)

I believe the issue is that in lib/mi-support.exp, *running matching assumes
there is -thread-id field
 
-re "220\\^running\[\r\n\]+\\*running,thread-id=\"\[^\"\]+\"\r\n$mi_gdb_prompt$" 
{}

I notice that there are other places that are outputing -thread-id="0", e.g.:

*stopped,reason="breakpoint-hit",disp="keep",bkptno="2",thread-id="0",frame={addr="0x000087e0",func="do_special_tests",args=[],file="../../../src/gdb/testsuite/gdb.mi/var-cmd.c",fullname="/home/pedro/gdb/monitor_always_a_thread/src/gdb/testsuite/gdb.mi/var-cmd.c",line="319"}
(gdb) 

So maybe we should make this consistent, and output -thread-id="0"
in *running as well?

-- 
Pedro Alves

Re: Don't try to report resumed thread it the thread list is empty.

[Vladimir Prus]

On Wednesday 09 July 2008 14:31:21 Pedro Alves wrote:
> A Saturday 05 July 2008 18:58:29, Vladimir Prus wrote:
> > I've checked in the below to make GDB not assert in MI mode on
> > targets where single-threaded programs have zero threads in the
> > GDB thread table.
> >
> 
> FYI, running the MI testsuite on a target that doesn't register
> the main thread in ST programs, 

For the record, which target did you test on?

> has failures that look like: 
> 
> -exec-continue
> ^running
> *running
> (gdb) 
> *stopped,reason="breakpoint-hit",disp="keep",bkptno="2",thread-id="0",frame={addr="0x000087e0",func="do_special_tests",args=[],file="../../../src/gdb/testsuite/gdb.mi/var-cmd.c",fullname="/home/pedro/gdb/monitor_always_a_thread/src/gdb/testsuite/gdb.mi/var-cmd.c",line="319"}
> (gdb) 
> FAIL: gdb.mi/mi2-var-display.exp: continue to do_special_tests (failed to 
> resume)
> 
> I believe the issue is that in lib/mi-support.exp, *running matching assumes
> there is -thread-id field
>  
> -re "220\\^running\[\r\n\]+\\*running,thread-id=\"\[^\"\]+\"\r\n$mi_gdb_prompt$" 
> {}
> 
> I notice that there are other places that are outputing -thread-id="0", e.g.:
> 
> *stopped,reason="breakpoint-hit",disp="keep",bkptno="2",thread-id="0",frame={addr="0x000087e0",func="do_special_tests",args=[],file="../../../src/gdb/testsuite/gdb.mi/var-cmd.c",fullname="/home/pedro/gdb/monitor_always_a_thread/src/gdb/testsuite/gdb.mi/var-cmd.c",line="319"} 
> (gdb) 
> 
> So maybe we should make this consistent, and output -thread-id="0"
> in *running as well?

I'm not sure which way we want to make it consistent. Outputiing "thread-id" 
with a "I don't exist" value of thread id seems a little bit distasteful.
I guess I can modify regexp to make thread-id optional.

- Volodya

Re: Don't try to report resumed thread it the thread list is empty.

[Pedro Alves]

A Wednesday 09 July 2008 13:04:11, Vladimir Prus wrote:
> On Wednesday 09 July 2008 14:31:21 Pedro Alves wrote:
> > A Saturday 05 July 2008 18:58:29, Vladimir Prus wrote:
> > > I've checked in the below to make GDB not assert in MI mode on
> > > targets where single-threaded programs have zero threads in the
> > > GDB thread table.
> >
> > FYI, running the MI testsuite on a target that doesn't register
> > the main thread in ST programs,
>
> For the record, which target did you test on?

arm-elf with target sim.  I've checked in the patch to make it
always have a thread, but it's just a matter of commenting out
the add_thread_silent call to get the old behaviour.

> I'm not sure which way we want to make it consistent. Outputiing
> "thread-id" with a "I don't exist" value of thread id seems a little bit
> distasteful. I guess I can modify regexp to make thread-id optional.

I meant either output -thread-id="0" everywhere, or don't output -thread=id
at all, nowhere.

-- 
Pedro Alves

Re: Don't try to report resumed thread it the thread list is empty.

[Pedro Alves]

A Wednesday 09 July 2008 13:52:28, Pedro Alves wrote:
> A Wednesday 09 July 2008 13:04:11, Vladimir Prus wrote:

> > I'm not sure which way we want to make it consistent. Outputiing
> > "thread-id" with a "I don't exist" value of thread id seems a little bit
> > distasteful. I guess I can modify regexp to make thread-id optional.
>
> I meant either output -thread-id="0" everywhere, or don't output -thread=id
> at all, nowhere.

To be clear, I don't care that much about this inconsistency, what
I really want to have all targets register at least a thread, so
modifying the regexp is also fine with me, FWIW.  Although I thought
that changing the code in this case would be a less invasive change, than
relaxing the testsuite.

-- 
Pedro Alves