Re: [PATCH 4/5] python: qmp_shell: add -e/--exit-on-error option

[Damien Hedde]

On 2/23/22 17:43, Damien Hedde wrote:
> On 2/23/22 16:44, Daniel P. Berrangé wrote:
> > On Wed, Feb 23, 2022 at 10:41:11AM -0500, John Snow wrote:
> > > On Wed, Feb 23, 2022 at 10:27 AM Daniel P. Berrangé 
> > > <berrange@redhat.com> wrote:
> > > > On Wed, Feb 23, 2022 at 10:22:11AM -0500, John Snow wrote:
> > > > > On Mon, Feb 21, 2022 at 10:55 AM Damien Hedde
> > > > > <damien.hedde@greensocs.com> wrote:
> > > > > > This option makes qmp_shell exit (with error code 1)
> > > > > > as soon as one of the following error occurs:
> > > > > > + command parsing error
> > > > > > + disconnection
> > > > > > + command failure (response is an error)
> > > > > >
> > > > > > _execute_cmd() method now returns None or the response
> > > > > > so that read_exec_command() can do the last check.
> > > > > >
> > > > > > This is meant to be used in combination with an input file
> > > > > > redirection. It allows to store a list of commands
> > > > > > into a file and try to run them by qmp_shell and easily
> > > > > > see if it failed or not.
> > > > > >
> > > > > > Signed-off-by: Damien Hedde <damien.hedde@greensocs.com>
> > > > >
> > > > > Based on this patch, it looks like you really want something
> > > > > scriptable, so I think the qemu-send idea that Dan has suggested might
> > > > > be the best way to go. Are you still hoping to use the interactive
> > > > > "short" QMP command format? That might be a bad idea, given how flaky
> > > > > the parsing is -- and how we don't actually have a published standard
> > > > > for that format. We've *never* liked the bad parsing here, so I have a
> > > > > reluctance to use it in more places.
>
> I don't really care of the command format. I was just using the one 
> expected by qmp-shell to avoid providing another script.
> I think it's better not to use some standard syntax like json.
I wanted to say the opposite: it's best to use json.
> As long as we can store the commands into a file and tests them easily, 
> it's ok. The crucial feature is the "stop as soon something unexpected 
> happens" so that we can easily spot an issue.
> > > > > I'm having the naive idea that a script file could be as simple as a
> > > > > list of QMP commands to send:
> > > > >
> > > > > [
> > > > >      {"execute": "block-dirty-bitmap-add", "arguments": { ... }},
> > > > >      ...
> > > > > ]
>
> I used this format at some point because it's so trivial to feed into 
> the QMP tools. Even used a yaml version of that to get the "human 
> readability" that goes with it.
>
> > > > I'd really recommend against creating a new format for the script
> > > > file, especially one needing opening & closing  [] like this, as
> > > > that isn't so amenable to dynamic usage/creation. ie you can't
> > > > just append an extcra command to an existing file.
> > > >
> > > > IMHO, the "file" format should be identical to the result of
> > > > capturing the socket data off the wire. ie just a concatenation
> > > > of QMP commands, with no extra wrapping / change in format.
> > > > >>
> > > Eugh. That's just so hard to parse, because there's no off-the-shelf
> > > tooling for "load a sequence of JSON documents". Nothing in Python
> > > does it. :\
> >
> > It isn't that hard if you require each JSON doc to be followed by
> > a newline.
> >
> > Feed one line at a time to the JSON parser, until you get a complete
> > JSON doc, process that, then re-init the parser and carry on feeding
> > it lines until it emits the next JSON doc, and so on.
>
> I agree it's doable. I can look into that.
>
> It makes me think that I've managed to modify the chardev 'file' backend
> a few months ago so that it can be used with an input file on the cli. 
> This allowed to give such raw qmp file directly with the -qmp option 
> instead of using an intermediate socket and a script issuing the same file.
> But I gave up with this approach because then it can't stop if a command 
> failed without hacking into the receiving side in qemu.
>
> --
> Damien