Re: [Qemu-devel] [PATCH v4 2/3] qapi: Add a primitive to include other files from a QAPI schema file

[Benoît Canet]

The Tuesday 04 Mar 2014 à 09:02:57 (+0100), Markus Armbruster wrote :
> Eric Blake <address@hidden> writes:
> 
> > On 03/03/2014 08:27 AM, Markus Armbruster wrote:
> >
> >>>>> +The QAPI schema definitions can be modularized using the 'include'
> >>>>> directive:
> >>>>> +
> >>>>> + include("sub-system/qapi.json")
> >>>
> >>>> And now it isn't JSON anymore.
> >>>
> >>>> To keep it JSON, use syntax like
> >>>
> >>>>     { "include": "sub-system/qapi.json" }
> >
> > I actually think this looks nicer - makes the file more consistent.
> 
> I suspect qapi.py would look nicer, too :)
> 
> >>>> If you absolutely must make it non-JSON, you better rename the .json
> >>>> files.
> >>>
> >>>> Hmm, we already are non-JSON, because we use ' instead of " for no sane
> >>>> reason.
> >
> > A weak argument: ' is easier than " to type (at least on US keyboards -
> > no shift key required).
> >
> > Another weak argument: using ' in the qapi files vs. " in actual QMP
> > makes it easy to interleave discussions about semantics vs. examples of
> > those semantics in use (you can see whether a code snippet is talking
> > about qapi or wire format based on what quoting it used)
> >
> > Our files are already non-JSON due to comments (JSON has no notion of #
> > introducing a comment to ignore text to the next newline).  But both our
> > use of comments and our use of ' instead of " can be remedied in a
> > one-pass sed script to get a true JSON output if such is needed, at
> > least as long as we don't need to quote any " characters in the schema.
> >
> > Therefore, I agree that making the include syntax closer to true JSON is
> > desirable, whether or not we also decide to use " in the files to begin
> > with.  I don't see any way around the fact that JSON doesn't define
> > comments, vs. our absolute need for comments in our schema files, though.
> 
> We certainly can't do without comments.
> 
> JSON is designed for easy data exchange, but we use it as programming
> language syntax.  Its restrictions make sense for easy data exchange,
> but hurt our use.  We're not the first ones experiencing that pain:
> http://json5.org/
> 
> No idea how much momentum this JSON5 thingy has...
> 
> >>>> Our JSON parser accepts ' as an extension, to save us quoting in C
> >>>> strings.  That reason doesn't apply to .json files.
> >>>
> >>> Is it a problem if they are not pure JSON? In the end, they are parsed by
> >>> qapi.py (which already knows about file syntax), and having a separate 
> >>> syntax
> >>> for includes makes it somewhat easier to spot when that happens.
> >> 
> >> I don't particularly care whether schema syntax is pure JSON, some
> >> bastardized variation of JSON, or something else entirely.  But as long
> >> as we advertize schema files it as .json, they better contain JSON.  If
> >> they contain something else, they should be called something else.
> >
> > Maybe .qapi? But the name qapi-schema.qapi sounds redundant...
> 
> schema.qapi?
> 
> Switch to JSON5 and call it qapi-schema.json5?
> 

Hmm don't we want something that python and other language know how to parse out
of the box ? Or will we write yet another delicate work of art to parse it ?

Best regards

Benoît