Re: [PATCH 2/9] qapi: Mark unstable QMP parts with feature 'unstable'

On Tue, Oct 26, 2021 at 3:56 AM Markus Armbruster <armbru(a)redhat.com&gt; wrote: Yeah, that's the part I don't like. If this isn't the first instance of breaking the seal, though, this is the wrong time for me to comment on it. Don't worry about it for this series. (Tangent below, unrelated to acceptance of this series) Yeah, I don't have an appetite for it at the moment either. I'll have to read Marc-Andre's recent sphinx patches and see if I want to do more work -- I had a bigger prototype a few months back I didn't bring all the way home, but I am still thinking about reworking our QAPIDoc format. It's ... well. I don't really want to, but I am not sure how else to bring some of the features I want home, and I think I need some more time to think carefully through exactly what I want to do and why. I wouldn't mind chatting about it with you sometime soon -- there's a few things to balance here, such as: (1) Reworking the doc format would be an obnoxious amount of churn, ... (2) ...but the Sphinx internals are really not meant to be used the way we're using them right now, ... (3) ...but I also don't want to write our QAPI docstrings in a way that *targets* Sphinx. Not that I think we'll be dropping it any time soon, but it still feels like the wrong idea to tie them so closely together. I'm hoping there's an opportunity to make the parsing nicer with minimal changes to the parsing and format, though. It just might require enforcing a *pinch* more structure. I could see how I feel about per-field annotations at that point. I consider them interesting for things like the Python SDK where I may want to enable/disable warnings for using deprecated stuff at the client-level. (e.g., let's say you're using Python SDK 6.2 to talk to a 6.1 client. Nothing stops you from doing this, but some commands will simply be rejected by QEMU as it won't know what you're talking about. Using deprecated fields or commands to talk to an older client will produce no visible warning from QEMU either, as it wasn't deprecated at that point in time. Still, the client may wish to know that they're asking for future trouble -- so it's a thought that client-side warnings have some play here.) Ehhhhhhhhhhhhhhhh don't worry about it for now, I guess I'll plow forward a bit with the Sphinx stuff I'm thinking of at some point Soon:tm: and worry about where the immovable objects are when I get there. This is foot-based landmine-detection, and it works 100% of the time. Yeh, just explaining why it's not an R-B. I'm trying to be a bit better about reviews by not forcing myself to do "all or nothing". I trust your work, of course -- I just also did not double check it :) I need to change the way in which I read and discuss code so that I can be more responsive.