innerText issues

[zcorpan]

When working on #1678 I found the following issues.

cc @rocallahan

Getter

• What should non-CSS UAs do? (45253f9)

• Is "content order" for CSS boxes defined? (Step 2.5.) (Issue moved to [css-text] Define "content order" for innerText w3c/csswg-drafts#421 )

• The rp check only works for direct children of rp, not descendants (e.g. <ruby><rp><span>(</span></rp>...).
  Do we want that to work? Or change the content model or rp to "text"?
  (Change <rp>'s content model to Text #1690)

• Issue inline in the spec:

  This algorithm is amenable to being generalized to work on a ranges. Then we can use it as the basis for Selection's stringifier and maybe expose it directly on ranges. See Bugzilla bug 10583.

• @annevk said:

  The only thing I'd consider changing further is making the recursion less declarative. Have that be some algorithm that is invoked for each child and also put the result in a variable of some kind the rest of the algorithm uses.

  @zcorpan said:

  Yeah, it could use some more cleanup. Possibly also switch to iterative traversal instead of recursive?

  2202c6c

Setter

• The setter needs to be better defined, in terms of DOM concepts so mutation observers are invoked correctly etc. (0783a61)

• innerText = null; WebKit/Chromium "", Gecko/IE "null". (618de84 , Test innerText = ""/null/undefined web-platform-tests/wpt#3482)

• Chromium throws for <br>.innerText = "x" etc, Gecko does not.
  https://cs.chromium.org/chromium/src/third_party/WebKit/Source/core/html/HTMLElement.cpp?sq=package:chromium&dr=CSs&rcl=1471276009&l=140
  https://cs.chromium.org/chromium/src/third_party/WebKit/Source/core/html/HTMLElement.cpp?sq=package:chromium&dr=CSs&rcl=1471276009&l=454

  (Test that setting innerText on <br> etc doesn't throw web-platform-tests/wpt#3491)

• Only update existing text node's data if element has just 1 text node?
  https://cs.chromium.org/chromium/src/third_party/WebKit/Source/core/editing/serializers/Serialization.cpp?sq=package:chromium&dr=CSs&rcl=1471276009&l=640

  (innerText setter should replace existing text node web-platform-tests/wpt#3493)

• If the assigned value starts with a newline, WebKit/Chromium insert an empty text node, Gecko/IE don't. (Current spec matches Gecko/IE.)

  (innerText setter should not result in empty text nodes web-platform-tests/wpt#3492)

(Filed chromium bug 639064, webkit bug 160971, edge bug 8536472 for some of the above bullet points)

Other known issues

• should the innerText getter require layout? rocallahan/innerText-spec#2
• suggested tweak for explicit newline rocallahan/innerText-spec#3
• maybe add some tests for successive <pre> items rocallahan/innerText-spec#4 (Test innerText getter for two <pre> siblings web-platform-tests/wpt#3589)
• innerText should be supported for the <option> elements in a <select> rocallahan/innerText-spec#5 (Make <select> special in innerText #1717)

[domenic]

Which of these do you think we'd need to resolve before merging #1678? In my opinion the setter issue should block merging but the rest are OK-ish...

[domenic]

rocallahan/innerText-spec#5 also seems kind of bad

[rocallahan]

What should non-CSS UAs do?

Just return textContent as if the node was display:none?

Chromium throws for
.innerText = "x" etc, Gecko does not.

That adds complexity and it's hard to imagine it's required for Web compat.

Only update existing text node's data if element has just 1 text node?

Ditto.

rocallahan/innerText-spec#5 also seems kind of bad

Yeah that should be fixed, but you could fix it after merging.

[zcorpan]

In my opinion the setter issue should block merging but the rest are OK-ish...

Fixed the setter in 0783a61

[zcorpan]

Found a new issue today (updated OP)

If the assigned value starts with a newline, WebKit/Chromium insert an empty text node, Gecko/IE don't. (Current spec matches Gecko/IE.)

[domenic]

• I think we forgot to move a mention of innerText as nonstandard in the CEReactions section
• we should send a PR to the downstream repo with a redirect or other pointer link so as not to cause confusion going forward.

[zcorpan]

The rendering section in HTML has this

User agents that do not support correct ruby rendering are expected to render parentheses
around the text of rt elements in the absence of rp elements.

Is this something that people want to do in innerText? (Edit: if we do that, maybe we should also change the algorithm to just ignore rp elements)