]> git.ipfire.org Git - thirdparty/binutils-gdb.git/commit
Centralize "[Thread ...exited]" notifications
authorPedro Alves <pedro@palves.net>
Tue, 21 Jun 2022 18:30:48 +0000 (19:30 +0100)
committerPedro Alves <pedro@palves.net>
Mon, 12 Dec 2022 20:01:20 +0000 (20:01 +0000)
commitdaf510bf56541306c5904af87b37385f7de0933c
tree5086cc63281e879b9618fb031fad16a163e92261
parent657a94e242485e6457215763e5c1055502ca6a79
Centralize "[Thread ...exited]" notifications

Currently, each target backend is responsible for printing "[Thread
...exited]" before deleting a thread.  This leads to unnecessary
differences between targets, like e.g. with the remote target, we
never print such messages, even though we do print "[New Thread ...]".

E.g., debugging the gdb.threads/attach-many-short-lived-threads.exp
with gdbserver, letting it run for a bit, and then pressing Ctrl-C, we
currently see:

 (gdb) c
 Continuing.
 ^C[New Thread 3850398.3887449]
 [New Thread 3850398.3887500]
 [New Thread 3850398.3887551]
 [New Thread 3850398.3887602]
 [New Thread 3850398.3887653]
 ...

 Thread 1 "attach-many-sho" received signal SIGINT, Interrupt.
 0x00007ffff7e6a23f in __GI___clock_nanosleep (clock_id=clock_id@entry=0, flags=flags@entry=0, req=req@entry=0x7fffffffda80, rem=rem@entry=0x7fffffffda80)
     at ../sysdeps/unix/sysv/linux/clock_nanosleep.c:78
 78      in ../sysdeps/unix/sysv/linux/clock_nanosleep.c
 (gdb)

Above, we only see "New Thread" notifications, even though threads
were deleted.

After this patch, we'll see:

 (gdb) c
 Continuing.
 ^C[Thread 3558643.3577053 exited]
 [Thread 3558643.3577104 exited]
 [Thread 3558643.3577155 exited]
 [Thread 3558643.3579603 exited]
 ...
 [New Thread 3558643.3597415]
 [New Thread 3558643.3600015]
 [New Thread 3558643.3599965]
 ...

 Thread 1 "attach-many-sho" received signal SIGINT, Interrupt.
 0x00007ffff7e6a23f in __GI___clock_nanosleep (clock_id=clock_id@entry=0, flags=flags@entry=0, req=req@entry=0x7fffffffda80, rem=rem@entry=0x7fffffffda80)
     at ../sysdeps/unix/sysv/linux/clock_nanosleep.c:78
 78      in ../sysdeps/unix/sysv/linux/clock_nanosleep.c
 (gdb) q

This commit fixes this by moving the thread exit printing to common
code instead, triggered from within delete_thread (or rather,
set_thread_exited).

There's one wrinkle, though.  While most targest want to print:

 [Thread ... exited]

the Windows target wants to print:

 [Thread ... exited with code <exit_code>]

... and sometimes wants to suppress the notification for the main
thread.  To address that, this commits adds a delete_thread_with_code
function, only used by that target (so far).

Change-Id: I06ec07b7c51527872a9713dd11cf7867b50fc5ff
14 files changed:
gdb/annotate.c
gdb/breakpoint.c
gdb/fbsd-nat.c
gdb/gdbthread.h
gdb/inferior.c
gdb/inferior.h
gdb/linux-nat.c
gdb/mi/mi-interp.c
gdb/netbsd-nat.c
gdb/observable.h
gdb/procfs.c
gdb/python/py-inferior.c
gdb/thread.c
gdb/windows-nat.c