Generate partial interface CSSStyleDeclaration from known CSS properties

[foolip]

https://drafts.csswg.org/cssom/#the-cssstyledeclaration-interface has some IDL "macros":

partial interface CSSStyleDeclaration {
  [CEReactions] attribute [TreatNullAs=EmptyString] CSSOMString _camel_cased_attribute;
};
partial interface CSSStyleDeclaration {
  [CEReactions] attribute [TreatNullAs=EmptyString] CSSOMString _webkit_cased_attribute;
};
partial interface CSSStyleDeclaration {
  [CEReactions] attribute [TreatNullAs=EmptyString] CSSOMString _dashed_attribute;
};

These should be turned into something like this:

partial interface CSSStyleDeclaration {
  [CEReactions] attribute [TreatNullAs=EmptyString] CSSOMString display;
  [CEReactions] attribute [TreatNullAs=EmptyString] CSSOMString align-content;
};

Gecko actually does generate IDL as part of the build step:
https://github.com/mozilla/gecko-dev/blob/86897859913403b68829dbf9a154f5a87c4b0638/dom/webidl/CSS2Properties.webidl.in

So, it would be great if reffy could output a cssom-generated.idl based on the properties found in all other specs, which could then be used in an idlharness.js test. Using something like the above snippet with display and align-content, I know that Chrome and Safari will fail all tests, Firefox fail the prototype ones (because it has CSS2Properties instead) and Edge pass except for dashed attributes.

I'll have a go at generating that IDL.

[foolip]

This would build on #113, which is already working.

@tidoust @lukebjerring, there are a number of options for when to generate the IDL and I'm wondering if you have preferences:

• In reffy itself, before things are pushed to reffy-reports, somewhere in src/cli/crawl-specs.js
• By the code in reffy-reports which creates branches/PRs for wpt (Trigger wpt interfaces/ updates in master Travis job in this repo webref#20)
• Copy the files in https://github.com/tidoust/reffy-reports/tree/master/whatwg/css to wpt and let JS in wpt fetch those files and generate the IDL on-the-fly.

I'm leaning towards the last option, and the main problem I see is how to ensure that all the specs that define properties actually are tested like this. Even if we did a single css/cssom/interfaces-generated.html test, that test would still have to list all the JSON files to load and check.

[tidoust]

I don't feel strongly but I'd prefer the last option as well. Although Reffy runs some analyses on the content it extracts, the dumps were meant to stay as close to the actual IDL (or meta-IDL) defined in the spec as possible, so that other tools can run on them with confidence that what they are seeing is what the spec defines.

[tidoust]

Reffy now generates a styleDeclaration property for CSS properties as discussed in w3c/webref#51. This approach was preferred to generating the partial interface definitions (which can easily be done with this new property).