[mediaqueries][css-contain] How to evaluate `<ratio>` queries?

[Loirooriol]

It seems hard for a viewport to have a degenerate ratio, so I guess @media (aspect-ratio) is mostly fine. But we have a problem for @container (aspect-ratio), there is no interoperability. Testcase

First, how should aspect-ratio be evaluated in boolean context?

Typically that checks if the condition is true for any non-zero value. Is that 0 / 1 for ratios? I'm not convinced this makes much sense, one of these options seems better:

• Check if any ratio would evaluate to true
• Check if any ratio different than 0 / 0 would evaluate to true
• Check if any non-degenerate ratio would evaluate to true

Also, how to compare ratios?
I think we could say 0/a = 0/b < c/d = e/f < g/0 = h/0 for any a,b,c,d,e,f,g,h > 0 s.t. c*f = d*e.
Then the tricky part is comparing with 0 / 0.

Some proposals:

1. We consider ratios in the [0,∞] interval, plus a special 0/0 value outside the interval. Strict inequalities between 0/0 and other ratios are false.
2. We consider ratios in the [0,∞] interval, with 0/0 being an indeterminate value in the interval. Most comparisons with 0/0 are unknown, but some inequalities with the limits of the interval are known.
3. Like 1 or 2, but using unknown for ratios that are not comparable, i.e. when none of =, < and > would be true.
4. Like 3, but using unknown when comparing a degenerate ratio with itself.

I think I'm leaning towards the 1st one.

Some notes from the implemented behaviors that I have observed:

• Evaluating min-aspect-ratio and max-aspect-ratio in a boolean context produces unknown as defined in the spec.
• aspect-ratio: <ratio> is a synonym of aspect-ratio = <ratio>, min-aspect-ratio: <ratio> is a synonym of aspect-ratio >= <ratio>, and max-aspect-ratio: <ratio> is a synonym of aspect-ratio <= <ratio>.
• Ratios are broken in Blink, e.g. 1 and 1 / 1 have different behaviors, same for 0 and 0 / 1.

[css-meeting-bot]

The CSS Working Group just discussed [mediaqueries][css-contain] How to evaluate `<ratio>` queries?, and agreed to the following:

• RESOLVED: aspect-ratio queries are always true (except when it can't apply at all)
• RESOLVED: Accept Oriol's option 1 (0/0 is comparable with itself, but false when compared with any other ratio)

The full IRC log of that discussion

<TabAtkins> oriol: CQs have the ability to compare the aspect-ratio to a value
<TabAtkins> oriol: You can also do in MQs, but more difficult to get into the degenerate cases there, so the CQ case is more important
<TabAtkins> oriol: So what should happen if you eval aspect-ratio in a boolean context?
<TabAtkins> oriol: Typically it's true if the feature is true when equal to any non-zero value
<TabAtkins> oriol: But should that apply here?
<TabAtkins> oriol: And do all 0/N ratios work the same?
<TabAtkins> oriol: Propose three options:
<TabAtkins> oriol: 1) Check if any ratio evaluates to true (not exlcuding a 0 ratio)
<TabAtkins> oriol: 2) Check if any ratio other than 0/0 would be true - treat *this* special ratio as the "none"
<TabAtkins> oriol: 3) Check if any non-degenerate ratio evals to true
<TabAtkins> oriol: And then there's a question of how to compare different ratios
<TabAtkins> oriol: If a ratio is > a value, how to compare if one is degenerate?
<fantasai> TabAtkins: taking up separately seems fine
<fantasai> TabAtkins: For the first, I propose a 4th option: all ratios are always true
<fantasai> TabAtkins: because there's no actual zero value
<fantasai> oriol: I guess that could work, too
<fantasai> TabAtkins: I don't think there's any meaningful behavior existing, so don't need to worry about it
<fantasai> florian: Conceptually the only one to answer false about is "I don't have a screen, I don't have an aspect ratio"
<fantasai> Rossen_: what about 0x0 viewport?
<Rossen_> ack fantasai
<TabAtkins> fantasai: Two options that could make sense is Tab's proposal with florian's amendment: only false if you don't have a viewport at all
<TabAtkins> fantasai: Even 0/0 is a ratio
<TabAtkins> fantasai: Other option is any degenerate ratio is false
<TabAtkins> fantasai: Includign zero
<TabAtkins> Rossen_: When printing, what's the a-r?
<TabAtkins> fantasai: page size
<fantasai> TabAtkins: We don't have a way to express "no viewport at all"...
<fantasai> TabAtkins: I'm fine with FLorian's amendment
<fantasai> Rossen_: other opinions?
<TabAtkins> (in effect, it's never false then)
<fantasai> Rossen_: Then proposed to choose option 4, false in the case of no viewport
<fantasai> oriol: This would only be for media queries
<TabAtkins> oriol: So always true in CQ, since there's always an element?
<TabAtkins> fantasai: I mean an aural browser could exist that doesn't have sizing
<fantasai> fantasai: Could have a non-visual browser, then can't have a box
<TabAtkins> florian: That's a broader question - geometry questions in a non-visual browser is a wider issue
<TabAtkins> proposed resolution: aspect-ratio queries in a boolean context are always true
<TabAtkins> RESOLVED: aspect-ratio queries are always true (except when it can't apply at all)
<TabAtkins> oriol: So othe rissue is comparison
<TabAtkins> oriol: [missed]
<TabAtkins> oriol: ratios of the form N/0 are all equal, and all greater than finites
<TabAtkins> oriol: I think this is straightforward, but would like confirmation
<TabAtkins> oriol: But then have to say what to do with 0/0
<TabAtkins> oriol: Doesn't fit into the expanded real line
<TabAtkins> oriol: I proposed some options
<TabAtkins> oriol: 1) Model this as a special value outside the real line - you can compare ratios in the real line, or outside the real line, but not mixed
<TabAtkins> oriol: 2) Instaed of 0/0 being special, it's a value in the interval that's unknowable
<TabAtkins> oriol: So we'd get unknown when comparing with real numbers
<TabAtkins> oriol: 3) Basically the same but with some other cases degenerate.
<TabAtkins> oriol: I'm leaning towards the first option
<TabAtkins> oriol: You can look at the tables I made for brwoser behavior
<TabAtkins> oriol: But I propose option 1 - always true or false, but comparing 0/0 with any real ratio is always false
<fantasai> TabAtkins: I agree, and I think Oriol's option 1 makes sense
<TabAtkins> Summary: 0/0 == 0/0 is true, 0/0 with any other ratio is false.
<TabAtkins> Rossen_: Objections?
<fantasai> wfm, given Tab's summary
<TabAtkins> RESOLVED: Accept Oriol's option 1 (0/0 is comparable with itself, but false when compared with any other ratio)

[Said182]

به نظر سخت است که یک ویوپورت نسبت انحطاط داشته باشد، بنابراین حدس می‌زنم @media (aspect-ratio)اکثراً خوب است. اما ما یک مشکل داریم @container (aspect-ratio)، هیچ قابلیت همکاری وجود ندارد. مورد آزمایشی

اول، چگونه باید در زمینه بولی ارزیابیaspect-ratio شود ؟

معمولاً بررسی می کند که آیا شرط برای هر مقدار غیر صفر درست است یا خیر. آیا این 0 / 1برای نسبت ها است؟ من متقاعد نیستم که این خیلی منطقی باشد، یکی از این گزینه ها بهتر به نظر می رسد:

• بررسی کنید که آیا نسبتی به درست ارزیابی می شود یا خیر
• بررسی کنید که آیا نسبتی متفاوت از آن است که 0 / 0به درستی ارزیابی می شود
• بررسی کنید که آیا نسبت غیر منحط به درست ارزیابی می شود یا خیر

همچنین چگونه نسبت ها را مقایسه کنیم؟ من فکر می کنم ما می توانیم برای هر خیابان بگوییم . سپس بخش دشوار مقایسه با .0/a = 0/b < c/d = e/f < g/0 = h/0``a,b,c,d,e,f,g,h > 0``c*f = d*e``0 / 0

برخی از پیشنهادات:

1. نسبت‌ها را در بازه [0,∞] به اضافه مقدار 0/0 ویژه خارج از بازه در نظر می‌گیریم. نابرابری های شدید بین 0/0 و سایر نسبت ها نادرست هستند.
2. نسبت‌ها را در بازه [0,∞] در نظر می‌گیریم که 0/0 یک مقدار نامشخص در بازه است. بیشتر مقایسه‌ها با 0/0 ناشناخته هستند، اما برخی نابرابری‌ها با حدود فاصله مشخص هستند.
3. مانند 1 یا 2، اما استفاده از مجهول برای نسبت هایی که قابل مقایسه نیستند، یعنی زمانی که هیچ یک از =، < و > درست نیست.
4. Like 3, but using unknown when comparing a degenerate ratio with itself.

I think I'm leaning towards the 1st one.

Some notes from the implemented behaviors that I have observed:

• Evaluating min-aspect-ratio and max-aspect-ratio in a boolean context produces unknown as defined in the spec.
• aspect-ratio: <ratio> is a synonym of aspect-ratio = <ratio>, min-aspect-ratio: <ratio> is a synonym of aspect-ratio >= <ratio>, and max-aspect-ratio: <ratio> is a synonym of aspect-ratio <= <ratio>.
• Ratios are broken in Blink, e.g. 1 and 1 / 1 have different behaviors, same for 0 and 0 / 1.