Skip to main content

makeover_webview/
lib.rs

1//! The webview renderer for [`makeover_layout`].
2//!
3//! <!-- wiki: makeover-webview -->
4//!
5//! # The renderer that needs no palette
6//!
7//! `makeover-immediate` and `makeover-tui` both take a `Palette`, because egui
8//! and a terminal need an actual colour before they can put anything on
9//! screen. A webview does not: `var(--surface-raised)` *is* the late binding,
10//! and the browser resolves it against whatever `themes.js` last wrote onto
11//! `:root`.
12//!
13//! So this crate emits text naming intents, and never learns a colour. It is
14//! the deferral rule with no adapter in the way, and it is why the webview was
15//! always the wrong renderer to derive a vocabulary from: it can express
16//! anything, so it never pushes back.
17//!
18//! # Phase A: the stylesheet
19//!
20//! This module emits component CSS and no markup, deliberately. GoingsOn has
21//! 145 `innerHTML` sites and Balanced Breakfast 175 `createElement` sites, so
22//! moving markup is a migration where adopting a generated stylesheet is not.
23//! The apps keep every line of their markup and gain the classes.
24//!
25//! It is not a deletion either, which this header claimed until the measurement
26//! came in. Adoption across goingson removed 49 declarations net and *added* 25
27//! lines: a rule loses its depth declarations and gains a variant selector next
28//! to it, so the file stays the same size. What phase A moves is where depth is
29//! defined, not how much CSS exists. Numbers and method in the wiki note under
30//! "The deletion test, run".
31//!
32//! The bevel properties are byte-identical to what both apps already
33//! hand-write, which is asserted below.
34//!
35//! Some of what phase A emits is not a look but the withdrawal of one. A
36//! renderer that picks its element from the description inherits that element's
37//! user-agent chrome, and [`reset`] is where a primitive says which parts of it
38//! were never asked for.
39//!
40//! # Phase B: the markup, one description at a time
41//!
42//! [`form`] renders [`makeover_layout::Field`], which is the half of phase B
43//! whose description is settled. It emits strings, because both apps
44//! interpolate their fields into larger string-built forms and returning nodes
45//! would rewrite those too. It owns its own escaping, on the reasoning in that
46//! module: a Rust encoder can cover element text and attribute values with one
47//! function, where the apps need four and have to choose correctly at every
48//! call site.
49//!
50//! [`facet`] renders `makeover_layout::Facet`: a dimension a set is narrowed by,
51//! and the one phase-B emitter whose markup an app is not keeping, because the
52//! markup it replaces was two mechanisms rather than one. A tag's selection and
53//! a tag's browse position were separate state on MNW's discover page, which is
54//! why every filter row there carries a tick box *and* a chevron; one gesture
55//! doing both is what lets the second one go.
56//!
57//! [`list`] is the other half: column tracks, the narrowing rules, and the cell
58//! containers a row is made of. It stops at the cell boundary and does not
59//! render what goes inside one, on the reasoning in that module. So phase B is
60//! now the frame around content in both directions, and what an app still owns
61//! is the content itself.
62//!
63//! # What phase A settled, and what it costs
64//!
65//! Measured against goingson's `styles.css` rather than against a component
66//! list: `.btn`, `.card` and `.tag, .badge` each hand-write the same
67//! composition, so three quarters of phase A is one rule with several names.
68//!
69//! Two of the four decisions change how goingson looks, and adoption should
70//! not be described as a pure deletion:
71//!
72//! - **Pressed carries its fill.** [`interactive_rules`] emits
73//!   [`Depth::pressed`] whole. goingson presses to `--surface-sunken` today and
74//!   will press to `--surface-well`, and hovers to `--surface-overlay` today
75//!   and will hover to `--hover-surface`. Since `surface-well` inverts by theme
76//!   where `surface-sunken` does not, a dark theme presses *lighter* than it
77//!   hovers. That falls out of `makeover`'s own derivation, which says outright
78//!   that `surface-sunken` cannot serve as a well, so if it reads wrong the
79//!   answer is there and not here.
80//! - **Badges go flat.** See [`token_rules`].
81//!
82//! The other two: the progress trough is renderer-local and the scrollbar
83//! track was dropped ([`component_rules`]), and no class prefix ships by
84//! default, so adoption means deleting the app's hand-written rule in the same
85//! commit that adds the generated one. `.card`, `.badge` and the tab classes
86//! all already exist in goingson, and while both rules exist the cascade order
87//! decides which wins. That is the one real risk in adopting this, and it is
88//! why the migration lands per component rather than in one commit.
89//!
90//! # Interaction states
91//!
92//! [`interactive_rules`] emits four states, in emission order, and the order is
93//! load-bearing: they are all specificity (0,2,0), so disabled beats hover by
94//! coming last and by nothing else. Nothing here reaches for `:not(:disabled)`,
95//! which would raise a selector this crate wraps in its own layer.
96//!
97//! Emitting the states here is what keeps an app from completing the primitive
98//! from outside, by out-specifying a rule it does not own. Those overrides are
99//! also what breaks under cascade layers: an app that declares `@layer` puts
100//! its own rules in a named layer, and unlayered declarations outrank every
101//! named layer regardless of specificity.
102//!
103//! Hover sits inside a capability query. `makeover-touch` answers whether a
104//! fingertip has hover and `makeover-geometry` spells the condition; this crate
105//! asks and does not decide, so no app has to take the hover state back on
106//! touch.
107//!
108//! # The layer contract
109//!
110//! [`stylesheet`] emits into the `makeover` cascade layer ([`CSS_LAYER`], which
111//! lives in `makeover-geometry` because that is the one crate every CSS emitter
112//! in the family already depends on). `makeover-geometry` does the same for
113//! `geometry.css`.
114//!
115//! The cascade resolves origin and importance, then layer, then specificity,
116//! then source order, and **unlayered normal declarations outrank every named
117//! layer**. An unlayered generated file therefore beats every rule an app owns,
118//! regardless of specificity and regardless of loading last. Nothing errors when
119//! that happens: the CSS is valid, the minifier is happy, and buttons and badges
120//! look subtly wrong. The layer belongs here rather than in each app, because an
121//! app cannot fix it from its own stylesheet: the fix is to layer the file it
122//! does not own.
123//!
124//! An app should declare the order once, or the layer's position is decided by
125//! whichever generated file the browser happens to see first:
126//!
127//! ```css
128//! @layer makeover, base, components, responsive;
129//! ```
130//!
131//! [`in_css_layer`] is re-exported for an app that assembles its own stylesheet
132//! from this crate's pieces: rules an app generates from them are as generated
133//! as the ones here, so they belong in the same layer and this crate cannot put
134//! them there on the app's behalf.
135//!
136//! # Suggestions
137//!
138//! `Outcome::Suggestions` carries `Candidate` rather than `Choice`, and a
139//! candidate has no `unavailable`: a suggestion that cannot be picked is a row
140//! a route should not have offered. What it has instead is a `detail`, the line
141//! that tells it from a row reading the same, and it is drawn in
142//! `--content-muted` rather than in the disabled token. A detail orients rather
143//! than refuses, and every other secondary line in this crate reads the same
144//! way. The class is `.form-suggestion-detail`.
145//!
146//! # An interval is one question with two ends
147//!
148//! [`makeover_layout::FieldKind::Interval`] emits a `role="group"` named by the
149//! field's label, holding one `<input type="number">` per end.
150//!
151//! - **The group carries the error and the descriptions**, on the split
152//!   [`makeover_layout::FieldKind::Radio`] already uses here: what is wrong is
153//!   the answer, and a crossed interval is not the fault of either end.
154//! - **Both boxes take the whole extent.** `min`, `max` and `step` describe the
155//!   axis, so they are written twice. The crossing rule is not emitted, because
156//!   HTML has no attribute for it and the description does not carry it: it
157//!   comes back as an error on the group, like every other refusal.
158//! - **Which end is which is `aria-label` and nothing more.** The description
159//!   states direction structurally, by which member holds which name, and never
160//!   in words. Visible Min and Max captions are a page's own and reach the
161//!   group through [`form::Filling::trailing`].
162//!
163//! [`form::Value::Between`] is the second value. A separator inside one string
164//! would make this crate the owner of a delimiter that either end could contain.
165//!
166//! # A number's unit is adjacent text
167//!
168//! HTML has no unit attribute and inventing one would be markup nothing reads,
169//! so `Field::unit` is a `<span>` after the control. It is named in
170//! `aria-describedby` rather than left as decoration, because a number and what
171//! it is measured in are one fact and reading the first without the second is
172//! reading it wrong. What that buys is a unit a consumer can read back rather
173//! than a suffix on a label it would have to parse.
174//!
175//! # A curve this renderer can carry, and one it declines
176//!
177//! A range takes its granularity from the curve (`Field::curve.step()`), every
178//! other kind keeps `Field::step`, and `Curve::Linear` emits a plain range.
179//!
180//! **A constant-ratio curve emits a linear track, and that is the answer, not a
181//! debt.** HTML has no logarithmic range input, so a described screen asking
182//! for one is asking the browser for something it does not have, the same class
183//! of request as [`makeover_layout::FieldKind::Date`] on a host with no
184//! calendar. The renderer answers with the nearest control the host really
185//! offers and keeps every fact that survives the translation: the extent, the
186//! granularity, and the value's own units. What does not survive is resolution
187//! at the small end. The value submitted is still a value in the field's own
188//! units, which is what every handler on this path reads.
189//!
190//! The alternatives are worse in the specific way this stack exists to avoid.
191//! Shipping JS that maps thumb position to value puts app code back in the
192//! renderer. Changing what the control submits from a value to a fraction moves
193//! the mapping to whoever reads the form, and a server reading these forms with
194//! its own handlers would take a fraction where a value is expected, silently.
195//!
196//! When this reopens: the day a described screen on the webview path asks for a
197//! non-linear range. The answer then is mapping in `quasi-router`, where one
198//! implementation serves every host, not JS here.
199//!
200//! # A markdown field gets a preview
201//!
202//! A [`makeover_layout::FieldKind::Rich`] field is marked
203//! `data-format="markdown"`, and [`form::editor_rules`] is what spends that
204//! mark. The Write/Preview pair is a segmented control, so it takes the depth,
205//! the focus ring and the chosen state from rules that already exist; the
206//! preview pane is a well, because it stands where the control stood. Both are
207//! gated on the attribute rather than on a class, which is what the attribute is
208//! for. A permission taken and not spent turns every conversion into a
209//! regression.
210//!
211//! **This crate renders no markdown.** The pane arrives empty and is filled by
212//! whatever binds the editor, which is where the host's sanitiser already is. A
213//! converter here would move that guarantee into a crate with no view of the
214//! host's content-security posture.
215//!
216//! # Ranges, ghost text, and an option that cannot be picked
217//!
218//! - `FieldKind::Range` emits `<input type="range">`, and `Field::step` emits
219//!   `step`. The step is emitted only when the description carries one: the
220//!   browser's own default is `step="1"`, which is what a description means by
221//!   saying nothing, and is also what turns a 0-to-1 threshold into a
222//!   two-position control.
223//! - A select with nothing chosen emits a disabled, selected, valueless first
224//!   option carrying `Field::placeholder`. HTML has no placeholder attribute on
225//!   `<select>`; this is the idiom, and `required` keeps working through it
226//!   because the option's value is empty.
227//! - `Choice::unavailable` emits `disabled` plus the reason. Where it goes
228//!   differs by control and the difference is forced: a radio group gets a
229//!   `.form-option-reason` span beside the label, and a `<select>` option has
230//!   room for no element at all, so the reason runs into its text.
231//! - `Choice::detail` takes the same split for the same reason: a
232//!   `.form-option-detail` span in a radio group, run into the text of a
233//!   `<select>`'s option. An option carrying both reads what it is before why it
234//!   cannot be picked.
235//!
236//! # A cell says what it holds
237//!
238//! [`CellPart`](makeover_layout::CellPart) names the four things a cell holds,
239//! and [`table_rules`] turns them into `.cell-value`, `.cell-tokens`,
240//! `.cell-actions` and `.cell-link`. The value and the link take a colour: a
241//! token carries its own tone and an action is a control rather than text.
242//!
243//! The colour goes on `.cell-value` rather than on `.cell`. On the container it
244//! cascades into the parts that are not text, and a control in a cell is painted
245//! as text, which is the drift
246//! [`RowPart::intent`](makeover_layout::RowPart) prevents for list rows.
247//!
248//! [`list::Cell::part`] is `Option<CellPart>` and never `Option<RowPart>`: the
249//! two answer different questions, and only one of them is about a cell.
250//!
251//! # A table narrows itself
252//!
253//! A table a description produced knows its columns at render time, so no rule
254//! can be written per table: it would have to travel with the markup, as a
255//! `<style>` element that needs `style-src 'unsafe-inline'` or in a head an htmx
256//! fragment swap does not carry. [`table_rules`] lays every table out as flex
257//! rows instead, each column starting at its declared floor and giving up room
258//! by its [`Priority`](makeover_layout::Priority), so the browser narrows a
259//! table at the width its floors add up to. [`list::column_classes`] is what
260//! puts the column's classes on a cell, and every cell's contents go in
261//! [`list::CELL_IN`]. **A header row emitted by a renderer's own code has to do
262//! both too**, or the header and the body disagree about which column just
263//! closed and how wide the rest are.
264//!
265//! `.button` takes the four tones as colour, off `data-tone`, the way the badge
266//! does, so a destructive button has somewhere for its tone to land. A list is
267//! reset rather than left as a bulleted list.
268//!
269//! `RowPart::revealed_on_hover` is not honoured. Hiding a row's actions until
270//! hover hides them from pointer users alone, who are the ones scanning a list
271//! to learn what can be done to a row, and every escape the rule grows
272//! (`focus-within` for the keyboard, a capability gate for a fingertip) is a
273//! report that hiding was wrong for somebody.
274//!
275//! # The depth classes are not controls
276//!
277//! `.raised` is a statement about shape and carries no interactive set, so the
278//! vocabulary has a raised surface that is merely an object. An app that wants
279//! one does not have to take a control class and cancel the control half.
280//!
281//! `.button` is the same depth *and* a control, and takes its states from
282//! [`surface_rules`], which is where a state belongs: on the thing that claims
283//! to answer a pointer. `.card` is a raised object, and it answers a pointer
284//! only when the element it is written on is a control -- `a`, `button`,
285//! `label`, or anything carrying `data-act` -- so one name serves the card
286//! that is read and the card that is pressed.
287//!
288//! # Substitution, three ways
289//!
290//! A theme with no `surface-well` is answered differently by each renderer,
291//! which is why substitution belongs to a renderer and not to the description:
292//!
293//! - `makeover-immediate` substitutes the page in Rust.
294//! - `makeover-tui` refuses to substitute and draws an edge instead, because a
295//!   terminal would quantise the two together.
296//! - here, CSS already has the mechanism: `var(--surface-well,
297//!   var(--surface-page))` falls back in the browser, and nothing in Rust
298//!   decides anything.
299
300#![forbid(unsafe_code)]
301
302pub mod chart;
303pub mod facet;
304pub mod figure;
305pub mod form;
306pub mod list;
307pub mod meter;
308pub mod placeholder;
309pub mod reset;
310pub mod vocabulary;
311
312mod table;
313use table::table_rules;
314
315/// A render of every emitter, scraped for the classes it wrote.
316///
317/// Test-only, and the guard behind [`vocabulary::names`]. See the module's own
318/// header for why the check renders rather than reads the source.
319#[cfg(test)]
320mod corpus;
321
322use crate::list::{cell_part_class, part_class};
323use crate::reset::{Chrome, Reset};
324use makeover_geometry::{Density, SizeClass};
325// Re-exported rather than redefined. An app assembling its own stylesheet out
326// of this crate's pieces needs the same layer name, and most such apps depend
327// on this crate and not on `makeover-geometry` directly.
328pub use makeover_geometry::{CSS_LAYER, in_css_layer};
329
330/// This crate's version, as the generated stylesheet reports it.
331///
332/// A consumer whose lockfile still pins an old `makeover-webview` gets a
333/// well-formed sheet with components missing and no error anywhere, so the
334/// emitter has to name itself in what it writes. Read by
335/// `makeover_build::layout_css` through [`stylesheet`].
336pub const VERSION: &str = env!("CARGO_PKG_VERSION");
337use makeover_layout::{
338    Bevel, CellPart, Depth, Fallback, Fill, Flow, Intent, RowPart, Selector, Sort, State, Token,
339    Tone,
340};
341use makeover_touch::Affordance;
342use std::fmt::Write as _;
343
344/// How the emitted CSS is shaped.
345#[derive(Debug, Clone, Copy, PartialEq, Eq)]
346pub struct Emit {
347    /// Bevel thickness, as a CSS length.
348    ///
349    /// A value, so it arrives from the caller: border widths belong to
350    /// `makeover-geometry` and will come from there once it carries them.
351    pub border_width: &'static str,
352    /// Focus ring thickness, as a CSS length.
353    ///
354    /// Separate from [`border_width`](Self::border_width), and never derived
355    /// from it: a bevel and a focus indicator answer different questions, and
356    /// only one of them has to be noticed from across a desk.
357    ///
358    /// The default is the measured consensus rather than a new opinion. Every
359    /// consumer had already written its own ring and all three chose at least
360    /// 2px: the MNW server 2px across 10 rules, Balanced Breakfast 2px,
361    /// goingson 2px on three rules and 3px on the one covering twelve
362    /// selectors. The design system was the only thing in the tree saying 1px.
363    pub focus_width: &'static str,
364    /// The smallest a pointer target may be drawn, as a CSS length.
365    ///
366    /// A value for [`border_width`](Self::border_width)'s reason, and it goes
367    /// to `makeover-geometry` on the same trip: a floor is a size, and this
368    /// crate names no sizes of its own.
369    ///
370    /// The default is WCAG 2.2's target minimum. It is an absolute rather than
371    /// a step off the geometry base, and that is deliberate on two counts: a
372    /// floor that scales with the reader's root size is not a floor, and the
373    /// touch density opens the gaps *between* targets rather than licensing a
374    /// smaller one for a pointer.
375    pub target_min: &'static str,
376    /// Prefix for emitted class names, without the leading dot.
377    pub class_prefix: &'static str,
378    /// What marks a compulsory field, appended to its label.
379    ///
380    /// `makeover-tui`'s `PieceStyle::required_marker` and
381    /// `makeover-immediate`'s `FieldStyle::required_marker`, said here for the
382    /// third renderer, with their default. A knob for their reason: it is the
383    /// one piece of *copy* this crate emits, and copy is not a renderer's
384    /// call.
385    ///
386    /// This crate was the outlier. `Field::required` reached it as the HTML
387    /// attribute and nothing else, so a browser refused an empty submit and
388    /// the screen never said which fields would do that, while both sibling
389    /// renderers had marked the label since they were written.
390    ///
391    /// # Why it is hidden from the accessibility tree here and nowhere else
392    ///
393    /// The control already carries `required`, which a screen reader
394    /// announces. A marker read out beside it is the same fact twice, so the
395    /// span is `aria-hidden`. The other two renderers have no such attribute
396    /// to lean on, which is why the marker is part of the label string there
397    /// and an element here.
398    pub required_marker: &'static str,
399}
400
401impl Default for Emit {
402    fn default() -> Self {
403        Self {
404            border_width: "1px",
405            focus_width: "2px",
406            target_min: "24px",
407            class_prefix: "",
408            required_marker: "*",
409        }
410    }
411}
412
413/// A string as CSS escapes, for a `content` value.
414///
415/// `\u{25B2}` becomes `\25B2`. Emitted escaped rather than literally so the
416/// stylesheet is ASCII whatever the description spells: a `content` string is
417/// read by whatever encoding the consumer serves the file as, and a caret that
418/// depends on that is a caret that works on one machine.
419///
420/// Terminated by the closing quote at every site here. A CSS hex escape takes
421/// up to six digits and ends at the first character that cannot be one, so an
422/// escape followed by more text would need a space that these do not.
423fn css_escape(text: &str) -> String {
424    text.chars().fold(String::new(), |mut out, c| {
425        let _ = write!(out, "\\{:X}", c as u32);
426        out
427    })
428}
429
430/// The CSS custom property holding a bevel's composition.
431#[must_use]
432pub fn bevel_var(bevel: Bevel) -> &'static str {
433    match bevel {
434        Bevel::Raised => "--bevel-raised",
435        Bevel::Inset => "--bevel-inset",
436        Bevel::RaisedOpen => "--bevel-raised-open",
437        // An edge added to the description since this renderer was last
438        // built. The raised edge rather than no edge: a control with no
439        // box-shadow reads as flat, which is a different claim, where a
440        // closed edge is at worst the same shape drawn one run too many.
441        _ => "--bevel-raised",
442    }
443}
444
445/// A `var()` reference to a fill intent, with the browser's own fallback where
446/// the intent may be absent.
447///
448/// The fallback is CSS syntax, not a decision made here. That is the whole
449/// difference between this renderer and the other two.
450#[must_use]
451pub fn fill_var(fill: Fill) -> String {
452    match fill {
453        Fill::Well => format!("var(--{}, var(--{}))", fill.token(), Fill::Page.token()),
454        other => format!("var(--{})", other.token()),
455    }
456}
457
458/// The two-tone edge as a `box-shadow` value.
459///
460/// Two inset shadows, one per corner pair: the light one offset down and
461/// right so it lands on the top and left edges, the dark one the other way.
462/// The same assignment `makeover-immediate` draws with polylines and
463/// `makeover-tui` draws with box-drawing characters.
464#[must_use]
465pub fn bevel_shadow(bevel: Bevel, opts: &Emit) -> String {
466    let (top_left, bottom_right) = bevel.edges();
467    let w = opts.border_width;
468    // The shaded run's vertical offset is what puts it on the bottom edge. At
469    // zero it lands on the right edge alone, which is the open bevel: three
470    // sides drawn and the fourth left for the surface below to continue
471    // through. Same two shadows either way, so nothing downstream has to know
472    // which it got.
473    // The sign belongs to the value rather than to the template: writing
474    // `-{down}` against a zero emits `-0`, which is a length no stylesheet
475    // should carry even where a parser accepts it.
476    let down = if bevel.draws_bottom() {
477        format!("-{w}")
478    } else {
479        "0".to_string()
480    };
481    format!(
482        "inset {w} {w} 0 var(--{}), inset -{w} {down} 0 var(--{})",
483        top_left.token(),
484        bottom_right.token()
485    )
486}
487
488/// The custom properties both bevels resolve through.
489///
490/// Emitted as properties rather than inlined into every rule because that is
491/// what the apps already do, and because a consumer that wants the edge
492/// without the fill reads the property directly.
493#[must_use]
494pub fn bevel_properties(opts: &Emit) -> String {
495    let mut css = String::new();
496    for bevel in [Bevel::Raised, Bevel::Inset, Bevel::RaisedOpen] {
497        let _ = writeln!(
498            css,
499            "    {}: {};",
500            bevel_var(bevel),
501            bevel_shadow(bevel, opts)
502        );
503    }
504    css.push_str(ELEVATION_PROPERTY);
505    css
506}
507
508/// The cast shadow of a surface that floats over the page.
509///
510/// Composed here for the reason the bevel pair is: `makeover` derives the tone,
511/// this crate owns the geometry, and neither has to know the other's numbers.
512///
513/// **Only for a surface that overlays the page.** A menu, a toast, a popover, a
514/// dropdown. A surface *in* the page takes `.raised` and its bevel, and a rule
515/// that reaches for this on a card or a plate has renamed a literal rather than
516/// replaced it.
517///
518/// Two lengths rather than one, because a single blur reads as a smudge at
519/// plate size and as a halo at menu size. The offset is small and downward: a
520/// Platinum-era menu sits just off the page rather than hovering above it.
521const ELEVATION_PROPERTY: &str =
522    "    --elevation-overlay: 0 2px 4px var(--elevation), 0 8px 24px var(--elevation);\n";
523
524/// The class name for a depth.
525#[must_use]
526pub fn depth_class(depth: Depth, opts: &Emit) -> Option<String> {
527    let name = match depth {
528        Depth::Flat => return None,
529        Depth::Raised => "raised",
530        Depth::Well => "well",
531        Depth::Sunken => "sunken",
532        // A depth added to the description since this renderer was last
533        // built. No class, on the same footing as Flat: emitting a name
534        // whose rule body we cannot write would put a class in the markup
535        // that the stylesheet never defines.
536        _ => return None,
537    };
538    Some(format!("{}{name}", opts.class_prefix))
539}
540
541/// A prefixed class name.
542///
543/// Public, for the renderers that emit markup this crate does not. A screen
544/// renderer writing `class="row"` has to prefix it the way the stylesheet half
545/// does, or a prefixed app gets rules matching everything except the elements
546/// that renderer wrote, and the failure is invisible: the CSS stays valid and
547/// one element is unstyled. Call this rather than copying it.
548#[must_use]
549pub fn class(name: &str, opts: &Emit) -> String {
550    let mut out = String::with_capacity(opts.class_prefix.len() + name.len());
551    push_class(&mut out, name, opts);
552    out
553}
554
555/// A prefixed class name, written into a buffer the caller already has.
556///
557/// The form the emitters use, and the reason it exists is [`escape_into`]'s:
558/// putting every class on every element through a `format!` allocates even in
559/// the default case, where the prefix is empty and the answer is the argument.
560/// A described table row carries roughly eighty transient allocations that way,
561/// and this and the escaper are most of them.
562///
563/// [`class`] stays for callers holding a name rather than a buffer.
564///
565/// [`escape_into`]: crate::form::escape_into
566pub fn push_class(out: &mut String, name: &str, opts: &Emit) {
567    out.push_str(opts.class_prefix);
568    out.push_str(name);
569}
570
571/// The attribute an element's custom properties ride in, instead of `style`.
572///
573/// A number this crate hands the stylesheet -- a bar's value and its axis, a
574/// facet's depth, a meter's fill -- is a custom property the rules divide or
575/// multiply. It used to be written as `style="--value: 4210"`, and a
576/// `style-src` without `'unsafe-inline'` refuses every style attribute, so a
577/// page under that policy drew every chart flat. The same text in this
578/// attribute is data a policy does not police, and the host's script sets each
579/// property through the CSSOM, which the policy allows: quasi-webview's
580/// `VARS_JS`.
581///
582/// The value is `--name: value` pairs separated by `;`, the declaration syntax
583/// the `style` attribute already had, so the numbers still reach the markup as
584/// themselves. That is the property the residual seam needs: see the
585/// [`chart`](crate::chart) module header.
586///
587/// Only custom properties. A script that applied any declaration from markup
588/// would hand the policy's refusal straight back.
589pub const VARS_ATTR: &str = "data-vars";
590
591/// The class an option of a selector carries, which is what the rules key off.
592///
593/// Named for the option and not for the group: [`selector_rules`] styles the
594/// thing that gets picked, so `Selector::Tabs` is `tab` and not `tabs`. The
595/// distinction is not pedantry. quasi-webview spelled these `tabs`, `segmented`
596/// and `option`, put `toggle` on the wrapping element rather than on the
597/// buttons inside it, and every described selector in that renderer came out
598/// with no depth, no focus ring and no chosen state, while the toggle group got
599/// a bevel meant for its buttons.
600///
601/// The chosen option additionally carries `chosen`, the same way a latched chip
602/// carries `latched`. That name is this crate's too; there is no reason for a
603/// caller to spell it, and [`selector_rules`] is where it is written down.
604#[must_use]
605pub fn option_class(selector: Selector) -> &'static str {
606    match selector {
607        Selector::Tabs => "tab",
608        Selector::Segmented => "segment",
609        Selector::Toggle => "toggle",
610    }
611}
612
613/// The fill and edge declarations for a depth, as a rule body.
614///
615/// Empty for [`Depth::Flat`], which has neither and inherits what it sits on.
616/// Callers lean on the emptiness to skip the rule rather than emit a class that
617/// sets nothing: a class that sets no properties is a class that means "I
618/// thought about this", which is what comments are for.
619///
620/// The two halves are emitted independently because [`Depth::Sunken`] has a
621/// fill and no bevel. Requiring both would silently drop the fill for exactly
622/// that case. Independent does not
623/// mean unpaired: both halves still come off one `Depth`, so they cannot
624/// disagree about what the region is.
625#[must_use]
626pub fn depth_declarations(depth: Depth) -> String {
627    let mut css = String::new();
628    if let Some(fill) = depth.fill() {
629        let _ = writeln!(css, "    background: {};", fill_var(fill));
630    }
631    if let Some(bevel) = depth.bevel() {
632        let _ = writeln!(css, "    box-shadow: var({});", bevel_var(bevel));
633    }
634    css
635}
636
637/// One rule giving a selector a depth, or nothing when the depth declares
638/// nothing.
639#[must_use]
640pub fn depth_rule(selector: &str, depth: Depth) -> String {
641    let body = depth_declarations(depth);
642    if body.is_empty() {
643        return String::new();
644    }
645    format!(".{selector} {{\n{body}}}\n")
646}
647
648/// The media condition a hover rule has to sit inside, or `None` if hover is
649/// unconditional.
650///
651/// Two crates answer this and neither answer is made here. `makeover-touch`
652/// owns *whether* hover exists at a density, and `makeover-geometry` owns how
653/// that capability is spelled as a media condition. Asking both is what stops
654/// this renderer minting a third opinion, which is what all three apps did:
655/// goingson sniffed the user agent, Balanced Breakfast used `(hover: none)`
656/// alone, and the MNW server had no gate at all.
657///
658/// [`SizeClass`] is required by [`Affordance::available`] and ignored by this
659/// member, which reports as much through `reads_size`. Passing Compact is not
660/// a claim about width; the test below pins that every class agrees.
661fn hover_condition() -> Option<&'static str> {
662    if Affordance::Hover.available(Density::Touch, SizeClass::Compact) {
663        // A fingertip grew a hover state. Nothing to gate, and this renderer
664        // should not invent a reason to gate anyway.
665        None
666    } else {
667        Some(Density::Pointer.media_condition())
668    }
669}
670
671/// Put a rule inside a media query, or leave it alone.
672fn gated(condition: Option<&str>, rule: &str) -> String {
673    let Some(condition) = condition else {
674        return rule.to_string();
675    };
676    let mut css = format!("@media {condition} {{\n");
677    for line in rule.lines() {
678        // Blank lines stay blank. Indenting one leaves trailing whitespace,
679        // which is the sort of thing a formatter later reverts and calls a diff.
680        if line.is_empty() {
681            css.push('\n');
682        } else {
683            let _ = writeln!(css, "    {line}");
684        }
685    }
686    css.push_str("}\n");
687    css
688}
689
690/// The keyboard focus ring, placed by the depth it lands on.
691///
692/// This is the webview's **focus ring** and nothing more. **Reach** and
693/// **focus** are both the browser's — the document decides what is reachable
694/// and `:focus-visible` decides which reached thing wears the ring — and no
695/// description states either. The three terms are defined once in
696/// `makeover_layout`'s crate header, "Reach, focus and the focus ring".
697///
698/// One ring for the whole system, because a focus ring's job is to be
699/// recognised and three apps having three of them is the failure. What varies
700/// is where it sits, and that comes off [`Depth`] rather than off a per-
701/// component choice: a well takes the ring inside its own edge, and anything
702/// standing proud of the page takes it outside.
703///
704/// `outline` rather than the composed `box-shadow` the invalid-field ring at
705/// [`field_rules`] uses, and deliberately the one place the two rings are built
706/// differently. A `box-shadow` ring has to restate the bevel beside it, because
707/// `box-shadow` is not additive and a lone ring silently drops the well out
708/// from under the element. That restatement is a second copy of the depth,
709/// living in a different function from the first, and it is exactly the
710/// duplication `Depth` exists to prevent. `outline` occupies its own property,
711/// so the bevel survives untouched and there is nothing to keep in agreement.
712/// They render the same: both are a flush ring one border-width wide.
713#[must_use]
714pub fn focus_rule(selector: &str, depth: Depth, opts: &Emit) -> String {
715    let w = opts.focus_width;
716    // Same magnitude either way, and only the sign comes off the depth. Both
717    // values are what the consumers had already converged on independently:
718    // 2px out is what all three wrote, and 2px in is the MNW server's own
719    // answer for the one inset ring it had.
720    let offset = match depth.bevel() {
721        // Inside the well, clear of its edge rather than painted over it.
722        Some(Bevel::Inset) => format!("calc(-1 * {w})"),
723        // Raised, or no edge at all. Outside, standing off by its own width.
724        _ => w.to_string(),
725    };
726    // The token by name. It is `makeover`'s, derived from the action colour,
727    // and reaching it through a description member was a second path to the
728    // same variable for as long as one existed.
729    format!(
730        ".{selector}:focus-visible {{\n    outline: {w} solid var(--focus-ring);\n    outline-offset: {offset};\n}}\n"
731    )
732}
733
734/// A rest depth said out loud on both axes, for a rule that has to beat the
735/// states above it.
736///
737/// [`depth_declarations`] states an axis only when the depth has something to
738/// say about it, which is right for a rest rule: a [`Depth::Flat`] region
739/// inherits what it sits on, and asserting `background: none` there would be
740/// the difference between level-with and painted-transparent. It is wrong for
741/// a rule whose whole job is to take a state back. An axis left unstated is an
742/// axis the state above keeps, so `Flat` re-asserted nothing at all and a
743/// disabled control kept whatever hover had given it.
744///
745/// So the axes the depth is silent on are withdrawn rather than skipped, and
746/// the withdrawal is spelled by [`reset`] rather than here, so a disabled
747/// control and a flat one say the same words. Reaches further than the fill:
748/// [`Depth::Sunken`] and [`Depth::Overlay`] have no bevel either, and the
749/// pressed rule above hands out an inset one.
750fn rest_declarations(depth: Depth) -> String {
751    let mut css = String::new();
752    match depth.fill() {
753        Some(fill) => {
754            let _ = writeln!(css, "    background: {};", fill_var(fill));
755        }
756        None => css.push_str(&Reset::NOTHING.and(Chrome::Fill).declarations()),
757    }
758    match depth.bevel() {
759        Some(bevel) => {
760            let _ = writeln!(css, "    box-shadow: var({});", bevel_var(bevel));
761        }
762        None => css.push_str(&Reset::NOTHING.and(Chrome::Shadow).declarations()),
763    }
764    css
765}
766
767/// Present, visible, and not answering.
768///
769/// Matches the ARIA attribute as well as the pseudo-class, because `:disabled`
770/// only matches form elements and half the things this crate emits are not
771/// one: a `div` carrying `.chip` or `.tab` can never be `:disabled`. Keying on
772/// the accessible state is the pattern [`field_rules`] already establishes for
773/// `aria-invalid`, on the reasoning that one fact read by both the styling and
774/// the accessibility tree cannot drift from itself.
775///
776/// The rest depth is re-asserted rather than assumed, because this rule has to
777/// beat the hover and pressed rules above it. It does that on source order at
778/// equal specificity, not by out-specifying them: every rule this function's
779/// caller emits is (0,2,0), and adding a `:not(:disabled)` anywhere would raise
780/// one of them and have to be unpicked when this output moves inside its own
781/// cascade layer.
782///
783/// Re-asserted on **both** axes, through [`rest_declarations`].
784/// `depth_declarations` alone is empty for [`Depth::Flat`], so a flat control
785/// would win the contest with nothing to say and keep the hover surface
786/// underneath a control that had stopped answering.
787#[must_use]
788pub fn disabled_rule(selector: &str, depth: Depth) -> String {
789    format!(
790        ".{selector}:disabled,\n.{selector}[aria-disabled=\"true\"] {{\n{}    color: var(--{});\n    cursor: not-allowed;\n}}\n",
791        rest_declarations(depth),
792        State::Disabled.token()
793    )
794}
795
796/// Every state a selector that answers a click implies: hover, pressed, focus
797/// and disabled, in that order.
798///
799/// Order is the whole cascade mechanism here. All four selectors are
800/// specificity (0,2,0), so disabled wins over hover and pressed by coming last
801/// and by nothing else.
802///
803/// Pressed emits [`Depth::pressed`] in full, fill and edge together. Emitting
804/// only the edge is what left goingson hand-writing `background:
805/// var(--surface-sunken)` on three separate rules, and a fill that does not
806/// travel with its edge is precisely the disagreement `Depth` exists to make
807/// unrepresentable. So the pressed fill comes from the description
808/// (`--surface-well`) rather than from whatever each app reached for.
809///
810/// Hover has no member in the description and is renderer policy: a terminal
811/// and an immediate-mode painter have no hover to express. It resolves against
812/// `--hover-surface`, which `makeover` already derives and which nothing
813/// consumed until now. What it *is* gated on is capability, via
814/// [`hover_condition`]. Before that gate existed the apps each wrote their own:
815/// goingson's section 60 exists solely to take back the hover state this
816/// function had just handed it, by out-specifying a rule it does not own.
817///
818/// `depth` is the selector's **rest** depth, used to place the focus ring and
819/// to restore the surface under a disabled control. The pressed rule keeps
820/// inverting from [`Depth::Raised`] regardless: a tab's unchosen depth is
821/// [`Depth::Sunken`], and `Sunken.pressed()` is `Sunken`, so deriving the press
822/// from the rest depth would leave a tab with no press at all.
823#[must_use]
824pub fn interactive_rules(selector: &str, depth: Depth, opts: &Emit) -> String {
825    let mut css = gated(
826        hover_condition(),
827        &format!(".{selector}:hover {{\n    background: var(--hover-surface);\n}}\n"),
828    );
829    css.push_str(&depth_rule(
830        &format!("{selector}:active"),
831        Depth::Raised.pressed(),
832    ));
833    css.push_str(&focus_rule(selector, depth, opts));
834    css.push_str(&disabled_rule(selector, depth));
835    css
836}
837
838/// The compound a card's interactive states are written against: the class
839/// on an element that is itself a control.
840///
841/// `label` for a card wrapping a choice, `[data-act]` for an act a renderer
842/// drew on some other element. Without the class prefix, the way every
843/// selector handed to [`interactive_rules`] is.
844#[must_use]
845pub fn pressable_card(card: &str) -> String {
846    format!("{card}:is(a, button, label, [data-act])")
847}
848
849/// One rule per depth: its fill and its edge, together.
850///
851/// A depth and nothing else. `.raised` says a surface sits on what is behind
852/// it, which is a statement about the shape and not about what happens when a
853/// pointer arrives, so it emits no hover, press, focus or disabled rule. The
854/// named surfaces are where interaction lives: `.button`, and `.card` written
855/// on a control, get their states from [`surface_rules`].
856///
857/// Giving this class the interactive set leaves the vocabulary with no raised
858/// surface that is merely an object, so a consumer that needs one has to take a
859/// control class and cancel half of it.
860#[must_use]
861pub fn depth_rules(opts: &Emit) -> String {
862    let mut css = String::new();
863    for depth in [Depth::Raised, Depth::Well] {
864        let Some(class) = depth_class(depth, opts) else {
865            continue;
866        };
867        css.push_str(&depth_rule(&class, depth));
868    }
869    css
870}
871
872/// The three surfaces that are a depth with a name.
873///
874/// `button` and `card` are both [`Depth::Raised`], and `field` is a
875/// [`Depth::Well`] because that is the reading `Depth`'s own documentation
876/// gives a text field. Their bodies come out identical by construction rather
877/// than by hand: three hand-written copies in goingson's stylesheet is what
878/// phase A deletes, and generating them from one call is what stops them
879/// drifting apart again.
880///
881/// A card's states hang off [`pressable_card`] rather than the bare class. A
882/// card is a raised object that answers a pointer only when it is written on a
883/// control, which is what lets one name cover both: a store tile that is one
884/// link, a tier picker that is a label round a radio, and a use-case card that
885/// does nothing and so has no hover to cancel.
886fn surface_rules(opts: &Emit) -> String {
887    let mut css = String::new();
888    let button = class("button", opts);
889    css.push_str(&depth_rule(&button, Depth::Raised));
890    css.push_str(&interactive_rules(&button, Depth::Raised, opts));
891
892    let card = class("card", opts);
893    css.push_str(&depth_rule(&card, Depth::Raised));
894    css.push_str(&interactive_rules(
895        &pressable_card(&card),
896        Depth::Raised,
897        opts,
898    ));
899
900    let field = class("field", opts);
901    css.push_str(&depth_rule(&field, Depth::Well));
902
903    // A field takes focus and refuses input like everything else here, and got
904    // neither until now, which is why all three apps hand-write a focus ring
905    // for it and no two of them match. No hover or pressed: a text field does
906    // not light up under the pointer and does not invert when clicked, so the
907    // two states `interactive_rules` would add are the two it does not have.
908    css.push_str(&focus_rule(&field, Depth::Well, opts));
909    css.push_str(&disabled_rule(&field, Depth::Well));
910
911    // A file field's own button is a button. `::file-selector-button` is the
912    // one part of an `<input type="file">` a stylesheet reaches, and without a
913    // rule the platform's grey control sits inside a well drawn in the theme's.
914    // Raised, with the hover and the press a button has; no focus ring, because
915    // the focus belongs to the input the pseudo-element is part of, and the
916    // field's own ring already draws it.
917    let chooser = format!("{field}::file-selector-button");
918    css.push_str(&depth_rule(&chooser, Depth::Raised));
919    let _ = writeln!(
920        css,
921        ".{chooser} {{\n    border: none;\n    color: inherit;\n    font: inherit;\n    padding: var(--gap-bound) var(--gap-peer);\n    margin-inline-end: var(--gap-peer);\n    cursor: pointer;\n}}"
922    );
923    css.push_str(&gated(
924        hover_condition(),
925        &format!(".{chooser}:hover {{\n    background: var(--hover-surface);\n}}\n"),
926    ));
927    css.push_str(&depth_rule(
928        &format!("{chooser}:active"),
929        Depth::Raised.pressed(),
930    ));
931
932    // Keyed on the ARIA attribute rather than on a class, so the visual state
933    // and the accessible state cannot drift apart: there is one fact and both
934    // read it. goingson already drove its invalid styling this way and was
935    // right to; the `.invalid` class this emitted before 0.5.0 was a second
936    // place to forget.
937    //
938    // The ring composes *after* the bevel rather than replacing it. box-shadow
939    // is not additive, so a lone ring silently dropped the well out from under
940    // an invalid field. Flat and unlit: this edge is saying "wrong", and
941    // lighting one side would have it say "raised" at the same time.
942    let _ = writeln!(
943        css,
944        ".{field}[aria-invalid=\"true\"] {{\n    box-shadow: var({}), 0 0 0 {} var(--danger);\n}}",
945        bevel_var(Bevel::Inset),
946        opts.border_width
947    );
948
949    // The default-button ring, on the one control that commits
950    // (`makeover_layout::Act::commits`, wiki `explicit-commit-affordance`).
951    // A flush frame twice the border width, composed after the bevel for the
952    // invalid ring's reason, and outside the element, so the focus outline,
953    // which stands off by its own width, lands beside it rather than over it.
954    // Pressed keeps the ring and inverts only the bevel; disabled keeps it in
955    // the muted ink, the way a disabled control keeps its bevel.
956    //
957    // An attribute rather than a class, for `data-tone`'s reason: it is a fact
958    // about the act, and a renderer writes it beside the tone.
959    let ring = format!("0 0 0 calc(2 * {})", opts.border_width);
960    let _ = writeln!(
961        css,
962        ".{button}[data-commits] {{\n    box-shadow: var({}), {ring} var(--border-strong);\n    font-weight: bold;\n}}",
963        bevel_var(Bevel::Raised)
964    );
965    let _ = writeln!(
966        css,
967        ".{button}[data-commits]:active {{\n    box-shadow: var({}), {ring} var(--border-strong);\n}}",
968        bevel_var(Bevel::Inset)
969    );
970    let _ = writeln!(
971        css,
972        ".{button}[data-commits]:is(:disabled, [aria-disabled=\"true\"]) {{\n    box-shadow: var({}), {ring} var(--content-muted);\n}}",
973        bevel_var(Bevel::Raised)
974    );
975    css
976}
977
978/// Badges and chips.
979///
980/// The one place phase A changes how goingson looks rather than only where its
981/// rules live. [`Token::Badge`] is [`Depth::Flat`], so a badge emits no fill
982/// and no edge at all, where goingson ships `.tag, .badge` as a single rule
983/// carrying the raised bevel. Splitting that means reading every call site to
984/// decide which of the two it always was.
985///
986/// What a badge does carry is a [`Tone`], the intent family it shares with
987/// notices and nothing else. Neutral is the bare class rather than a variant,
988/// because it is the absence of a status and not a status called "none".
989/// Text that goes somewhere.
990///
991/// The one inline control. A table cell carries `cell-link` instead, which
992/// reads as its row's title rather than as a link; this is the same thing
993/// outside a table, which is what a described run holds when a sentence
994/// contains a link.
995///
996/// Colour and underline only. Whether a link is inline in a sentence or sitting
997/// on its own line is the app's layout, and how much room it takes is
998/// `makeover-geometry`'s. What is here is the pair of signals that say "this
999/// goes somewhere" and nothing that says where it sits.
1000///
1001/// The visited arm is deliberately absent. A link inside an app points at the
1002/// app's own screens, which the user is expected to have been to, so painting
1003/// them differently marks almost everything and distinguishes nothing.
1004fn link_rules(opts: &Emit) -> String {
1005    let mut css = String::new();
1006    let link = class("link", opts);
1007
1008    let _ = writeln!(
1009        css,
1010        ".{link} {{\n    color: var(--action);\n    \
1011         text-decoration: underline;\n}}"
1012    );
1013    // The hover step is the same one every other control takes, and it is a
1014    // colour rather than a surface: a link has no box to raise.
1015    let _ = writeln!(
1016        css,
1017        "@media (hover: hover) and (pointer: fine) {{\n    .{link}:hover \
1018         {{\n        color: var(--action-hover);\n    }}\n}}"
1019    );
1020    let _ = writeln!(
1021        css,
1022        ".{link}:focus-visible {{\n    outline: {} solid var(--focus-ring);\n    \
1023         outline-offset: 2px;\n}}",
1024        opts.focus_width
1025    );
1026    // A link is often a `<button>` rather than an `<a>`: a renderer picks the
1027    // element from the method, so a link that writes is a button that has to
1028    // stop looking like one. What that costs is named in [`reset`].
1029    css.push_str(&Reset::TEXT_BUTTON.rule(&format!("button.{link}")));
1030    css
1031}
1032
1033fn token_rules(opts: &Emit) -> String {
1034    let mut css = String::new();
1035
1036    // No `depth_rule` call here, deliberately: `Token::Badge.depth(_)` is Flat,
1037    // and a label with an edge says it can be pressed.
1038    let badge = class("badge", opts);
1039    // `content-muted` literally, not `Tone::Neutral.token()`. What makes a
1040    // badge quiet is `Token::Badge` answering no click, which this crate holds
1041    // and `Tone` genuinely does not know. Routing it through Neutral put the
1042    // claim where the evidence was not, and the bill arrived on the figure
1043    // value: it took the same muting from the same call and read as its own
1044    // caption. Neutral answers `content` from makeover-layout 0.36.0.
1045    //
1046    // A badge is a chip, wiki `table-model`: an opaque fill inside an edge, in
1047    // the theme's own ink at normal weight. Coloured text was never ratified
1048    // and measured 1.34:1 on goingson's warning; muted text 2.78:1 on
1049    // solarized-dark. So the ink is `content` and what says which kind of
1050    // token it is moves to the fill and the edge: the table's raised ground for
1051    // a neutral chip, makeover's `*-surface` for a status, edged in its tone.
1052    // Still no bevel: a label with a raised edge says it can be pressed.
1053    let bw = opts.border_width;
1054    let _ = writeln!(
1055        css,
1056        ".{badge} {{\n    display: inline-flex;\n    align-items: center;\n    \
1057         padding: 0 var(--step-snug);\n    border: {bw} solid var(--border);\n    \
1058         border-radius: var(--radius-fine);\n    background: var(--surface-raised);\n    \
1059         color: var(--content);\n    font-weight: normal;\n}}"
1060    );
1061    for tone in [Tone::Info, Tone::Success, Tone::Warning, Tone::Danger] {
1062        let _ = writeln!(
1063            css,
1064            ".{badge}[data-tone=\"{0}\"] {{\n    background: var(--{0}-surface);\n    \
1065             border-color: var(--{0});\n}}",
1066            tone.token()
1067        );
1068    }
1069
1070    // A button carries the four tones a badge does. It had none, on the reading
1071    // that a control's colour is its surface rather than its text, and that
1072    // reading has one hole big enough to matter: the button that destroys
1073    // something. Every consumer had written that rule itself, and a description
1074    // that says `Tone::Danger` on an act had nowhere for it to land.
1075    //
1076    // A fill and not the text, matching the badge exactly. This emitted
1077    // `color: var(--danger)` until 0.86.0, which is the shape wiki
1078    // `table-model` ratified against for status text and `{tone}-surface`'s own
1079    // derivation comment names: the tone colours are fills, and a fill used as
1080    // ink measures 3.36:1 for danger and 1.84:1 for success on goingson's
1081    // theme. Both are under the 4.5:1 floor, and success is under it by more
1082    // than the floor itself.
1083    //
1084    // It read as ratified for longer than it was because goingson set
1085    // `color` on `.button` in a later layer, so no described tone ever reached
1086    // a button and nobody saw the colour it would have drawn.
1087    //
1088    // The earlier note here said a red surface belongs to an app's own layer.
1089    // That argument was against a *saturated* fill, and it still holds: this is
1090    // the ground with 12 percent of the tone in it, which is the same tint a
1091    // status chip carries and is not something an app would draw over.
1092    //
1093    // The fill and NOT the edge, which is where this differs from the badge.
1094    // A badge's border is makeover's, so the badge tones it. A button's is the
1095    // app's: goingson writes `border: ... solid var(--border)` on `.button` in
1096    // a later layer, and a shorthand cancels a generated `border-color` the
1097    // same way its `color` cancelled the tone for months. goingson's own drift
1098    // check caught it on the first build. An edge an app silently overrides is
1099    // worse than no edge, so the tint carries the tone alone.
1100    let button = class("button", opts);
1101    for tone in [Tone::Info, Tone::Success, Tone::Warning, Tone::Danger] {
1102        let _ = writeln!(
1103            css,
1104            ".{button}[data-tone=\"{0}\"] {{\n    background: var(--{0}-surface);\n}}",
1105            tone.token()
1106        );
1107        // The pointer answer, at the specificity the tone rule just took. The
1108        // plain `.button:hover` is (0,2,0) and sits earlier in this sheet, so
1109        // without this a toned button is the one control on the screen that
1110        // does not move under a pointer.
1111        //
1112        // The fallback is for the version skew, and it is not decoration: the
1113        // hover tint is a `makeover` 3.8 token and this sheet is emitted from a
1114        // crate an app can upgrade on its own. An unresolved custom property
1115        // makes the whole declaration invalid at computed-value time rather
1116        // than dropping to the cascade, so without it a toned button on an
1117        // older theme would hover to `transparent`.
1118        css.push_str(&gated(
1119            hover_condition(),
1120            &format!(
1121                ".{button}[data-tone=\"{0}\"]:hover {{\n    background: \
1122                 var(--{0}-surface-hover, var(--hover-surface));\n}}\n",
1123                tone.token()
1124            ),
1125        ));
1126    }
1127
1128    // A chip holds itself down, which is `Depth::pressed` arrived at
1129    // independently by two apps. `removable` is a remove affordance, so it is
1130    // markup and waits for phase B.
1131    let chip = class("chip", opts);
1132    let unlatched = Token::Chip { removable: false };
1133
1134    // The pointer floor, and the one number in this file that does not scale.
1135    // A chip drawn from its padding alone came out 23px at every width and
1136    // every density in goingson: 36 of one ui-fuzz run's 51 target findings
1137    // were this single rule. The length arrives from the caller, the way every
1138    // other length in this crate does -- see `Emit::target_min` for why the
1139    // floor is an absolute rather than a `--step-*`.
1140    //
1141    // A floor and not a height: a chip with two lines of text or a larger root
1142    // size still grows. `.chip` is already `inline-flex` with its items
1143    // centred, from quasi-webview's box rules, so the extra block size opens
1144    // around the label rather than under it. `.chip.latched` takes it through
1145    // the same class; `.badge` is a separate class and deliberately does not,
1146    // because a badge is a label and never a target.
1147    let _ = writeln!(
1148        css,
1149        ".{chip} {{\n    min-block-size: {};\n}}",
1150        opts.target_min
1151    );
1152
1153    css.push_str(&depth_rule(&chip, unlatched.depth(false)));
1154    css.push_str(&interactive_rules(&chip, unlatched.depth(false), opts));
1155    css.push_str(&depth_rule(
1156        &format!("{chip}.latched"),
1157        unlatched.depth(true),
1158    ));
1159    css
1160}
1161
1162/// The three selectors, each named by what it picks.
1163///
1164/// A tab comes *forward* to join the pane it opens, which is why
1165/// [`Selector::Tabs`] chooses [`Depth::Raised`] where a segment and a toggle
1166/// are held in. That is the folder semantic, and it is the whole reason the
1167/// three are not one member with a flag.
1168///
1169/// [`Selector::abutting`] is not emitted: whether the options touch is
1170/// spacing, and spacing is `makeover-geometry`'s question to answer.
1171///
1172/// Both states emit. Naming only the chosen option leaves an unchosen one
1173/// falling through to [`Depth::Flat`] with nothing drawn for it, so an app has
1174/// to hand-write the recess that makes its chosen tab read as forward.
1175fn selector_rules(opts: &Emit) -> String {
1176    let mut css = String::new();
1177    for selector in [Selector::Tabs, Selector::Segmented, Selector::Toggle] {
1178        let c = class(option_class(selector), opts);
1179        css.push_str(&depth_rule(&c, selector.unchosen()));
1180        css.push_str(&interactive_rules(&c, selector.unchosen(), opts));
1181        if selector.joins_its_pane() {
1182            // Every tab in a folder strip rounds at the top and stays square
1183            // at the bottom, chosen or not: the corners belong to the strip's
1184            // shape rather than to which option is up. On the base class and
1185            // not only on the chosen arm, so that an app has no reason left to
1186            // carry a radius of its own. goingson's unlayered `.tab` rule beat
1187            // a chosen-only version of this outright, an unlayered rule being
1188            // ahead of every layer whatever its specificity.
1189            let _ = writeln!(
1190                css,
1191                ".{c} {{\n    border-radius: var(--radius-control) var(--radius-control) 0 0;\n}}"
1192            );
1193            css.push_str(&joined_rule(&c, selector.chosen(), opts));
1194        } else {
1195            css.push_str(&depth_rule(&format!("{c}.chosen"), selector.chosen()));
1196        }
1197    }
1198    css
1199}
1200
1201/// The chosen option of a selector that joins what it opens.
1202///
1203/// [`depth_rule`] draws a closed box, which is right for anything standing on
1204/// its own and wrong for a folder tab: a closed bottom run reads as a chip
1205/// resting near its pane rather than as the front of it. Three things make the
1206/// join, and all three are needed, which is why this is one rule rather than a
1207/// property added to the depth:
1208///
1209/// - The depth's own fill, so the tab and the pane are the same surface. Taken
1210///   from [`Depth::fill`] rather than named here, so the pairing rule holds.
1211/// - [`Bevel::RaisedOpen`], so the run between the two is not drawn.
1212/// - A pull down by one border width, so the tab covers the pane's own top
1213///   run where it passes underneath. Without it the pane's lit edge draws
1214///   straight across the join and the tab reads as sitting on a line.
1215///
1216/// The corners round at the top and stay square at the bottom for the same
1217/// reason: a rounded bottom corner puts a notch of page colour either side of
1218/// the join. `makeover-geometry` has no per-corner member and does not need
1219/// one, since which corners round is a consequence of the join rather than a
1220/// scale a consumer picks.
1221fn joined_rule(selector: &str, depth: Depth, opts: &Emit) -> String {
1222    let mut body = String::new();
1223    if let Some(fill) = depth.fill() {
1224        let _ = writeln!(body, "    background: {};", fill_var(fill));
1225    }
1226    let _ = writeln!(
1227        body,
1228        "    box-shadow: var({});",
1229        bevel_var(Bevel::RaisedOpen)
1230    );
1231    // No radius here. Every tab in a joining strip already rounds at the top
1232    // from the base rule, chosen or not, and repeating it on this arm would be
1233    // a second copy of one decision.
1234    let _ = writeln!(
1235        body,
1236        "    margin-block-end: calc(-1 * {});",
1237        opts.border_width
1238    );
1239    // And it has to paint above what it overlaps. Both the tab and the pane are
1240    // in flow, the pane is the later sibling, so without this the pane draws
1241    // over the run the pull-down just covered and the join is undone by the
1242    // pane's own lit edge. Positioned, not raised by `z-index`: being
1243    // positioned is enough to paint above an unpositioned sibling, and a stack
1244    // index here would be a number this crate has no way to keep true against
1245    // whatever an app stacks around it.
1246    let _ = writeln!(body, "    position: relative;");
1247    let mut css = format!(".{selector}.chosen {{\n{body}}}\n");
1248
1249    // A strip that runs down its pane joins it on the side, and the join above
1250    // is written for a strip that runs across. Which way a strip runs is not
1251    // on [`Selector`], and it should not be: it is the arrangement, and the
1252    // renderer drawing the arrangement is the one that knows.
1253    //
1254    // `aria-orientation` is how that renderer has to say it anyway. It is
1255    // ARIA's word rather than a private attribute, required on any vertical
1256    // tablist for a screen reader, so keying on it couples this to the
1257    // standard and not to one caller's markup. Until the side join is drawn,
1258    // a vertical strip takes the closed edge back and reads as a raised chip
1259    // beside its pane, which is what it drew before this rule existed.
1260    let _ = writeln!(
1261        css,
1262        "[aria-orientation=\"vertical\"] .{selector}.chosen {{\n    \
1263         box-shadow: var({});\n    border-radius: var(--radius-control);\n    \
1264         margin-block-end: 0;\n}}",
1265        bevel_var(Bevel::Raised)
1266    );
1267    css
1268}
1269
1270/// The parts of a list row.
1271///
1272/// The list is written out rather than derived because `RowPart` is
1273/// `#[non_exhaustive]`, so there is nothing to iterate. A member added upstream emits no rule until it is named here, which
1274/// is the trade `non_exhaustive` makes: a silent gap instead of a build break.
1275/// [`part_class`] carries the same list and the same obligation.
1276fn row_rules(opts: &Emit) -> String {
1277    let mut css = String::new();
1278
1279    // The container the rows sit in, giving back what a `<ul>` brought. Not a
1280    // size: there is no magnitude in it, which is the line this crate holds.
1281    css.push_str(&Reset::BULLETS.rule(&format!(".{}", class("list", opts))));
1282
1283    for part in [
1284        RowPart::Primary,
1285        RowPart::Secondary,
1286        RowPart::Meta,
1287        RowPart::Actions,
1288        RowPart::Tokens,
1289        RowPart::Proportion,
1290    ] {
1291        let c = class(part_class(part), opts);
1292
1293        // Actions carry controls rather than text, and `RowPart::intent` says
1294        // so by returning the same intent inheriting already gives. Pinning it
1295        // would be louder than saying nothing. Tokens answer alike, for their
1296        // own reason: each token carries its own tone, and a colour on the
1297        // strip would fight the things sitting in it. A proportion is the same
1298        // case again: the meter inside carries the tone.
1299        if !matches!(
1300            part,
1301            RowPart::Actions | RowPart::Tokens | RowPart::Proportion
1302        ) {
1303            let _ = writeln!(css, ".{c} {{\n    color: var(--{});\n}}", part.intent());
1304        }
1305
1306        // A row's actions are shown at rest. `RowPart::revealed_on_hover` said
1307        // otherwise and was not honoured here from 0.23.0; makeover-layout
1308        // 0.13.0 retired the method, so there is no longer a description saying
1309        // one thing and a renderer doing another.
1310        //
1311        // The rule was `opacity: 0` gated to pointer devices, revealed on
1312        // `:hover` and on `:focus-within`. Each escape it needed was a report
1313        // that hiding was wrong for somebody: `focus-within` because tabbing
1314        // could never reach an action; the gate because a fingertip had no way
1315        // to unhide, which both webview apps had already hand-written
1316        // `opacity: 1` to undo. What was left was a control hidden from
1317        // exactly one group: people using a pointer, who are also the group
1318        // scanning a list to find out what can be done to a row.
1319        //
1320        // A settings screen is where that reads worst: the whole reason to be
1321        // on it is to remove a key, and the button doing so was invisible
1322        // until pointed at. A table cell's actions were never hidden, so the
1323        // two arrangements now agree.
1324    }
1325
1326    // A part that may take two lines. `Flow::Tight` gets no rule: one line is
1327    // what a run already does, and restating it here would put a declaration on
1328    // every part in every row to say nothing.
1329    //
1330    // This is the shape both webview apps had already written by hand and
1331    // commented -- Balanced Breakfast on a feed row's title, goingson on a
1332    // problem's body -- which is the whole argument for the description
1333    // carrying it. `-webkit-` prefixed and unprefixed together: the prefixed
1334    // trio is what every engine actually implements, and `line-clamp` is the
1335    // standard property landing behind it.
1336    let _ = writeln!(
1337        css,
1338        ".{} {{\n    display: -webkit-box;\n    -webkit-box-orient: vertical;\n    -webkit-line-clamp: {lines};\n    line-clamp: {lines};\n    overflow: hidden;\n}}",
1339        class("row-relaxed", opts),
1340        lines = Flow::Relaxed.lines()
1341    );
1342
1343    // A row inside a hierarchy: a tree, an outline, a threaded list.
1344    // `edf33114`, decided 2026-08-30 (Max). makeover-layout names the concept
1345    // as `Nesting` -- deliberately not `Depth`, which is surface bevel in that
1346    // crate -- and this is the rule.
1347    //
1348    // Measured 2026-08-30: quasi-webview has been emitting `row-nested` with
1349    // `style="--row-depth:N"` on every described hierarchy, and **no stylesheet
1350    // anywhere in the tree read any of it**. So a described outline in a
1351    // browser was a flat list with chevrons in it -- the folding worked and the
1352    // indent did not exist.
1353    //
1354    // The magnitude is a custom property with a fallback, which is this crate's
1355    // own shape: `--awaiting-gap` above is the precedent, and an app overriding
1356    // `--row-indent` is how a level becomes worth more or less. What a level IS
1357    // stays the description's; what it is WORTH is a renderer's, and a terminal
1358    // spending columns for the same fact is not disagreeing.
1359    //
1360    // Here rather than in each app, and that was a live option rather than an
1361    // oversight. `row-select`, `row-current` and `row-chosen` are app-styled by
1362    // design; the indent is not decoration, it is what the description MEANS,
1363    // and three apps agreeing about it by accident is not agreement.
1364    //
1365    // `padding` and not `margin`: a row is a box that can be selected and
1366    // hovered, and indenting with margin would take the indent out of the
1367    // highlight, so the shading under a nested row would start where its text
1368    // does rather than where its row does.
1369    let _ = writeln!(
1370        css,
1371        ".{} {{\n    padding-inline-start: calc(var(--row-depth, 0) * var(--row-indent, 1.5ch));\n}}",
1372        class("row-nested", opts)
1373    );
1374
1375    // The two halves of a branch, emitted by quasi-webview and unstyled until
1376    // now for the same reason.
1377    //
1378    // A branch row is the one a reader can fold, and the chevron is its hit
1379    // target. The chevron is drawn by the app or the description -- this says
1380    // where it sits and how big the target is, which is the accessibility fact
1381    // rather than the decorative one: a control smaller than this is one a
1382    // finger misses.
1383    let _ = writeln!(
1384        css,
1385        ".{} {{\n    display: flex;\n    align-items: baseline;\n    gap: var(--row-disclose-gap, 0.5ch);\n}}",
1386        class("row-branch", opts)
1387    );
1388    let _ = writeln!(
1389        css,
1390        ".{} {{\n    flex: none;\n    min-inline-size: var(--tap-target, 2rem);\n    min-block-size: var(--tap-target, 2rem);\n    background: none;\n    border: 0;\n    color: inherit;\n    cursor: pointer;\n}}",
1391        class("row-disclose", opts)
1392    );
1393
1394    // How a part holds what is in it, for a row and for a table cell. These
1395    // were quasi-webview's arrangement sheet's until 0.82.0, which ruled names
1396    // this crate emits, so a row an app wrote by hand got none of it. The row
1397    // itself stays quasi-webview's: where a part sits in the line is the screen
1398    // renderer's arrangement, and nothing here says it.
1399    //
1400    // A token or a control never breaks across two lines. What gives when the
1401    // line runs short is the row's text.
1402    let tokens = class(part_class(RowPart::Tokens), opts);
1403    let actions = class(part_class(RowPart::Actions), opts);
1404    let proportion = class(part_class(RowPart::Proportion), opts);
1405    let fallback = class("row-part", opts);
1406    let cell_tokens = class(cell_part_class(CellPart::Tokens), opts);
1407    let cell_actions = class(cell_part_class(CellPart::Actions), opts);
1408    let cell_fallback = class("cell-part", opts);
1409    let _ = writeln!(
1410        css,
1411        ".{fallback},\n.{cell_fallback} {{\n    min-width: 0;\n}}"
1412    );
1413    let _ = writeln!(
1414        css,
1415        ".{tokens} {{\n    display: flex;\n    flex-wrap: wrap;\n    align-items: center;\n    gap: var(--gap-bound);\n}}"
1416    );
1417    let _ = writeln!(
1418        css,
1419        ".{actions} {{\n    display: flex;\n    flex: none;\n    align-items: center;\n    gap: var(--gap-bound);\n}}"
1420    );
1421    // A cell's parts sit inside the cell's own box, so they are inline: the
1422    // cell's alignment, which the column kind sets, still places them.
1423    let _ = writeln!(
1424        css,
1425        ".{cell_tokens},\n.{cell_actions} {{\n    display: inline-flex;\n    align-items: center;\n    gap: var(--gap-bound);\n}}"
1426    );
1427    let _ = writeln!(css, ".{cell_tokens} {{\n    flex-wrap: wrap;\n}}");
1428    let _ = writeln!(
1429        css,
1430        ".{tokens} > *,\n.{actions} > *,\n.{cell_tokens} > *,\n.{cell_actions} > * {{\n    white-space: nowrap;\n}}"
1431    );
1432    // A proportion is a meter, and a meter is as wide as the row lets it be.
1433    let _ = writeln!(css, ".{proportion} {{\n    flex: 1 1 auto;\n}}");
1434
1435    css
1436}
1437
1438/// A row of things that share their space, and what each fallback gets here.
1439///
1440/// Ruling: wiki `layout-room-and-fallback`, Max. Rule 1 is that every described
1441/// member is in flow, and these rules are how that is kept rather than asked
1442/// for. A member taken out of flow with `position: absolute` contributes zero
1443/// width to the row it shares, so nothing can collide with it and nothing
1444/// prevents the collision.
1445///
1446/// # The floor, which is most of the fix
1447///
1448/// `.run > *` gets `min-width: min-content`. That is the derived minimum the
1449/// ruling asks for, in this renderer's own unit and stated by the browser
1450/// rather than by anybody: a member cannot be squeezed narrower than what is
1451/// in it, so members in one flow push each other instead of overlapping. It
1452/// costs no query and no number, and it is what fixes all four measured widths
1453/// whichever fallback the group declared.
1454///
1455/// # A member that asks to fill
1456///
1457/// `.run > [data-width="fill"]` gets `flex: 1 1 0`, which is the second half of
1458/// what a column has always been able to say, reaching a row of regions.
1459/// `flex-basis: 0` and not `auto` is what makes several fills divide the room
1460/// equally rather than dividing the leftovers in proportion to their contents;
1461/// equal division is [`makeover_layout::Width::Fill`]'s own stated rule. The
1462/// floor above still applies, so a fill cannot shrink under what is in it.
1463///
1464/// A member that says nothing gets nothing, because a flex item with the floor
1465/// and no grow is already content-sized. That is why the omitted value here is
1466/// `Content` while a control omits `Fill`: each position leaves out what it
1467/// already did.
1468///
1469/// # What each fallback gets, exactly
1470///
1471/// [`Fallback::Wrap`] is `flex-wrap: wrap`, and a member that asks to fill gets
1472/// a stated room to start from.
1473///
1474/// Wrapping on its own was not exact, which `4f5705b1` measured. A flex line
1475/// breaks on each member's hypothetical size, a fill member's basis is `0`, and
1476/// the floor above is `min-content`, so the line breaks only once the members'
1477/// intrinsic widths no longer fit. A member holding a table reports almost
1478/// nothing for that: a table's headings and cells are `container-type:
1479/// inline-size`, so the size container refuses to be measured through and a few
1480/// headings are all that is left. Two fill members each holding a table drew as
1481/// two slivers side by side at 420 while both tables overflowed, and the page
1482/// scrolled sideways.
1483///
1484/// So under `Wrap` a fill member's `flex-basis` is
1485/// `min(100%, var(--run-room, 20rem))` rather than `0`: room the description
1486/// states, rather than room measured through a container that will not report
1487/// it. `min(100%, ...)` is what keeps a lone member from overflowing a run
1488/// narrower than the room, and it is the shape quasi's own `main.list-detail`
1489/// regions already fold on. The custom property is how an app says what the
1490/// room is, because the number is a size and sizes are the app's; the fallback
1491/// is a default rather than this crate settling a width.
1492///
1493/// Only under `Wrap`. A fill member in a run that does not wrap keeps
1494/// `flex: 1 1 0`, where equal division is the whole of what it asked for and
1495/// there is no line to break.
1496///
1497/// [`Fallback::Stack`] is wrap plus `flex: 1 1 max-content` on the members, so
1498/// a member that cannot sit beside its sibling takes a line of its own and
1499/// fills it. For the two-member run this was ruled on -- a tab strip and a
1500/// band -- that is precisely "a row becomes a column".
1501///
1502/// [`Fallback::Shed`] and [`Fallback::Menu`] get wrap, and this renderer is
1503/// honouring less than the description says. **CSS cannot express either one
1504/// without breaking the ruling's own first constraint.** Both need to know that
1505/// the run is out of room in order to take a member out of it, a container
1506/// query is the only construct that can ask, and `@container` compares against
1507/// a `<length>` -- there is no `@container (inline-size < min-content)`. So
1508/// every honest spelling of Shed here needs an authored breakpoint, which is
1509/// the thing the ruling exists to forbid, and the dishonest ones are worse: a
1510/// clamped height clips by document order rather than by [`Priority`], and
1511/// `display: none` under a viewport `@media` is the `nth-child(n+5)` bug the
1512/// vocabulary replaced.
1513///
1514/// Wrapping is the right thing to do instead. It keeps every member reachable,
1515/// which is the property that was actually broken -- goingson's new-contact
1516/// button left the viewport entirely at 560 -- and it keeps rule 1. A renderer
1517/// answering with less than was described is precedented and deliberate here:
1518/// [`makeover_layout::Region::Columns`] says a terminal stacking a board's
1519/// columns is honouring the description rather than degrading it.
1520///
1521/// The real mechanism needs the shed members to have somewhere to go, which is
1522/// markup and belongs to quasi-webview: an overflow control is a member of the
1523/// run, and the description does not yet say that a member *is* one.
1524///
1525/// # Menu, once a script is measuring
1526///
1527/// That is now built, in `quasi-webview`'s `menu.js`, and this crate's half of
1528/// it is two classes and one override. A script that has taken a menu run over
1529/// marks it `data-menu`, and a marked run goes back to `nowrap`: wrapping is
1530/// what hides the overflow condition the script is trying to measure. The
1531/// unmarked rule above is untouched, so a page that ships no script still
1532/// wraps, which is rule 1 — every member reachable — rather than a strip with
1533/// tabs squeezed off the end.
1534///
1535/// `.run-overflow` is the control the shed members move into and
1536/// `.run-overflow-items` is where they land. The geometry is this crate's the
1537/// way every other surface's is; what is *in* it is the script's, because which
1538/// members no longer fit is a measurement and not a description.
1539fn run_rules(opts: &Emit) -> String {
1540    let run = class("run", opts);
1541    let mut css = String::new();
1542
1543    // `flex-wrap: nowrap` is stated rather than left to the default, because
1544    // the fallbacks below are read as overrides of this line and a reader
1545    // should not have to know which way flexbox leans to see that.
1546    //
1547    // No gap. Spacing between members is the app's, the same way this crate
1548    // states no margins anywhere else; a gap here would be a size, and the one
1549    // hardcoded size in the mechanism is makeover-geometry's contact patch.
1550    let _ = writeln!(
1551        css,
1552        ".{run} {{\n    display: flex;\n    flex-wrap: nowrap;\n    align-items: center;\n}}"
1553    );
1554
1555    // The derived minimum, and the whole reason a member can no longer be
1556    // overlapped. `min-width: auto` is flexbox's default for a flex item and is
1557    // *not* the same thing: auto lets an item be compressed below its content
1558    // in a nowrap run, which is how a toolbar ends up drawn over a tab strip
1559    // even without anything leaving the flow.
1560    let _ = writeln!(css, ".{run} > * {{\n    min-width: min-content;\n}}");
1561
1562    // A member that absorbs what is left. `flex-basis: 0` rather than `auto` is
1563    // what makes several fills divide the room equally instead of dividing the
1564    // leftovers in proportion to what is already in them, which is
1565    // `Width::Fill`'s own rule and the one thing that type states about more
1566    // than one of them.
1567    //
1568    // The `min-width: min-content` floor above is deliberately not overridden.
1569    // A fill that could shrink below its contents would overlap its neighbour,
1570    // which is rule 1, and equal division under a floor is still equal division
1571    // everywhere the floor is not reached.
1572    //
1573    // Attribute rather than class, because the width is a fact the description
1574    // carried rather than a hook this crate invented: the same division
1575    // `data-tone` and `data-selector` are on the right side of. It beats the
1576    // `Stack` rule below on specificity whichever order they are written in,
1577    // which is what a member asking to fill should do to a blanket.
1578    let _ = writeln!(
1579        css,
1580        ".{run} > [data-width=\"fill\"] {{\n    flex: 1 1 0;\n}}"
1581    );
1582
1583    for fallback in [
1584        Fallback::Wrap,
1585        Fallback::Stack,
1586        Fallback::Shed,
1587        Fallback::Menu,
1588    ] {
1589        let name = fallback_class(fallback);
1590        let c = class(name, opts);
1591        let _ = writeln!(css, ".{c} {{\n    flex-wrap: wrap;\n}}");
1592        if matches!(fallback, Fallback::Stack) {
1593            let _ = writeln!(css, ".{c} > * {{\n    flex: 1 1 max-content;\n}}");
1594        }
1595        // The stated room, and only here. Written after the blanket fill rule
1596        // above so it overrides the basis it set; the grow and shrink that rule
1597        // states are deliberately left alone, so several fills still divide one
1598        // line equally and the min-content floor still stops any of them
1599        // shrinking under its contents.
1600        if matches!(fallback, Fallback::Wrap) {
1601            let _ = writeln!(
1602                css,
1603                ".{c} > [data-width=\"fill\"] {{\n    \
1604                 flex-basis: min(100%, var(--run-room, 20rem));\n}}"
1605            );
1606        }
1607    }
1608
1609    // A menu run a script has taken over. The mark is the script's and this is
1610    // the only rule that reads it: wrapping is what a run does when nothing is
1611    // measuring, and it is also what makes the overflow unmeasurable, since a
1612    // wrapped run always fits. The two cannot both be on.
1613    let menu = class(fallback_class(Fallback::Menu), opts);
1614    let _ = writeln!(css, ".{menu}[data-menu] {{\n    flex-wrap: nowrap;\n}}");
1615
1616    // The overflow control, and it is a member of the run like any other: in
1617    // flow, at the end, taking the width of what is in it. `relative` is what
1618    // the items hang off.
1619    let overflow = class("run-overflow", opts);
1620    let items = class("run-overflow-items", opts);
1621    let _ = writeln!(css, ".{overflow} {{\n    position: relative;\n}}");
1622
1623    // Overlaid rather than in flow, for the reason every menu is: a control
1624    // that pushed the page down when it opened would change the layout it was
1625    // opened to escape. `inset-inline-end: 0` rather than a left, so the panel
1626    // stays on the page in both writing directions.
1627    //
1628    // No width, no padding and no border. All three are sizes and sizes are
1629    // makeover-geometry's; what is stated here is placement, the surface and
1630    // the shadow that separates it from the page, which is the same division
1631    // `figure_rules` and the timeline entry make. The elevation shadow is what
1632    // an overlaid surface takes instead of an edge -- see `ELEVATION_PROPERTY`.
1633    let _ = writeln!(
1634        css,
1635        ".{items} {{\n    \
1636         position: absolute;\n    \
1637         inset-block-start: 100%;\n    \
1638         inset-inline-end: 0;\n    \
1639         z-index: 1;\n    \
1640         display: flex;\n    \
1641         flex-direction: column;\n    \
1642         align-items: stretch;\n    \
1643         background: var(--surface-raised);\n    \
1644         box-shadow: var(--elevation-overlay);\n\
1645         }}"
1646    );
1647
1648    // `hidden` is how the script closes it, and a flex display would otherwise
1649    // beat the attribute's own `display: none`.
1650    let _ = writeln!(css, ".{items}[hidden] {{\n    display: none;\n}}");
1651
1652    css
1653}
1654
1655/// Every class [`fallback_class`] can return, plus the run itself.
1656///
1657/// [`ROW_PART_CLASSES`](crate::list::ROW_PART_CLASSES)'s reasoning and the same
1658/// obligation: a `match` over a `#[non_exhaustive]` enum cannot be enumerated
1659/// from outside, so the list sits beside it and a test holds the two together.
1660/// `run` is in it because it is emitted in its own right rather than only as a
1661/// fallback's fallback.
1662pub const RUN_CLASSES: &[&str] = &[
1663    "run",
1664    "run-wrap",
1665    "run-stack",
1666    "run-shed",
1667    "run-menu",
1668    // Not returned by `fallback_class`: these two are the overflow control a
1669    // measuring renderer builds, and they are in the list because the list is
1670    // what a host seals its vocabulary against. A class emitted by a script and
1671    // missing from here is a control with no surface and no edge.
1672    "run-overflow",
1673    "run-overflow-items",
1674];
1675
1676/// The class a run carries for what it does when it is tight.
1677///
1678/// A run always carries `.run` as well, so an unrecognised fallback -- the enum
1679/// is `#[non_exhaustive]` -- lands as a plain nowrap row with the min-content
1680/// floor still under it. That is the safe failure: every member in flow and
1681/// none overlapped, which is the property, with only the rearrangement missing.
1682#[must_use]
1683pub fn fallback_class(fallback: Fallback) -> &'static str {
1684    match fallback {
1685        Fallback::Wrap => "run-wrap",
1686        Fallback::Stack => "run-stack",
1687        Fallback::Shed => "run-shed",
1688        Fallback::Menu => "run-menu",
1689        _ => "run",
1690    }
1691}
1692
1693/// The progress trough these rules fill.
1694///
1695/// [`meter::meter_html`](crate::meter::meter_html) is what fills these.
1696///
1697/// The rules stay a superset of what a description can ask for. An app drawing
1698/// its own bar keeps these classes, which is what the four goingson grew
1699/// independently were adopted onto.
1700///
1701/// The trough is a [`Depth::Well`], the same reading a text field gets:
1702/// something with its content down inside it.
1703fn progress_rules(opts: &Emit) -> String {
1704    let progress = class("progress", opts);
1705    // `progress-fill` rather than a bare `fill`: an unprefixed build claims
1706    // these names in the app's own stylesheet, and `.fill` is grabby enough to
1707    // catch things that have nothing to do with progress. goingson already
1708    // calls it `.progress-fill`, so this is also the name that deletes.
1709    let fill = class("progress-fill", opts);
1710    let mut css = depth_rule(&progress, Depth::Well);
1711
1712    // The trough's block size, and the fill filling it. The fill is an empty
1713    // block, so a trough nothing sized had no height and the meter drew nothing:
1714    // goingson carried the size in its own sheet, and MNW, which did not, drew
1715    // its Cloud Sync storage meter as a blank gap. Width stays the app's, below;
1716    // a height is not taste, it is whether the bar exists.
1717    let _ = writeln!(
1718        css,
1719        ".{progress} {{\n    block-size: var(--meter-block, 0.625rem);\n    overflow: hidden;\n}}"
1720    );
1721
1722    // The untoned bar is `--action`, not [`Tone::Neutral`]. That is the one
1723    // place this differs from the badge rules, and deliberately: a badge with
1724    // no status is a muted label, while a bar with no status is still
1725    // reporting progress, and `content-muted` would read as disabled.
1726    // The width is the fill the meter emitter hands over in `data-vars`. The
1727    // fallback is an empty bar rather than a full one: an unset width on a block
1728    // is the whole trough, which would read as done.
1729    //
1730    // On the fill alone, not under `.progress >`: a rule names every class in
1731    // its selector to a consumer's drift check, and the trough's width is the
1732    // app's to set. goingson sizes `.progress` and failed its build on a width
1733    // this rule never gave the trough.
1734    let _ = writeln!(
1735        css,
1736        ".{fill} {{\n    width: var(--meter-fill, 0%);\n    block-size: 100%;\n}}"
1737    );
1738    let _ = writeln!(
1739        css,
1740        ".{progress} > .{fill} {{\n    background: var(--action);\n}}"
1741    );
1742
1743    // A bar can be saying something, same as a badge: goingson colours subtask
1744    // progress as success and an over-estimate as danger, which is real
1745    // information rather than decoration. Emitting the tones is what lets that
1746    // survive adoption instead of staying hand-written.
1747    for tone in [Tone::Info, Tone::Success, Tone::Warning, Tone::Danger] {
1748        let _ = writeln!(
1749            css,
1750            ".{progress} > .{fill}[data-tone=\"{0}\"] {{\n    background: var(--{0});\n}}",
1751            tone.token()
1752        );
1753    }
1754    css
1755}
1756
1757/// What a wait looks like, for the attribute that has been describing one to
1758/// nobody.
1759///
1760/// Wiki `loading-and-progress-standard`, phase 2. `data-awaiting` is emitted
1761/// as `data-awaiting="determinate"` with a `data-awaiting-amount` beside it, or
1762/// `data-awaiting="indeterminate"` alone. This is the half that styles them,
1763/// without which the two render identically.
1764///
1765/// # Why it is keyed on `aria-busy` and not on the attribute alone
1766///
1767/// `data-awaiting` is a fact about the control: pressing this waits. It is true
1768/// when the page is painted and it stays true. Whether a wait is *running* is
1769/// true only between two events, so it is the binder's to set, and `aria-busy`
1770/// is the standard spelling of it — announced as well as drawn, which a class
1771/// of our own would not be.
1772///
1773/// That also keeps this crate out of any one client library's vocabulary.
1774/// `quasi-webview` sets `aria-busy` from htmx's request events; a host driving
1775/// the same markup another way sets it the same way and gets the same drawing.
1776///
1777/// # The two drawings
1778///
1779/// One pseudo-element either way, so no renderer has to emit an extra node.
1780///
1781/// Indeterminate is the activity mark of rule 2: a small square that blinks.
1782/// Determinate is a trough with a fill, drawn as a single gradient whose stop is
1783/// `--awaiting-share`, a plain number from 0 to 1 that the binder sets from
1784/// bytes it has actually watched land. A determinate control with nothing
1785/// setting the share draws an empty trough rather than a full one, which is the
1786/// honest reading: the size is known and the delivery is not.
1787///
1788/// **What the bar may not do**, from rule 1 and from `Awaiting`'s own docs:
1789/// what is done over what there is, and never a remaining time, an arrival time
1790/// or a rate extrapolated forward. Nothing here can express one, which is
1791/// deliberate — the only input is a share of a measured payload.
1792///
1793/// # Sizes, and the deferral rule
1794///
1795/// `progress_rules` emits the tones and never the width, because the width is
1796/// not this crate's to know. A pseudo-element has no intrinsic size at all, so
1797/// the same treatment would render nothing anywhere. Both sizes are therefore
1798/// custom properties with defaults: an app that wants a different mark sets
1799/// `--awaiting-mark` and `--awaiting-bar` once, and one that says nothing gets a
1800/// mark that is visible.
1801///
1802/// # The cadence, and what happens without it
1803///
1804/// `--cadence-activity` comes from `makeover-timing` through `makeover-build`,
1805/// and is a half-period, so a full cycle is twice it. It is used bare rather
1806/// than with a fallback: a number written here would be a second heartbeat for
1807/// a mark three renderers draw.
1808///
1809/// A sheet assembled without the time axis leaves `animation-duration` invalid,
1810/// which resolves to `0s`, which runs no animation and leaves the base style —
1811/// a lit, still mark. That is also exactly what the reduced-motion block does,
1812/// since it sets the cadence to `0ms`. Both fall out of one rule because the
1813/// base style is lit and the keyframes do the dimming, which is the ordering
1814/// `makeover_timing::reduced_motion_css` asks its consumers for by name.
1815fn awaiting_rules(opts: &Emit) -> String {
1816    let mut css = String::new();
1817
1818    // Dimming rather than lighting, so a zero-length animation leaves a lit
1819    // mark rather than a blank one. See `makeover_timing::reduced_motion_css`.
1820    css.push_str(
1821        "@keyframes makeover-activity {\n    \
1822         0%, 49.99% {\n        background: var(--action);\n    }\n    \
1823         50%, 100% {\n        background: var(--surface-sunken);\n    }\n\
1824         }\n",
1825    );
1826
1827    // Nothing is drawn until something is waiting. `content` on the base rule
1828    // rather than on the busy one keeps the box the same box across the
1829    // transition, so a mark appearing does not reflow the line it is in.
1830    let _ = writeln!(
1831        css,
1832        "[data-awaiting]::after {{\n    \
1833         content: \"\";\n    \
1834         display: none;\n    \
1835         margin-inline-start: var(--awaiting-gap, 0.5ch);\n    \
1836         vertical-align: baseline;\n\
1837         }}"
1838    );
1839
1840    let _ = writeln!(
1841        css,
1842        "[data-awaiting][aria-busy=\"true\"]::after {{\n    \
1843         display: inline-block;\n    \
1844         inline-size: var(--awaiting-mark, 0.5em);\n    \
1845         block-size: var(--awaiting-mark, 0.5em);\n    \
1846         background: var(--action);\n    \
1847         opacity: 1;\n    \
1848         animation: makeover-activity calc(var(--cadence-activity) * 2) \
1849         step-end infinite;\n\
1850         }}"
1851    );
1852
1853    // The measured half. A wider box, no blink, and a gradient whose stop is
1854    // the share: the fill and the trough in one paint, so the markup stays one
1855    // pseudo-element on both branches.
1856    //
1857    // `--awaiting-share` unset is an empty trough, not a full one. A bar that
1858    // read full because nobody was counting would be the confidently-wrong
1859    // drawing rule 1 exists to forbid.
1860    //
1861    // The trough takes an edge for the reason a well does: a bar at zero share
1862    // is otherwise a rectangle of the surface it sits on, which is nothing at
1863    // all. `border_width` rather than a literal, the way `depth_rule` and
1864    // `track_rules` write theirs.
1865    let _ = writeln!(
1866        css,
1867        "[data-awaiting=\"determinate\"][aria-busy=\"true\"]::after {{\n    \
1868         inline-size: var(--awaiting-bar, 6em);\n    \
1869         animation: none;\n    \
1870         outline: {} solid var(--border);\n    \
1871         outline-offset: -{};\n    \
1872         background: linear-gradient(\n        \
1873         to inline-end,\n        \
1874         var(--action) 0 calc(var(--awaiting-share, 0) * 100%),\n        \
1875         var(--surface-sunken) 0\n    \
1876         );\n\
1877         }}",
1878        opts.border_width, opts.border_width
1879    );
1880
1881    css
1882}
1883
1884/// A strip of figures, and the two spans inside each one.
1885///
1886/// Colour only, which is the deferral rule applied to a component that badly
1887/// wants to break it. A figure reads as a figure because the value is set large
1888/// over a small caption, and that is a size: `makeover-geometry` answers how
1889/// much space and this crate answers what the thing is. Emitting `font-size`
1890/// here would be this crate naming a value, which is the one thing it is defined
1891/// by not doing, and `progress_rules` is the precedent — it emits the tones and
1892/// never the width, because the width is not its to know.
1893///
1894/// So the type scale is the app's, and what is generated is the part an app
1895/// cannot get right by itself: which of the two spans carries the tone, and,
1896/// from 0.82.0, the strip the figures sit in.
1897/// A definition list, and the two tracks that make its values line up.
1898///
1899/// **The alignment is the whole member.** Five goingson panes were saying
1900/// labelled facts as a row each, and rendered that way every value started
1901/// after its own label, so no column formed and the eye had nothing to run
1902/// down. A grid with one label track sized to the longest label is the fix, and
1903/// it is something only the set can do -- which is why `Node::Facts` carries
1904/// the set rather than each pair.
1905///
1906/// `max-content` on the first track and not a fixed width: the label column is
1907/// as wide as the longest label and no wider, which is what keeps a pane of
1908/// short labels from leaving a canyon before the values.
1909///
1910/// The value takes `content` and the label reads back. goingson had that the
1911/// other way round -- muted values against unmuted labels -- so the part a
1912/// person came for was the quieter of the two. Same rule `figure_rules` states
1913/// for a value against its caption, and `row` for primary against meta.
1914fn facts_rules(opts: &Emit) -> String {
1915    let facts = class("facts", opts);
1916    let label = class("facts-label", opts);
1917    let value = class("facts-value", opts);
1918    let mut css = String::new();
1919
1920    let _ = writeln!(
1921        css,
1922        ".{facts} {{\n    display: grid;\n    grid-template-columns: max-content 1fr;\n    \
1923         column-gap: var(--gap-section);\n    row-gap: var(--gap-peer);\n    margin: 0;\n}}"
1924    );
1925    // `dt` and `dd` both carry a browser margin that would break the tracks.
1926    let _ = writeln!(
1927        css,
1928        ".{facts} > .{label} {{\n    margin: 0;\n    color: var(--content-secondary);\n}}"
1929    );
1930    let _ = writeln!(
1931        css,
1932        ".{facts} > .{value} {{\n    margin: 0;\n    color: var(--content);\n}}"
1933    );
1934    css
1935}
1936
1937fn figure_rules(opts: &Emit) -> String {
1938    let figures = class("figures", opts);
1939    let figure = class("figure", opts);
1940    let value = class("figure-value", opts);
1941    let caption = class("figure-caption", opts);
1942    let change = class("figure-change", opts);
1943    let mut css = String::new();
1944
1945    // A strip of tiles that wraps rather than overflows, a section apart. It
1946    // was quasi-webview's arrangement sheet's to say, for a name this crate
1947    // emits.
1948    let _ = writeln!(
1949        css,
1950        ".{figures} {{\n    display: flex;\n    flex-wrap: wrap;\n    gap: var(--gap-section);\n}}"
1951    );
1952
1953    // `content` literally. A figure's value is the thing itself, at full
1954    // weight -- wiki `three-tone-convention` classes it "active, emphasised",
1955    // and both other renderers already draw it that way (makeover-tui
1956    // `piece.rs:391` bold, makeover-immediate `widget.rs:244`). It reached
1957    // here through `Tone::Neutral.token()` and came out `content-muted`, so
1958    // the headline number sat at the colour of its own caption.
1959    let _ = writeln!(
1960        css,
1961        ".{figure} > .{value} {{\n    color: var(--content);\n}}"
1962    );
1963    let _ = writeln!(
1964        css,
1965        ".{figure} > .{caption} {{\n    color: var(--content-muted);\n}}"
1966    );
1967    let _ = writeln!(
1968        css,
1969        ".{figure} > .{change} {{\n    color: var(--content-muted);\n}}"
1970    );
1971
1972    // A toned figure tones one part and never the caption. The caption is the
1973    // noun and stays muted.
1974    //
1975    // Which part depends on whether there is a change, and that is the whole of
1976    // what 0.13.0 changed here. A figure with a delta is an ordinary number that
1977    // has moved in a direction worth reading, so the delta takes the colour and
1978    // the number stays plain; a figure without one has nowhere else to put it.
1979    // `:has` is what lets one attribute mean both, and the alternative was the
1980    // emitter deciding by writing the attribute onto a different element, which
1981    // leaves two elements able to disagree about a figure's one meaning.
1982    // The tone is an edge, and neither number is drawn in it. Both were until
1983    // 0.86.0, and both failed the same way `{tone}-surface`'s derivation
1984    // comment describes: measured on goingson, a toned value read 1.24:1, 1.53:1
1985    // and 2.27:1 against the ground it sits on. A figure's value is the
1986    // largest text on the screen and the one most worth reading, so it keeps
1987    // `content` whatever the tone is.
1988    //
1989    // An edge rather than the tinted fill a chip and a button take. A figure
1990    // has no box of its own -- it is a bare pair of spans in a flex strip --
1991    // and giving the toned ones a fill would give a strip of figures two
1992    // different shapes depending on which of them had news. The rule sits on
1993    // the side so a neutral figure is untouched and the tint never has to exist.
1994    //
1995    // This also retires the `:has` split. It existed to choose which of the two
1996    // elements took the colour; with neither taking it, a figure with a delta
1997    // and a figure without one want the same rule, and the two elements can no
1998    // longer disagree about one tone because neither states it.
1999    for tone in [Tone::Info, Tone::Success, Tone::Warning, Tone::Danger] {
2000        let _ = writeln!(
2001            css,
2002            ".{figure}[data-tone=\"{0}\"] {{\n    border-inline-start: {1} solid \
2003             var(--{0});\n    padding-inline-start: var(--step-snug);\n}}",
2004            tone.token(),
2005            opts.focus_width
2006        );
2007    }
2008    css
2009}
2010
2011/// A picture, its frame and its caption.
2012///
2013/// # The frame is a border, and this is the one place a bevel is wrong
2014///
2015/// Emitting [`Depth::Raised`] here through [`depth_rule`], the same call
2016/// `button` and `card` make, is wrong, and MNW's landing page is where it
2017/// showed: **the frames had no visible edge at all.**
2018///
2019/// `Depth::Raised`'s edge is `--bevel-raised`, which is an *inset* shadow — a
2020/// 1px light run at the top-left and a dark one at the bottom-right, drawn
2021/// **inside** the element's box. On a button or a card that box is a surface
2022/// this crate owns, so an inset edge reads as the surface catching the light.
2023/// On a picture it is drawn on top of the picture, over whatever pixels the
2024/// image happens to have at its border. MNW's screenshots are light-on-light
2025/// parchment, so the light half landed on a light image and the frame
2026/// disappeared.
2027///
2028/// **You cannot bevel a surface you do not own.** A picture's content is the
2029/// app's, arrives at request time, and can be any colour, so its edge has to
2030/// sit *outside* the content rather than on it. That is a border.
2031///
2032/// The fill stays, and it is not decoration: it is what shows through a
2033/// transparent PNG and what stands in the frame's place while the image is
2034/// still loading.
2035///
2036/// This is a real limit on [`Depth`] rather than a special case. Every other
2037/// consumer of a depth draws its own surface; a picture is the first member
2038/// whose surface belongs to someone else.
2039///
2040/// # No size
2041///
2042/// `width: 100%` and nothing else. How large a picture is depends on the box it
2043/// was put in, which is the app's arrangement and `makeover-geometry`'s scales,
2044/// and a renderer that picked one would be answering for every consumer at
2045/// once. This is where `figure_rules` landed for the same reason.
2046fn picture_rules(opts: &Emit) -> String {
2047    let picture = class("picture", opts);
2048    let img = class("picture-img", opts);
2049    let caption = class("picture-caption", opts);
2050    let mut css = String::new();
2051
2052    // Block, or an inline image sits on the text baseline and carries a
2053    // descender's worth of space under it that no app ever wants and every app
2054    // deletes by hand. The border is the frame; see the type docs for why it is
2055    // not the bevel every other surface here gets.
2056    let _ = writeln!(
2057        css,
2058        ".{img} {{\n    display: block;\n    width: 100%;\n    height: auto;\n    \
2059         background: var(--{});\n    border: {} solid var(--border);\n}}",
2060        Fill::Raised.token(),
2061        opts.border_width
2062    );
2063
2064    // The two fits that need a rule. `Fit::Natural` emits no attribute at all,
2065    // so it is the bare rule above and needs nothing here.
2066    let _ = writeln!(
2067        css,
2068        ".{img}[data-fit=\"cover\"] {{\n    height: 100%;\n    object-fit: cover;\n}}"
2069    );
2070    let _ = writeln!(
2071        css,
2072        ".{img}[data-fit=\"contain\"] {{\n    height: 100%;\n    object-fit: contain;\n}}"
2073    );
2074
2075    // A caption reads back one step, which is `figure-caption`'s answer and the
2076    // same claim: it says what the thing above it is, and it is not the thing.
2077    let _ = writeln!(
2078        css,
2079        ".{picture} > .{caption} {{\n    color: var(--content-muted);\n}}"
2080    );
2081
2082    css
2083}
2084
2085/// A region showing one child at a time, and the chrome that moves between them.
2086///
2087/// [`makeover_layout::Showing`] lets a description say that a region holds
2088/// several children and shows some of them. A renderer derives its own chrome
2089/// from that, which is what stops every renderer growing a `match` on a widget
2090/// name; these are the rules the derived chrome needs.
2091///
2092/// # Why the default is every child, and the enhancement takes them away
2093///
2094/// The controls are a lie until something binds them. A prev button rendered
2095/// into a document with no script is a control that looks live and answers
2096/// nothing, and the reader it lies to is exactly the one who cannot see the
2097/// other children either — the collapsing and the moving are the same half.
2098///
2099/// So the rules run in the direction the enhancement does. Nothing here hides a
2100/// child and nothing here shows a control. Whatever binds the region sets
2101/// `data-ready` on it, and that is what collapses the stack to one and reveals
2102/// the row that moves it. A reader with no script gets every child in order and
2103/// no controls, which is more content rather than less, and a reader with
2104/// script gets a settled page rather than a stack that jumps to one frame after
2105/// load.
2106///
2107/// MNW proved this shape by hand — a `<noscript>` stylesheet opening its
2108/// carousel back out — and it is here rather than there because the property is
2109/// the description's, not one app's.
2110///
2111/// # Not spacing
2112///
2113/// The row's gaps are `makeover-geometry`'s question and are absent for
2114/// [`row_rules`]'s reason. What is here is `display`, which carries no
2115/// magnitude, and the muted readout, which is the same claim `picture-caption`
2116/// makes: it says where you are among the children and it is not one of them.
2117fn showing_rules(opts: &Emit) -> String {
2118    let controls = class("showing", opts);
2119    let position = class("showing-position", opts);
2120    let frame = class("showing-frame", opts);
2121    let mut css = String::new();
2122
2123    // Hidden until something binds it, which is the whole argument above.
2124    let _ = writeln!(css, ".{controls} {{\n    display: none;\n}}");
2125    // Block, and nothing about how the three sit in it. A button and a span are
2126    // inline already, so they make a row without this crate saying so, and
2127    // saying so is where `align-items` and a gap would follow -- both spacing,
2128    // both `makeover-geometry`'s, and `row_rules` refuses them for the same
2129    // reason.
2130    let _ = writeln!(
2131        css,
2132        "[data-ready] > .{controls} {{\n    display: block;\n}}"
2133    );
2134
2135    // A child is in flow until the region is bound, and then only the current
2136    // one is. `.current` is a modifier for the reason `.chosen` and `.latched`
2137    // are: one name for the state, set by whoever knows it.
2138    let _ = writeln!(
2139        css,
2140        "[data-ready] > .{frame}:not(.current) {{\n    display: none;\n}}"
2141    );
2142
2143    // Reads back one step. `picture-caption`'s rule and its reason.
2144    let _ = writeln!(css, ".{position} {{\n    color: var(--content-muted);\n}}");
2145
2146    css
2147}
2148
2149/// A time axis: the container, its gridlines and ruler, and the placed things.
2150///
2151/// [`Track`](makeover_layout::Track), the one member here whose whole point is
2152/// *position*. That makes the magnitude line this crate keeps worth restating
2153/// rather than assuming.
2154///
2155/// # Where the numbers come from
2156///
2157/// Every value that varies per item is a custom property the caller sets
2158/// inline, and every rule here reads one. `--track-at` and `--track-for` are
2159/// percentages of the span, which
2160/// [`Track::fraction`](makeover_layout::Track::fraction) computes once so three
2161/// renderers cannot disagree about it. Nothing here knows a pixel.
2162///
2163/// That is what lets the stylesheet stay static while the items move: an entry
2164/// carries `style="--track-at: 37.5%; --track-for: 4.166%"` and the rule below
2165/// turns it into a box. The alternative was emitting a rule per item, which is
2166/// a stylesheet that grows with the data.
2167///
2168/// **The height of the track is the app's**, not this crate's. 96 quarter-hour
2169/// slots at some slot height is a size, and a size is `makeover-geometry`'s
2170/// question -- the same refusal `figure_rules` and `placeholder_rules` make.
2171/// Percentages need a resolved height above them, so `.track` gets
2172/// `position: relative` and nothing else; the app says how tall a day is.
2173///
2174/// # Overlap
2175///
2176/// Two things at the same time is intrinsic to an axis and has no analogue in a
2177/// list. The description does not declare it -- `Placement::overlaps` derives it
2178/// from the times -- so what arrives here is a lane index and a lane count, and
2179/// the rule divides the width. A renderer that would rather stack them ignores
2180/// both properties and the defaults give it one full-width lane.
2181fn track_rules(opts: &Emit) -> String {
2182    let track = class("track", opts);
2183    let slot = class("track-slot", opts);
2184    let tick = class("track-tick", opts);
2185    let entry = class("track-entry", opts);
2186    let mut css = String::new();
2187
2188    // The positioning context every entry resolves against, and the whole of
2189    // what this crate says about the container. No height: see the doc above.
2190    let _ = writeln!(css, ".{track} {{\n    position: relative;\n}}");
2191
2192    // Gridlines and the ruler are the axis reading itself back, which is
2193    // `picture-caption` and `showing-position`'s claim: about the thing rather
2194    // than one of the things.
2195    //
2196    // `border_width` rather than a literal, the way `depth_rule` writes its
2197    // edge. A gridline is the one place a timeline would most naturally reach
2198    // for a hardcoded 1px, and a hardcoded 1px is this crate naming a size.
2199    let _ = writeln!(
2200        css,
2201        ".{slot} {{\n    border-top: {} solid var(--border);\n}}",
2202        opts.border_width
2203    );
2204    let _ = writeln!(css, ".{tick} {{\n    color: var(--content-muted);\n}}");
2205
2206    // The one rule that does real work. Top and height are the placement;
2207    // left and width are the lane. Both lane properties default so an entry
2208    // that names neither is full width, which is the common case and the one a
2209    // renderer gets for free.
2210    let _ = writeln!(
2211        css,
2212        ".{entry} {{\n    \
2213         position: absolute;\n    \
2214         top: var(--track-at, 0%);\n    \
2215         height: var(--track-for, 100%);\n    \
2216         left: calc(var(--track-lane, 0) / var(--track-lanes, 1) * 100%);\n    \
2217         width: calc(100% / var(--track-lanes, 1));\n\
2218         }}"
2219    );
2220
2221    css
2222}
2223
2224/// A region's stand-in, and the header of a table that can be reordered.
2225///
2226/// Both are colour and affordance only, the same place `figure_rules` lands:
2227/// emitting a type scale is not this crate's. How much room
2228/// a stand-in gets is a size — goingson has the same one at three, as
2229/// `--compact`, `--dashboard` and `--padded` — and a size is
2230/// `makeover-geometry`'s question.
2231///
2232/// The caret is the one thing here that is neither colour nor affordance, and it
2233/// is a renderer's own expression rather than a value the description named:
2234/// `aria-sort` is what the table actually says, and this turns it into something
2235/// visible for everyone not using a screen reader. A terminal draws its own; an
2236/// immediate-mode painter draws its own.
2237fn state_rules(opts: &Emit) -> String {
2238    let placeholder = class("placeholder", opts);
2239    let text = class("placeholder-text", opts);
2240    let heading = class("table-heading", opts);
2241    let mut css = String::new();
2242
2243    let _ = writeln!(
2244        css,
2245        ".{placeholder} > .{text} {{\n    color: var(--content-muted);\n}}"
2246    );
2247    // Only the failure is toned. An empty list is the normal state of a new
2248    // install, and `Readiness::tone` is what says so.
2249    let _ = writeln!(
2250        css,
2251        ".{placeholder}[data-tone=\"{0}\"] > .{text} {{\n    color: var(--{0});\n}}",
2252        Tone::Danger.token()
2253    );
2254    // The way out stands a group apart from the sentence it answers. A gap
2255    // step rather than a size: how much room the stand-in gets is still the
2256    // app's.
2257    let _ = writeln!(
2258        css,
2259        ".{placeholder} > .{} {{\n    margin-block-start: var(--gap-group);\n}}",
2260        class("placeholder-action", opts)
2261    );
2262
2263    // A header that reorders the table is a control, and the pointer is the
2264    // only part of saying so that is not the app's own type and spacing.
2265    let _ = writeln!(
2266        css,
2267        ".{heading}[data-sortable] {{\n    cursor: pointer;\n}}"
2268    );
2269    // The caret carries its own leading space, the way `makeover-tui` and
2270    // `makeover-immediate` both write `" \u{25B2}"`. It used to be emitted bare,
2271    // and both apps that adopted the vocabulary had to put the gap back in their
2272    // own stylesheets on the same afternoon -- each having to work out first that
2273    // app CSS outranks this crate's cascade layer, so adding the space the
2274    // obvious way, as `content`, silently wins over the glyph and leaves the
2275    // heading with no caret at all. A consumer should not have to know that, and
2276    // with the space emitted here there is nothing left for one to add.
2277    //
2278    // Three states, three tones, on the convention in wiki
2279    // `three-tone-convention`. A column in force is `content`; a column offering
2280    // to reorder and not doing it now is `content-secondary`, because it still
2281    // answers a press; a column that is not sortable emits no caret at all and
2282    // takes nothing. The idle arm used to hide its glyph and reserve the box,
2283    // which cost a reflow-free press and said nothing. It draws now, and the
2284    // reservation stops being a thing to get right.
2285    //
2286    // The glyph is `Sort::glyph`, escaped rather than written: a CSS `content`
2287    // string cannot carry the character literally through this file's own
2288    // escaping, and spelling it here as well would put the third copy back that
2289    // `makeover-layout` 0.27.5 exists to remove.
2290    //
2291    // On the heading's `.cell-in`, so the caret sits beside the label inside
2292    // its padding rather than at the far edge of a column that grew.
2293    let cell_in = class("cell-in", opts);
2294    let _ = writeln!(
2295        css,
2296        ".{heading}[data-sortable] > .{cell_in}::after \
2297         {{\n    content: \" {}\";\n    color: var(--content-secondary);\n}}",
2298        css_escape(Sort::Ascending.glyph())
2299    );
2300    for direction in [Sort::Ascending, Sort::Descending] {
2301        let _ = writeln!(
2302            css,
2303            ".{heading}[aria-sort=\"{}\"] > .{cell_in}::after \
2304             {{\n    content: \" {}\";\n    color: var(--content);\n}}",
2305            direction.as_str(),
2306            css_escape(direction.glyph())
2307        );
2308    }
2309    css
2310}
2311
2312/// The component layer: every named thing phase A emits.
2313///
2314/// No scrollbar track. It was on the phase A list and came off: eight lines of
2315/// `::-webkit-scrollbar` with no shape a terminal or an immediate-mode painter
2316/// would want handed to it, so it stays with the apps.
2317#[must_use]
2318pub fn component_rules(opts: &Emit) -> String {
2319    let mut css = String::new();
2320    css.push_str(&surface_rules(opts));
2321    css.push_str(&link_rules(opts));
2322    css.push_str(&token_rules(opts));
2323    css.push_str(&selector_rules(opts));
2324    css.push_str(&row_rules(opts));
2325    css.push_str(&run_rules(opts));
2326    css.push_str(&progress_rules(opts));
2327    css.push_str(&chart::rules(opts));
2328    css.push_str(&awaiting_rules(opts));
2329    css.push_str(&figure_rules(opts));
2330    css.push_str(&facts_rules(opts));
2331    css.push_str(&picture_rules(opts));
2332    css.push_str(&showing_rules(opts));
2333    css.push_str(&track_rules(opts));
2334    css.push_str(&state_rules(opts));
2335    css.push_str(&table_rules(opts));
2336    css.push_str(&facet::facet_rules(opts));
2337    css.push_str(&form::group_rules(opts));
2338    css.push_str(&form::editor_rules(opts));
2339    css.push_str(&form::suggestion_rules(opts));
2340    css.push_str(&form::unit_rules(opts));
2341    css.push_str(&form::option_detail_rules(opts));
2342    css.push_str(&form::note_rules(opts));
2343    css.push_str(&leaving_rules());
2344    css
2345}
2346
2347/// How a transient notice goes away.
2348///
2349/// `makeover-timing` says that `Intent::Dismiss` is how long a notice lives
2350/// *before it starts to leave*, and that the leaving itself is `Motion::Fade`.
2351/// `makeover-build` writes `--motion-fade` into every consumer's `timing.css`,
2352/// and this is the rule that reads it. Removing the node the moment the dismiss
2353/// is up skips the leaving entirely.
2354///
2355/// # Why an attribute and not a class
2356///
2357/// [`Emit`]'s prefix moves every class this crate writes, so a class here would
2358/// have to be resolved through `class()` by whoever sets it -- and the party
2359/// setting it is a script, which has no prefix to hand. `data-leaving` is
2360/// outside that namespace, so a renderer can set it from JavaScript with no
2361/// coordination.
2362///
2363/// The transition sits on the notice and the opacity on the leaving state, so
2364/// the element is transitionable before the attribute arrives; a transition
2365/// declared in the same rule as the value it changes has nothing to animate
2366/// from.
2367///
2368/// # Reduced motion is handled by the token, not by a second rule here
2369///
2370/// `timing.css` already zeroes `--motion-fade` under `prefers-reduced-motion`.
2371/// The reader who asked for less motion gets an instant change rather than a
2372/// fade, and the renderer that sets the attribute must still remove the node on
2373/// a timer rather than on `transitionend` -- a zero-length transition may fire
2374/// no event at all, and a node waiting on one that never comes stays forever.
2375///
2376/// The fallback is `0ms` and not a guessed duration: a page with no timing
2377/// sheet has not opted into this vocabulary, and the honest answer there is the
2378/// behaviour it had before, which is the notice going away at once.
2379fn leaving_rules() -> String {
2380    let mut css = String::new();
2381    let _ = writeln!(
2382        css,
2383        "[data-notice] {{\n    transition: opacity var(--motion-fade, 0ms) \
2384         ease-out;\n}}"
2385    );
2386    let _ = writeln!(css, "[data-notice][data-leaving] {{\n    opacity: 0;\n}}");
2387    css
2388}
2389
2390/// The whole phase-A stylesheet: properties, depth rules and components, in
2391/// [`CSS_LAYER`], under a generated-file banner.
2392///
2393/// The banner sits outside the layer, because a comment participates in no
2394/// cascade and a reader opening the file should see what it is before seeing
2395/// an at-rule.
2396#[must_use]
2397pub fn stylesheet(opts: &Emit) -> String {
2398    let body = in_css_layer(&format!(
2399        ":root {{\n{}}}\n\n{}\n{}",
2400        bevel_properties(opts),
2401        depth_rules(opts),
2402        component_rules(opts)
2403    ));
2404    // Counted from the body rather than through `vocabulary::names`, which
2405    // calls back into here.
2406    let classes = vocabulary::classes_in_css(&body).len();
2407    let version = VERSION;
2408    format!(
2409        "/* Generated by makeover-webview {version} from makeover-layout, \
2410         {classes} classes.\n   \
2411         Do not edit. The version and the count are here because a stale\n   \
2412         lockfile fails silently: an older emitter writes a well-formed sheet\n   \
2413         with components missing, and nothing else in the file says so. If\n   \
2414         this version trails what the manifest asks for, re-resolve.\n\n   \
2415         Depth is a fill and an edge together; naming them apart is what let\n   \
2416         them disagree. See the crate's README and wiki note makeover-layout.\n\n   \
2417         Everything below is in the `{CSS_LAYER}` cascade layer. Declare the\n   \
2418         order once in your own stylesheet, or this layer's position is decided\n   \
2419         by whichever generated file the browser happens to see first:\n\n   \
2420         @layer {CSS_LAYER}, base, components, responsive; */\n{body}"
2421    )
2422}
2423
2424#[cfg(test)]
2425mod tests;