testsuite: Don't use expect_background to reap gdbserver
authorPedro Alves <palves@redhat.com>
Tue, 7 Apr 2015 17:19:30 +0000 (18:19 +0100)
committerPedro Alves <palves@redhat.com>
Tue, 7 Apr 2015 17:27:55 +0000 (18:27 +0100)
commit6423214fcb912a5d402ffcb84d6f61f2b900a900
tree38f9b12a82f53186e18d249004995ea44c4612c1
parent71c0ee8cb9d7d77e8b480aaad715cc5343737993
testsuite: Don't use expect_background to reap gdbserver

I adjusted a test to do 'expect -i $server_spawn_id -re ...', and saw
really strange behavior.  Whether that expect would work, depended on
whether GDB would also send output and the same expect matched it too
(on $gdb_spawn_id).  I was perplexed until I noticed that
gdbserver_spawn spawns gdbserver and then uses expect_background to
reap gdbserver.  That expect_background conflicts/races with any
"expect -i $server_spawn_id" done anywhere else in parallel...

In order to make it possible for tests to read inferior I/O out of
$server_spawn_id, we to get rid of that expect_background.  This patch
makes us instead reap gdbserver's spawn id when GDB exits.  If GDB is
still around, this gives a chance for gdbserver to exit cleanly.  The
current code in gdb_finish uses "kill", but that doesn't work with
extended-remote (gdbserver doesn't exit).  We now use "monitor exit"
instead which works in both remote and extended-remote modes.

gdb/testsuite/ChangeLog:
2015-04-07  Pedro Alves  <palves@redhat.com>

* lib/gdb.exp (gdb_finish): Delete persistent gdbserver handling.
* lib/gdbserver-support.exp (gdbserver_start): Make
$server_spawn_id global.
(gdbserver_start): Don't wait for gdbserver's spawn id with
expect_background.
(close_gdbserver): New procedure.
(gdb_exit): Rename the default version and reimplement.
gdb/testsuite/ChangeLog
gdb/testsuite/lib/gdb.exp
gdb/testsuite/lib/gdbserver-support.exp