Not only +1 to selective toggling, I think there is another possible problem in the current design.

Let's say authors set html { text-decoration-skip: ink; } to their pages today.

The spec says:

Level 4 of this specification defines additional values (leading-spaces and trailing-spaces), and uses them in the initial value of this property, meaning that by default they are skipped.

So L4 UA must render decorations to leading/trailing spaces only on such pages, correct? This is probably unexpected, breaking behavior.

IIUC with the current design, or even with selective design as long as we have shorthands, we can't add new values that are skipping by default without breaking existing pages.

WDYT?

/cc @drott

So L4 UA must render decorations to leading/trailing spaces only on such pages, correct? This is probably unexpected, breaking behavior.

IIUC with the current design, or even with selective design as long as we have shorthands, we can't add new values that are skipping by default without breaking existing pages.

You're right, any new "skip-by-default", whether introduced now or later, with or without shorthand, would be a breaking change, since it would be a change in behavior. I don't think there's any way around that, the only thing we can do is decide if the breakage is a problematic one or a negligible one, and whether the benefit brought by the new skip-by-default is outweights that breakage or not.

In the case of leading and trailing space, the difference is noticeable if there are  s or if white-space is pre or pre-wrap. In all the cases I can think of, skipping by default is an improvement, but that's subjective, and authors are infinitely creative. In terms of compat risk, I'll note that if there's no   but white-space is pre or pre-wrap, Safari already skips trailing spaces (but not the leading one). This (small) gap in compatibility probably helps us (a little) pick which ever behavior we prefer.

If we switched to an "auto|none" shorthand, and move everything else to longhands, and made leading-spaces and trailing-spaces skipping the (non controlable) default in L3, we would avoid breakage between L4 and L3, but we'd still have to introduce the breaking change now.

Thank you for the confirmation. With that, and with the problem you mentioned, I have mild preference not to add the shorthand for now. We can add it whenever we want to, but it looks to me that the risks win over the benefit at this point, and unless I miss something, the benefit of the shorthand for this function isn't very high?

@kojiishi I am not sure I understand what you mean. Do you mean you prefer to keep the spec as-is with only the text-decoration-skip shorthand, no text-decoration-skip-* longhands, or to change to add only the text-decoration-skip-* longhands, and remove the text-decoration-skip shorthand?

Well, there's two part to my suggestion.

1. add a text-decoration-skip-* as longhands of text-decoration-skip.

2. limit the shorthand to only none and auto, requiring authors to use longhands for everything else

If we only want to do 1, we can do it any time we want (but I would still argue that the sooner the better). If we want to do 2 as well, we need to do it before it gets significant usage.

If you'd like to only add the longhands and remove the shorthand (possibly adding it back latter), we would need to do it now, but I am not sure what the benefit is compared to 1+2.

I think limiting the shorthand to resets only (none and auto), and using the longhands to toggle on and off the various aspects of skipping will lead to more robust and DRY code. This includes avoiding accidental overrides of whatever the browser's default value is, but also makes it easier to avoid fighting with yourself or your colleagues when you want different types of skipping to cascade separately.

For example, something like this is pretty painful to do without longhands:

* { text-decoration-skip-ink: all }
:lang(ja,zh,ko) { text-decoration-skip-ink: none }
:lang(zh), a[href] { text-decoration-skip-edges: all; }
img { text-decoration-skip-objects: all; }
pre code { text-decoration-skip-spaces: leading trailing; }

If that's the preferred coding style (and I think it should be), having any other value than none and auto is mostly a footgun.

Also, while any change to the default behavior is a “breaking” one, regardless of syntax, some breakage may be minor and advantageous. For example, I am not sure we will do it, but it doesn't seem crazy to me that we would eventually decide that the initial value of text-decoration-skip-ink is neither on nor off, but some kind of script dependent auto. If we limit the shorthand to auto and none and require authors to use longhands instead, everyone who hasn't explicitely set ink skipping one way or another will benefit. If we require (because we didn't introduce longhands) or allow (because we introduce longhands, but leave the full syntax on the shorthand) authors to use the shorthand, effectively everyone who has used this property, including for changes unrelated to ink-skipping, will miss out.

OK, I've drafted up the changes discussed here and resolved at a high level in https://lists.w3.org/Archives/Public/www-style/2017Feb/0049.html

@frivoal Let me know if these edits look OK, or if there's something that should be fixed before closing out this overaching issue. (There are a number of detail issues which I think we should address separately.)

@fantasai Thanks, this looks great.

I do have a few remarks. If you think any of these is an obvious bugfix, please fold that in and tick it off below. After that, I'm happy to re-file the rest into separate issues so that we can track them individually. For now, I'm dumping it all here as the result of a one-pass review, but I do expect most of these to move to separate issue.

• text-decoration-skip-self says “Applies to: all elements”, but its object value's definition says: “Skip this element [...] if it is an atomic inline”. I wonder if this should be changed to “Applies to: atomic inlines”, or if you also envision a value that that skips the element even if it is not an atomic inline? If so, we may need to rename the existing value.
• text-decoration-skip-inset's auto has “The UA should [...]”. Are you sure this is a should? As much as it makes sense to leave the size of the inset UA defined (and possibly context dependent), I'm not sure what the benefit is to allow UAs to implement this on/off switch at the syntax level and yet have no effect.
• The grammar of text-decoration-skip-spaces defines start and end, but the initial value and prose calls them leading and trailing. I don't strongly care which one we pick (and we can bikeshed that separately if need be, as noted in the inline issue), but the spec should be self-consistent.

@fantasai I've applied the two tweaks you agreed about, and fixed a couple more bugs editorial / markup bugs along the way. Closing, as I'm happy with the resolution and its execution.