[css-fonts] Make larger/smaller simple ratios

[Manishearth]

As of https://bugzilla.mozilla.org/show_bug.cgi?id=1361550, Firefox, Chrome, Safari, IE, and Servo all treat font-size: smaller and larger as simple ratios of the parent font size.

The spec specifies a complicated size selection mechanism (see https://drafts.csswg.org/css-fonts-3/#relative-size-value), based off of finding the relative position on the table and using that.

Given that font-size is pretty complicated already due to the inheriting dependence of keywords on the language/family, this only muddies things further.

We probably should make this a simple ratio (with the exact ratio values left up to implementors?)

[litherum]

The spec says:

<relative-size>
A <relative-size> keyword is interpreted relative to the table of font sizes and the computed ‘font-size’ of the parent element. Possible values are:
[ larger | smaller ]

For example, if the parent element has a font size of ‘medium’, a value of ‘larger’ will make the font size of the current element be ‘large’.

And, about absolute-size:

<absolute-size>
An <absolute-size> keyword refers to an entry in a table of font sizes computed and kept by the user agent. Possible values are:
[ xx-small | x-small | small | medium | large | x-large | xx-large ]

According to the current spec text, it seems to me:

• The relative sizes are only well-defined when an absolute size is inherited
• In-between absolute sizes, relative-sizes are fuzzy (and therefore can be implemented using a simple ratio)
• The UA has total control over the absolute sizes

Therefore, it seems to me that, if UAs choose absolute sizes which are a simple ratio from each other, then using a simple ratio would be spec compliant. However, no browser that I tested does this. Because of web compatibility, I can't imagine that any browser could change their absolute sizes. Therefore, we should change the spec.

Because of the existing fuzziness in the current spec, I don't think there is any value is specifying exactly which mechanism is used to increase/decrease the font size (for example, a UA could use a ratio, or a lookup table, or asking an oracle, etc.). The spec should simply say that "smaller" means smaller and "larger" means larger.

[litherum]

It's also worth mentioning that the three browsers I tested on all agree on their absolute font sizes.

[svgeesus]

Therefore, we should change the spec.

@litherum Not clear to me whether that means "edits needed" or whether you meant "Therefore, we shouldn't change the spec."

[litherum]

Therefore we should change the spec, but to make it more general. Browsers should be able to make the ratio different based on user settings. For example, users with impaired eyesight might need a larger ratio.

The spec should not define what “larger” means, other than that it is, well, larger.

[dbaron]

That sounds like the sort of thing that's likely to break content that's actually using larger. I suspect a significant chunk of that content assumes it means a certain ratio, or within a pretty tight range.

Browsers used to offer preferences for the default font size, but we've now settled on 16px because many sites don't work with other defaults, and if we want to show different font sizes we use various sorts of zooming.

[css-meeting-bot]

The Working Group just discussed [css-fonts] Make larger/smaller simple ratios, and agreed to the following resolutions:

• RESOLVED: add 1.2-1.5 as the conventional ratio range for the relative sizes when they're not modifying an absolute keyword size

The full IRC log of that discussion

<dael> Topic: [css-fonts] Make larger/smaller simple ratios
<dael> github: https://github.com//issues/1711
<dael> Chris: myles said he tested this and on 3 browsers they agreed on absolute font size but not relative. dbaron you commented you thinkt he content in practice depends on ratios. If we allow free choice the content wouldn't be web compat.
<dael> Chris: Do you have specific language you want to see?
<dael> dbaron: I think what I"m saying is I'm in favor of manish's original proposal, but I would spec the ratio
<dael> Chris: Is the ratio doc?
<dael> dbaron: I don't know.
<dael> Chris: He's also saying existing spec has complex table and no one does that.
<dael> Chris: Should we remove that? Mark at risk?
<dael> dbaron: I'd be in favor of removing the table.
<dael> dbaron: The link doesn't work.
<dael> Chris: I'll have a look.
<dael> Chris: See what's happened. It hasn't disappeared since Aug I don't think. Spec hasn't changed much.
<astearns> https://drafts.csswg.org/css-fonts-3/#typedef-absolute-size
<Rossen_> https://drafts.csswg.org/css-fonts-3/#font-size-adjust-prop
<dael> astearns: Is it the table below ^?
<dael> astearns: It's not in that def but it's below. There are 2 absolute-size anchors.
<dael> astearns: The idea is that since nobody is using info in table we remove it?
<dael> Chris: Right.
<dael> dbaron: I'm not sure it was this table. It's about hte absolute keywords rather then larger and smaller
<dael> Chris: Yes
<dael> Chris: Also that just says guidelines so there's nothing to conform to.
<dael> Chris: It says UA is free to interpolate or round to the closest. Seems to say relative sizes are relative to the absolute size table.
<dael> astearns: I think I've confused myself.
<dael> astearns: We are just talking about the larger and smaller keywords and what they do?
<dael> Chris: Right.
<dael> astearns: Okay.
<dael> Chris: Spec says you go up or down a size in the table and you read what that computes to.
<dael> dbaron: The issue is with the prose int he definition that refers to the table, but not the table itself.
<dael> Chris: Right
<dael> astearns: So myles was saying make the spec more general, but I don't see how we could b/c definition says UA is free to do wahtever it wants.
<dael> dbaron: Thought I think if everyone does 1.2 then maybe we shouldn't say it's free.
<dael> Chris: We went from 1.5 to 1.2 and we've pushed 1.2 strongly for quite some time.
<dael> Chris: Maybe spec says 1.2 is convention? There needs to be guidance for the mythical new impl.
<dael> dbaron: I believe Gecko treats larger and smaller as factors of 1.2 That may be pretty interop but I don't remember for sure.
<dael> astearns: What about adding 1.2 as a conventional ratio but not requiring.
<dael> Chris: Works for me.
<dbaron> https://bugzilla.mozilla.org/show_bug.cgi?id=1361550 is where Gecko changed to use 1.2 as a ratio
<dael> astearns: dbaron is that enough for you?
<dael> dbaron: Fine for me. I linked to the bug when Gecko changed the behavior, but I haven't read it. It says Chrome and Safari use 1.2 and 0.85.
<dael> dbaron: In Gecko we used 1.2 and 1/1.2 which isn't exactly the same.
<dael> astearns: Obj to adding 1.2 as the conventional ratio for the relative sizes when they're not modifying an absolute keyword size?
<dael> Rossen_: Looks like we're 1.5. I don't believe I've seen interop bugs on this.
<dael> Rossen_: If anything I would reserve the right to come back next week and ask for a change, but I don't obj.
<dael> Chris: If someone says larger and smaller they don't much care. In practice I don't think they're much used.
<dael> dbaron: Agreed people don't use, partly bc no interop
<dael> Rossen_: We're currently using 1.5. Yep. Reading through both larger and smaller we're consistant on 1.5x
<dael> Chris: That's what CSS1 said, then we changed it in like '98
<dael> Rossen_: And that's likely the last time we changed this code.
<dael> astearns: We culd do a guidance ratio range where browsers in practice tend to use 1.2-1.5
<dael> Rossen_: I'd be happy with that.
<dael> astearns: Obj to a ratio range?
<gregwhitworth> smaller = 1.6% usage | larger .47% usage https://developer.microsoft.com/en-us/microsoft-edge/platform/usage/css/font-size/
<dael> RESOLVED: add 1.2-1.5 as the conventional ratio range for the relative sizes when they're not modifying an absolute keyword size
<dael> Chris: I don't understand gregwhitworth's IRC which impl 1.6
<dael> astearns: That's usage
<dael> astearns: 1.6% of sites use smaller...that's high.
<aja> smaller in UA sheets for <small>
<dael> gregwhitworth: That's over hte 1.5 mil random URLs.
<dael> Chris: Thanks.
<dael> gregwhitworth: To Rossen_ 's point we haven't seen interop issues.
<dael> Chris: Okay. I'm happy with this.

[FremyCompany]

FWIW, Edge behavior is to use the next font-enum size in order (so if you are 16px (normal) or 17px etc... you will be mapped to the next keyword size "large"). If you exceed xx-large and xx-small, then we apply a 1.5 factor. If other browsers are just doing a 1.2 factor, this would actually help remove a bunch of code.

Do we have test cases for that?

[svgeesus]

@FremyCompany so that would mean not changing the spec to say "1.2 to 1.5" and instead keeping it at 1.2?