From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([2001:4830:134:3::10]:58109) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1VCUi7-0005N1-S9 for qemu-devel@nongnu.org; Thu, 22 Aug 2013 09:14:06 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1VCUi1-0001RK-Qz for qemu-devel@nongnu.org; Thu, 22 Aug 2013 09:13:59 -0400 Received: from mx1.redhat.com ([209.132.183.28]:42484) by eggs.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1VCUi1-0001RD-JF for qemu-devel@nongnu.org; Thu, 22 Aug 2013 09:13:53 -0400 Date: Thu, 22 Aug 2013 09:12:48 -0400 From: Luiz Capitulino Message-ID: <20130822091248.19ad2e2a@redhat.com> In-Reply-To: <5215D6E7.30708@linux.vnet.ibm.com> References: <1374808842-11051-1-git-send-email-xiawenc@linux.vnet.ibm.com> <20130730120311.7f1e44a7@redhat.com> <20130820100442.1abc070d@redhat.com> <5215D6E7.30708@linux.vnet.ibm.com> Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: quoted-printable Subject: Re: [Qemu-devel] [PATCH V8 00/13] monitor: support sub command group in auto completion and help List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: Wenchao Xia Cc: pbonzini@redhat.com, armbru@redhat.com, qemu-devel@nongnu.org On Thu, 22 Aug 2013 17:16:23 +0800 Wenchao Xia wrote: > =E4=BA=8E 2013-8-20 22:04, Luiz Capitulino =E5=86=99=E9=81=93: > > On Tue, 30 Jul 2013 12:03:11 -0400 > > Luiz Capitulino wrote: > > > >> On Fri, 26 Jul 2013 11:20:29 +0800 > >> Wenchao Xia wrote: > >> > >>> This series make auto completion and help functions works normal for = sub > >>> command, by using reentrant functions. In order to do that, global va= riables > >>> are not directly used in those functions any more. With this series, = cmd_table > >>> is a member of structure Monitor so it is possible to create a monito= r with > >>> different command table now, auto completion will work in that monito= r. In > >>> short, "info" is not treated as a special case now, this series ensur= e help > >>> and auto complete function works normal for any sub command added in = the future. > >>> > >>> Patch 5 replaced cur_mon with rs->mon, it is safe because: > >>> monitor_init() calls readline_init() which initialize mon->rs, result= is > >>> mon->rs->mon =3D=3D mon. Then qemu_chr_add_handlers() is called, whic= h make > >>> monitor_read() function take *mon as its opaque. Later, when user inp= ut, > >>> monitor_read() is called, where cur_mon is set to *mon by "cur_mon = =3D opaque". > >>> If qemu's monitors run in one thread, then later in readline_handle_b= yte() > >>> and readline_comletion(), cur_mon is actually equal to rs->mon, in an= other > >>> word, it points to the monitor instance, so it is safe to replace *cu= r_mon > >>> in those functions. > >> > >> I've applied this to qmp-next with the change I suggested for > >> patch 09/13. > > > > Unfortunately this series brakes make check: > > > > GTESTER check-qtest-x86_64 > > Broken pipe > > GTester: last random seed: R02S3492bd34f44dd17460851643383be44d > > main-loop: WARNING: I/O thread spun for 1000 iterations > > make: *** [check-qtest-x86_64] Error 1 > > > > I debugged it (with some help from Laszlo) and the problem is that it > > broke the human-monitor-command command. Any usage of this command > > triggers the bug like: > > > > { "execute": "human-monitor-command", > > "arguments": { "command-line": "info registers" } } > > > > It seems simple to fix, I think you just have to initialize > > mon->cmd_table in qmp_human_monitor_command(), but I'd recommend two > > things: > > > > 1. It's better to split off some/all QMP initialization from > > monitor_init() and call it from qmp_human_monitor_command() > > > > 2. Can you please take the opportunity and test all commands using > > cur_mon? Just grep for it > > > > Sorry for noticing this only now, but I only run make check before > > sending a pull request (although this very likely shows you didn't > > run it either). > > > About the fd related qmp interface, to test it, send_msg() is needed, > which was not supported in python 2, but new added python 3.3. I think > there are three ways to add test cases for fd qmp APIs: > 1 test only when python > 3.3. > 2 python plus C: compile a .so and call it with ctypes. > 3 a new test framework: pure C code to call qmp interfaces. > Which one do you prefer? Can't we have a C program plus a shell script to test this? Anyway, if this gets complicated you can skip having the test-case. This series took a long way already and holding it because of that test-case isn't fair.