Export QMP functions

[bensmrs]

Closes: #1226

[stgraber]

@bensmrs I wonder, wouldn't it be cleaner if we just added a single command, run_command which would take one un-named argument (the command) and treat all kwargs as the arguments?

So we'd then have run_qmp that handles the raw JSON in/out QMP and run_command that lets you use a nicer syntax like:

run_command("qom-get", path="/machine/peripheral/dev-incus_root", property="serial")

I'm basically trying to avoid us having to duplicate the entire QMP command list in here.

[stgraber]

Actually, I think I'd be willing to make a bit of a compromise because I do like the shorter syntax when you don't need to name all the arguments, but I still don't want to set a precedent which will get us to have to implement every single QMP command.

So I think we may want to go with:

Generic:

• run_qmp(JSON)
• run_command(CMD, key=value...)

Specific:

• qom_{get,list,set}
• object_{add,del}
• device_{add,del}
• blockdev_{add,del}
• chardev_{add,change,remove}
• netdev_{add,del}

Those are all modern QMP commands for core objects so not terribly likely to change or go away and the most likely to be called from the hook.

The other more niche commands can go through run_command where for commands with no arguments (quit/system_reset), this will remain pretty nice and short.

[bensmrs]

I expected the remark but I agree with your last message, I’ll look at it tomorrow

[stgraber]

Thanks!

[bensmrs]

Just for a bit of context, I’m exploring the call stack to get the position in the Starlark code where there is a problem (in big scriptlets, it’s pretty useful). I haven’t found any other way, and Starlark doesn’t seem to offer much introspection into AST nodes, so that should do the trick…