Re: [PATCH 1/2] mi: Restore original thread/frame when specifying --thread or --thread-group

[Pedro Alves <palves at redhat dot com>]

On 08/02/2016 11:14 PM, Simon Marchi wrote:
> On 16-08-02 01:38 PM, Simon Marchi wrote:

>> Ok, I kinda had the same design idea, but it was blurry.  Now that you explain it
>> (I didn't know what previous_inferior_ptid was), it's clear.  I'll try to prototype it
>> quickly to see if it's viable.
> 
> I started to prototype what you described, but it's definitely not a small job.  It would
> also be too big of a change to put in 7.12.  I'd be really interested to work on that for
> the next development cycle though.
> 
> In the mean time, here is a new version of the patch that implements what I had described
> (with the global flag), which I think is less risky.

I think you need to consider infcalls, and canned sequence of commands
Python/Scheme, etc., basically any place that sets ui->async = 0 ...

guile/guile.c:169:  current_ui->async = 0;
guile/guile.c:202:  current_ui->async = 0;
guile/guile.c:332:      current_ui->async = 0;
guile/scm-ports.c:521:  current_ui->async = 0;
top.c:700:  current_ui->async = 0;
compile/compile.c:95:  current_ui->async = 0;
compile/compile.c:137:  current_ui->async = 0;
compile/compile.c:191:  current_ui->async = 0;
infcall.c:586:  current_ui->async = 0;
python/python.c:326:  current_ui->async = 0;
python/python.c:471:  current_ui->async = 0;
python/python.c:655:      current_ui->async = 0;
cli/cli-script.c:384:  current_ui->async = 0;
cli/cli-script.c:666:  current_ui->async = 0;
cli/cli-script.c:690:  current_ui->async = 0;
cli/cli-script.c:1697:  current_ui->async = 0;

... because these will only return when execution next stops,
so the cleanup runs after, and undoes the thread change.

E.g., this infcall stops due to a breakpoint on another
thread, but MI still reverts back to the "--thread" thread:

 (gdb)
 b usleep
 [...]
 ^done
 (gdb)

 thread 1
 [...]
 ^done
 =thread-selected,id="1"
 (gdb) 

 -interpreter-exec --thread 1 console "print sleep (10)"
 [...]
 ~"[Switching to Thread 0x7ffff6ffb700 (LWP 7818)]\n"
 ~"\n"
 ~"Thread 3 \"threads\" hit Breakpoint 2, usleep (useconds=1) at ../sysdeps/posix/usleep.c:26\n"
 ~"26\t  struct timespec ts = { .tv_sec = (long int) (useconds / 1000000),\n"
 *stopped,reason="breakpoint-hit",disp="keep",bkptno="2",frame={addr="0x00007ffff78f6030",func="usleep",args=[{name="useconds",value="1"}],file="../sysdeps/posix/usleep.c",fullname="/usr/src/debug/glibc-2.22/sysdeps/posix/usleep.c",line="26"},thread-id="3",stopped-threads="all",core="2"
 &"The program stopped in another thread while making a function call from GDB.\n"
 &"Evaluation of the expression containing the function\n"
 &"(__sleep) will be abandoned.\n"
 &"When the function is done executing, GDB will silently stop.\n"
 ^error,msg="The program stopped in another thread while making a function call from GDB.\nEvaluation of the expression containing the function\n(__sleep) will be abandoned.\nWhen the function is done executing, GDB will silently stop."
 (gdb)

 thread
 [...]
 ~"[Current thread is 1 (Thread 0x7ffff7fc8700 (LWP 7813))]\n"
 ^done
 (gdb) 

Above should have been 3.

E.g., this user-defined command starts the program and stops in
thread 2.1, but MI still reverts back to the "--thread" thread:

 (gdb) define foo
 ~">" inferior 2
 ~">" start
 ~">" end
 (gdb)

 info inferiors
 ~"  Num  Description       Executable        \n"
 ~"* 1    process 8216      /home/pedro/brno/pedro/gdb/tests/threads \n"
 ~"  2    <null>            /home/pedro/brno/pedro/gdb/tests/threads \n"
 ^done
 (gdb)

 info threads
 ~"  Id   Target Id         Frame \n"
 ~"* 1.1  Thread 0x7ffff7fc8700 (LWP 8216) \"threads\" main () at threads.c:54\n"
 ^done
 (gdb)

 -interpreter-exec --thread 1 console "foo"
 ~"[Switching to inferior 2 [<null>] (/home/pedro/brno/pedro/gdb/tests/threads)]\n"
 [...]
 ~"\n"
 ~"Thread 2.1 \"threads\" hit Temporary breakpoint 3, main () at threads.c:54\n"
 ~"54\t    long i = 0;\n"
 *stopped,reason="breakpoint-hit",disp="del",bkptno="3",frame={addr="0x000000000040076e",func="main",args=[],file="threads.c",fullname="/home/pedro/brno/pedro/gdb/tests/threads.c",line="54"},thread-id="3",stopped-threads="all",core="7"
 =breakpoint-deleted,id="3"
 (gdb)

 thread
 &"thread\n"
 ~"[Current thread is 1.1 (Thread 0x7ffff7fc8700 (LWP 8216))]\n"
 ^done
 (gdb) 


 While at it, "inferior 2" is a command that explicitly changes
 the thread, but misses setting the new flag:

-interpreter-exec --thread 1 console "inferior 2"
~"[Switching to inferior 2 [process 8603] (/home/pedro/brno/pedro/gdb/tests/threads)]\n"
~"[Switching to thread 2.1 (Thread 0x7ffff7fc8700 (LWP 8603))] \n"
~"#0  main () at threads.c:54\n"
~"54\t    long i = 0;\n"
^done
(gdb) 

 thread 
 &"thread \n"
 ~"[Current thread is 1.1 (Thread 0x7ffff7fc8700 (LWP 8599))]\n"
 ^done
 (gdb) 


It may be this requires setting the flag in most of the places
that we'd set the user-selected thread in the other approach,
not sure.

Thanks,
Pedro Alves