[cssom-view-1] Can scrollIntoView param type be more specifc?

[saschanaz]

The IDL of Element partial interface defines scrollIntoView receives boolean or object. Can the IDL be more specific and directly refer ScrollIntoViewOptions as other scroll methods do?

void scrollIntoView();
void scrollIntoView((boolean or object) arg);
void scroll(optional ScrollToOptions options);
void scroll(unrestricted double x, unrestricted double y);
void scrollTo(optional ScrollToOptions options);
void scrollTo(unrestricted double x, unrestricted double y);
void scrollBy(optional ScrollToOptions options);
void scrollBy(unrestricted double x, unrestricted double y);

[zcorpan]

Nope, unfortunately.

WebIDL requires dictionary arguments (unless they have required members) to be marked as optional. But legacy requires scrollIntoView() to be like scrollIntoView(true). To achieve this the spec uses an overload and a union, and then converts the object to a dictionary in prose. There's no IDL syntax to achieve this weird-but-required-for-compat behavior.

[saschanaz]

Looking at the internal steps I think it still can be void scrollIntoView(optional ScrollToOptions options); and void scrollIntoView(boolean arg);, as it can still do the step If arg is not specified or is true.

[zcorpan]

No, that has different semantics. An omitted dictionary is equivalent to specifying {} per WebIDL. So with that IDL, the prose cannot differentiate between omitted argument and an empty dictionary.

[saschanaz]

Ah, you're right, my fault. But then how about void scrollIntoView(optional (boolean or ScrollIntoViewOptions) arg = true);?

Web IDL 2.2.3

Such arguments are always considered to have a default value of an empty dictionary, unless otherwise specified.

Dictionary types cannot have a default value specified explicitly, so the “unless otherwise specified” clause above can only be invoked for a union type that has a dictionary type as one of its flattened member types.

So I think this way it won't get the unwanted default empty dictionary value {}.

[zcorpan]

Hmmmm... yes, that could work, I think!

cc @bzbarsky to confirm.

[bzbarsky]

Presumably the point is that for compat we want scrollIntoView() to act like be scrollIntoView(true) but scrollIntoView(undefined) to act like scrollIntoView(false); that was certainly the reason for the initial non-use of an optional argument here and use of an overload instead, back when the first arg of the overload was not a dictionary. If that's the case, it sure would be nice if someone just wrote this down, in a note, or comment, or something, so we don't have to keep coming back to it over and over again.

That said, I don't think Gecko's current IDL for this (presumably based on an earlier spec draft?) has this property. It looks like this:

void scrollIntoView(boolean top);
void scrollIntoView(optional ScrollIntoViewOptions options);

where ScrollIntoViewOptions has ScrollLogicalPosition block = "start";. So passing either no arg or undefined will give the same behavior as passing { block: "center" } or true. The IDL in the spec now does have the property that passing nothing and passing undefined have different behavior. @zcorpan When was it changed from the thing Gecko implements and why? That might tell us what's going on here.

[bzbarsky]

And ccing @kearwood because he might know what was going on with the Gecko behavior change there, when we went from:

void scrollIntoView();
void scrollIntoView(boolean top, optional ScrollOptions options);

to what we have now.

[zcorpan]

Thanks @bzbarsky. Yes, undefined is intended to be like false. There is a comment in the source; I should change that to a visible comment:

<!-- scrollIntoView() means arg=true but scrollIntoView(undefined) means arg=false, to align with impls. -->

Another thing is that passing {} is intended to be like { block: "center", inline: "center" }, because that is argued to be a common case in https://www.w3.org/Bugs/Public/show_bug.cgi?id=17152

[zcorpan]

If Gecko already treats undefined the same as true, that suggests undefined meaning false is not a strong web compat invariant we need to maintain. If so, we can switch to

void scrollIntoView(optional (boolean or ScrollIntoViewOptions) arg = true);

[bzbarsky]

That's possible. Have you tested what other UAs do?

[zcorpan]

http://software.hixie.ch/utilities/js/live-dom-viewer/saved/5238

Gecko/WebKit/Chromium treat explicit undefined like omitted argument and top-aligns.

EdgeHTML 15 treats it like false and bottom-aligns.

[bzbarsky]

Sounds like aligning to the majority behavior, which is also simpler to spec, is not unreasonable. @travisleithead, would that be OK with you?