[Bug gdb/17503] New: "mt set target-async on" breaks post hook routines

[Bug gdb/17503] New: "mt set target-async on" breaks post hook routines

[dje at google dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=17503

            Bug ID: 17503
           Summary: "mt set target-async on" breaks post hook routines
           Product: gdb
           Version: HEAD
            Status: NEW
          Severity: normal
          Priority: P2
         Component: gdb
          Assignee: unassigned at sourceware dot org
          Reporter: dje at google dot com

The current way we implement post hook routines does not work with mt set
target-async on.

The hook is called long before the command actually completes, and in
particular before the current thread is marked as no longer executing, which
breaks one of my user's post hooks.

Testcase will be attached.
Below are sessions with target-async = off/on.

Repro steps:

bash$ gcc -g hookpost-step.c hookpost-step-2.c -o hookpost-step.x64
bash$ gdb hookpost-step.x64
(gdb) file hookpost-step.x64
(gdb) source hookpost-step.gdbinit
(gdb) # set debug infrun 1
(gdb) # set debug lin-lwp 1
(gdb) # mt set target-async off or on
(gdb) start
(gdb) step

Here's the session with target-async = off.

---snip-target-async-off---
(gdb) start
Temporary breakpoint 1 at 0x4004f1: file hookpost-step.c, line 7.
Starting program: /home/dje/src/play/hookpost-step.x64

Temporary breakpoint 1, main () at hookpost-step.c:7
7         foo ();
(gdb) step
foo () at hookpost-step-2.c:4
4       }
hookpost-step called ...
(gdb) finish
main () at hookpost-step.c:8
8         return 0;
----
(gdb) step
9       }
---- ^^ skipped 1 functions ^^ ----
#0  main () at hookpost-step.c:9
9       }
(gdb) 
---end-snip-target-async-off---

Here's the session with target-async = on.

---snip-target-async-on---
(gdb) start
Temporary breakpoint 2 at 0x4004f1: file hookpost-step.c, line 7.
Starting program: /home/dje/src/play/hookpost-step.x64

Temporary breakpoint 2, main () at hookpost-step.c:7
7         foo ();
(gdb) step
hookpost-step called ...
Traceback (most recent call last):
  File "<string>", line 3, in <module>
gdb.error: No frame is currently selected.
Error while executing Python code.
(gdb) foo () at hookpost-step-2.c:4
4       }
---end-snip-target-async-on---

Also note that in the target-async on case we got the (gdb) prompt and then
further output was produced.

Turning infrun and lin-lwp debug flags on provides more information.
In the "target-async = off" session, the hook is called after
gdb notices we've stepped into a subroutine (which is as expected).

---snip---
(gdb) step
infrun: clear_proceed_status_thread (process 13427)
infrun: proceed (addr=0xffffffffffffffff, signal=GDB_SIGNAL_DEFAULT, step=1)
infrun: resume (step=1, signal=GDB_SIGNAL_0), trap_expected=0, current thread
[process 13427] at 0x4004f1
LLR: Preparing to step process 13427, 0, inferior_ptid process 13427
LLR: PTRACE_SINGLESTEP process 13427, 0 (resume event thread)
infrun: wait_for_inferior ()
[... lots of output elided ...]
infrun:   status->kind = stopped, signal = GDB_SIGNAL_TRAP
infrun: TARGET_WAITKIND_STOPPED
infrun: stop_pc = 0x400506
infrun: BPSTAT_WHAT_STEP_RESUME
infrun: stepped into subroutine
infrun: stop_waiting
foo () at hookpost-step-2.c:4
4       }
hookpost-step called ...
---end-snip---

In the "target-async = on" session, the hook is called after the first
PTRACE_SINGLESTEP.  Yikes.

---snip---
(gdb) step
infrun: clear_proceed_status_thread (process 13539)
infrun: proceed (addr=0xffffffffffffffff, signal=GDB_SIGNAL_DEFAULT, step=1)
infrun: resume (step=1, signal=GDB_SIGNAL_0), trap_expected=0, current thread
[process 13539] at 0x4004f1
LLR: Preparing to step process 13539, 0, inferior_ptid process 13539
LLR: PTRACE_SINGLESTEP process 13539, 0 (resume event thread)
sigchld
hookpost-step called ...
Traceback (most recent call last):
  File "<string>", line 3, in <module>
gdb.error: No frame is currently selected.
Error while executing Python code.
(gdb) linux_nat_wait: [process -1], [TARGET_WNOHANG]
LLW: enter
LNW: waitpid(-1, ...) returned 13539, No child processes
LLW: waitpid 13539 received Trace/breakpoint trap (stopped)
LLW: Candidate event Trace/breakpoint trap (stopped) in process 13539.
[... lots of output elided ...]
---end-snip---

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug gdb/17503] "mt set target-async on" breaks post hook routines

[dje at google dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=17503

--- Comment #1 from dje at google dot com ---
Created attachment 7840
  --> https://sourceware.org/bugzilla/attachment.cgi?id=7840&action=edit
testcase part 1

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug gdb/17503] "mt set target-async on" breaks post hook routines

[dje at google dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=17503

--- Comment #2 from dje at google dot com ---
Created attachment 7841
  --> https://sourceware.org/bugzilla/attachment.cgi?id=7841&action=edit
testcase part 2

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug gdb/17503] "mt set target-async on" breaks post hook routines

[dje at google dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=17503

--- Comment #3 from dje at google dot com ---
Created attachment 7842
  --> https://sourceware.org/bugzilla/attachment.cgi?id=7842&action=edit
testcase gdbinit

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug gdb/17503] "mt set target-async on" breaks post hook routines

[dje at google dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=17503

--- Comment #4 from dje at google dot com ---
I'm thinking there's no need to block 7.8.1 for this.
A workaround is to do "mt set target-async off".

-- 
You are receiving this mail because:
You are on the CC list for the bug.