All Web IDL constants are missing

[foolip]

I was using confluence to research w3c/svgwg#291 a bit today, to see if https://svgwg.org/svg2-draft/types.html#InterfaceSVGUnitTypes was implemented everywhere.

The interface object is there (everywhere), but the constants aren't listed. They are own properties of the interface object, and I do see SVG_UNIT_TYPE_UNKNOWN in the data collected in https://github.com/mdittmer/web-apis.

Other interfaces I checked to confirm missing constants: Event, Node. HTMLMediaElement.

Per #202 (comment) this seems to not be an accident, but it still seems like something to fix.

[mdittmer]

This was by design after we observed that "actual" constants that (usually) look like SomeConstructor.SOME_CONSTANT were bloating API counts. I have implemented a CONST_CASE check at #296 and it is staged with fresh data here. The biggest surprise I saw there was Firefox's jump in browser-specific APIs.

[mdittmer]

@bzbarsky do you think the data at the const-case staging instance is more or less accurate than prod wrt Firefox APIs?

[bzbarsky]

I wish I could tell you. The staging instance shows mostly "..." for me various places without enough label text to actually say anything about anything.

[bzbarsky]

That said, if you are just adding constants, then I would expect to see 372 things added from the DOM_VK_* constants on KeyEvent and KeyboardEvent. Is that about what you're seeing?

[bzbarsky]

(Those are the constants from https://www.w3.org/TR/1999/WD-DOM-Level-2-19990923/events.html#Events-KeyEvent fwiw.)

[mdittmer]

The change is actually to include APIs that appear to be constants (i.e., have a "value" in descriptor, rather than getter/setter) but are *not* CONST_CASE. Prod exclude valued properties regardless of their name. So, the change should be purely additive, and should not introduce CONST_CASE properties.

…

On Fri, May 18, 2018, 3:57 PM Boris Zbarsky, ***@***.***> wrote: (Those are the constants from https://www.w3.org/TR/1999/WD-DOM-Level-2-19990923/events.html#Events-KeyEvent fwiw.) — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#260 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABsWSCVXCHqbWHxgFGqOLaASdbFGxwaJks5tzye4gaJpZM4Rz3oh> .

[bzbarsky]

Well, https://const-case-data-dot-web-confluence.appspot.com/ is certainly showing CONST_CASE things.

[mdittmer]

Oops. I fixed the code without updating the data. Please try the staging URL now. Most (not all) CONST_CASEs are now filtered out, but other "value"d things are included.

Looks like some additional filtering is needed to drop non-APIs that accidentally shipped with browsers or bleed through from the data collection methods (e.g., interface names +toString+ and a)

[foolip]

@mdittmer, https://const-case-data-dot-web-confluence.appspot.com/ includes stuff that isn't in any of the 4 listed browsers, is that intentional? Also I'm not able to scroll with page up/down, that would be nice.

[foolip]

Another somewhat surprising thing is that searching for "HTMLMedia" doesn't find HTMLMediaElement. I know that's because the data isn't loaded yet, but that didn't stop me from trying.

[foolip]

HTMLMediaElement#HAVE_NOTHING now shows up, and I don't see HAVE_NOTHING anywhere else, so constants on interface objects (constructors) aren't being filtered out, that's good! I wonder if there are any things that are filtered out that aren't included on some other "object" instead?

[bzbarsky]

At this point I'm really not sure what the new thing is trying to measure. If it's trying to measure webidl constants, why is it excluding CONST_CASE stuff?

@mdittmer is there an existing issue on the new UI being a lot less usable than the old UI? Or do you need one filed?

[foolip]

@bzbarsky, the problem was that some things that weren't CONST_CASE were previously excluded, like in #259. Including those is the most important thing. But IDL constants also show up both on the interface object and on instances, so there's a risk of double counting. @mdittmer, can you list which APIs are excluded by the new logic?

[bzbarsky]

But IDL constants also show up both on the interface object and on instances

Well, on the interface object and on the prototype. And yes, there is risk of double-counting.

I don't have a strong opinion about whether IDL constants should be included or not; I agree it's important to include things like static methods.

If you just want to exclude IDL constants, then excluding value properties which are non-writable is possibly a better approach than excluding based on name, imo. In Chrome, location.href is writable, of course.