Skip to main content

max / makeover-webview

60.6 KB · 1400 lines History Blame Raw
1 //! Phase B, the forms half: [`makeover_layout::Field`] rendered to HTML.
2 //!
3 //! # Why this emits strings
4 //!
5 //! Both webview apps build their markup as strings and hand it to `innerHTML`:
6 //! goingson's `renderFormField` returns a template literal that fifteen call
7 //! sites interpolate into larger literals, and Balanced Breakfast's builds
8 //! nodes but appends them into the same string-built forms. Returning nodes
9 //! would rewrite the surrounding templates as well, which makes it a migration
10 //! rather than an adoption. So: strings, and the escaping comes with them.
11 //!
12 //! # Why one escaper is enough here
13 //!
14 //! goingson carries four escapers and 543 call sites that must pick between
15 //! them, because `escapeHtml` is built on `textContent` serialization and
16 //! **`textContent` refuses to encode `"`**. That is what makes it unsound in an
17 //! attribute, and it is the whole reason the choice exists. Its `escape.js`
18 //! records the finding as the CHRONIC-XSS seal, and its test suite has a gate
19 //! keeping the unsafe one off the namespace.
20 //!
21 //! [`escape`] here is not built on that, so it encodes the quote along with
22 //! everything else, which makes one function sound in both sinks. The four-way
23 //! choice does not move into Rust: it disappears. Nothing in this module hands
24 //! an unescaped value to the output except through [`Markup`], which a caller
25 //! has to name.
26 //!
27 //! # What the description does not carry
28 //!
29 //! One thing: the **current value**, which arrives in [`Filling`]. The
30 //! placeholder and a select's options are not renderer state: the first is
31 //! user-facing text sitting with `label` and `hint`, and the second is needed by
32 //! every renderer, so both are read off [`Field`].
33 //!
34 //! The value stays, and it is not a leftover. A webview reads it back out of
35 //! the DOM, an immediate-mode renderer writes through a `&mut`, and a terminal
36 //! keeps an edit buffer; a description carrying it would have to carry a way to
37 //! write it back, at which point it is a form model.
38
39 use crate::{Emit, class, push_class};
40 use makeover_layout::{Choice, Depth, Field, FieldKind, Intent as _, Selector, ThemeVariant, Tone};
41 use std::fmt::Write as _;
42
43 /// Every class this module can put in markup.
44 ///
45 /// [`crate::facet::FACET_CLASSES`]' obligation, and the module where it was
46 /// missing longest. Most of these carry no rule and never will: `.form-group`,
47 /// `.form-label`, `.form-hint` and `.form-error` are the apps' own names, kept
48 /// so adoption deletes goingson's `renderFormField` rather than restyling
49 /// anything, and phase A emits only what it can generate from the description.
50 /// A class with no rule is invisible to [`crate::vocabulary::vocabulary`],
51 /// which reads the generated sheet, so the unruled half of a renderer's
52 /// vocabulary can only be written down.
53 ///
54 /// What goes wrong without it: an app checking its stylesheet against
55 /// [`crate::vocabulary::names`] concludes that its live `.form-group` and
56 /// `.form-label` rules match nothing and are safe to delete.
57 pub const FIELD_CLASSES: &[&str] = &[
58 "field",
59 "form-checkbox-label",
60 "form-editor-modes",
61 "form-editor-preview",
62 "form-error",
63 "form-group",
64 "form-hint",
65 "form-interval",
66 "form-label",
67 "form-note",
68 "form-option-detail",
69 "form-option-reason",
70 "form-radio-group",
71 "form-radio-label",
72 "form-unit",
73 ];
74
75 // `form-suggestions`, `form-suggestion` and `form-suggestion-detail` are
76 // deliberately absent: [`suggestion_rules`] writes their look and
77 // `quasi-webview` writes their markup, because a suggestion source is a route
78 // and no description layer carries one. They reach the vocabulary through the
79 // generated sheet, which is where a name this crate rules but does not emit
80 // belongs.
81
82 /// The state classes a field carries, which take no prefix.
83 ///
84 /// `chosen` and `latched`'s convention, stated in
85 /// [`crate::vocabulary::vocabulary`]: a state qualifies a prefixed component
86 /// (`.mk-form-group.has-error`) rather than standing on its own, so a prefix
87 /// moves the thing and not its state.
88 ///
89 /// `has-error` marks the group and `visible` marks the message, which is
90 /// [`makeover_layout::Field::invalid`]'s own reasoning: a renderer with no
91 /// descendant selectors cannot find the group from the message, so both are
92 /// told.
93 pub const FIELD_STATE_CLASSES: &[&str] = &["has-error", "visible"];
94
95 /// A string that is already markup, and is emitted without escaping.
96 ///
97 /// The one hole in the escaping, and it has to be named to be used. goingson
98 /// has two live callers that need it, both passing a recurrence-config block
99 /// built elsewhere, and both would otherwise have their markup rendered as
100 /// visible angle brackets. A caller constructing this is stating that the
101 /// contents are trusted; nothing here can check that for them.
102 #[derive(Debug, Clone, Copy, PartialEq, Eq)]
103 pub struct Markup<'a>(pub &'a str);
104
105 /// What the field currently holds.
106 ///
107 /// An enum rather than a bag of optional fields, on the same reasoning
108 /// [`makeover_layout::Depth`] is one: a checkbox holding a string is unsayable
109 /// here, where a struct would let it be said and then have to cope.
110 #[derive(Debug, Clone, Copy, PartialEq, Eq, Default)]
111 pub enum Value<'a> {
112 /// Nothing yet.
113 #[default]
114 Absent,
115 /// The value of anything that takes typed text, a select included: what a
116 /// select holds is the `value` of one of [`Field::options`]'s
117 /// [`Choice`]s.
118 ///
119 /// The options are the field's and never this type's, which is what keeps
120 /// a `Chosen { options, value }` variant from existing.
121 /// `makeover-immediate` carries the same single-variant shape.
122 Text(&'a str),
123 /// A checkbox, on or off.
124 On(bool),
125 /// Both ends of a [`FieldKind::Interval`], lower first.
126 ///
127 /// Two values rather than one string with a separator, for
128 /// [`makeover_layout::Field::upper_name`]'s reason one level down: an
129 /// interval submits under two names, so it comes back as two values, and a
130 /// delimiter this crate owned could appear inside either of them.
131 ///
132 /// Either end may be empty while the other stands. "Over 120 BPM" is a
133 /// lower end and no upper one, and it is an answer rather than a
134 /// half-filled form.
135 Between {
136 /// What the lower box holds now.
137 lower: &'a str,
138 /// What the upper box holds now.
139 upper: &'a str,
140 },
141 }
142
143 impl<'a> Value<'a> {
144 /// The value as text, for the kinds that submit one.
145 const fn as_text(&self) -> &'a str {
146 match self {
147 Self::Text(text) | Self::Between { lower: text, .. } => text,
148 Self::Absent | Self::On(_) => "",
149 }
150 }
151 }
152
153 impl<'a> Value<'a> {
154 /// The upper end, for the one variant that has one.
155 const fn upper_text(&self) -> &'a str {
156 match self {
157 Self::Between { upper, .. } => upper,
158 Self::Absent | Self::Text(_) | Self::On(_) => "",
159 }
160 }
161 }
162
163 /// Everything about the field that the description does not carry.
164 #[derive(Debug, Clone, Copy, Default)]
165 pub struct Filling<'a> {
166 /// What the field holds now.
167 pub value: Value<'a>,
168 /// Markup appended inside the group, after the hint. Not escaped.
169 pub trailing: Option<Markup<'a>>,
170 /// Attributes written onto the control element itself. Not escaped.
171 ///
172 /// [`trailing`](Self::trailing)'s argument at attribute scale: a host knows
173 /// facts about the control that no description layer carries, and until
174 /// this existed the only way to attach one was to stop calling this emitter
175 /// and write a second one. quasi's suggestion source is the first caller —
176 /// a field that owns a list of candidates is a `role="combobox"` pointing
177 /// at the list it owns, and neither half is anything
178 /// [`makeover_layout::Field`] can say.
179 ///
180 /// Written verbatim, so a caller supplies `attr="value"` pairs with no
181 /// leading space and does its own escaping. It is [`Markup`]'s hole in the
182 /// same wall, named the same way so a caller has to state that the contents
183 /// are trusted.
184 ///
185 /// A [`FieldKind::Radio`] drops them, and that is deliberate rather than an
186 /// oversight: a radio group is a set of sibling inputs with no one control
187 /// element, so there is nowhere honest to put an attribute meant for the
188 /// control. The group carries the descriptions for the same reason.
189 pub control_attrs: Option<Markup<'a>>,
190 /// Scopes the `id` attributes to one instance of the form.
191 ///
192 /// The field's `name` is what the value submits under and is the same
193 /// wherever the form appears; its `id` has to be unique in the document,
194 /// and those two facts stop agreeing the moment a form appears twice.
195 /// goingson hits this directly: its new-task and edit-task modals are the
196 /// same field set, so it prefixes `form-modal-task-new` or `-edit` to keep
197 /// `label for` and `aria-describedby` pointing at the right control.
198 ///
199 /// Applies to `id`, `for` and the `-hint` / `-error` associations. Never to
200 /// `name`, which would change what the form submits.
201 pub id_prefix: Option<&'a str>,
202 }
203
204 impl<'a> Filling<'a> {
205 /// A filling that carries a value and nothing else.
206 #[must_use]
207 pub const fn of(value: Value<'a>) -> Self {
208 Self {
209 value,
210 trailing: None,
211 control_attrs: None,
212 id_prefix: None,
213 }
214 }
215
216 /// The document-unique id for a field of this name.
217 fn id_for(&self, name: &str) -> String {
218 let mut id = String::new();
219 if let Some(prefix) = self.id_prefix {
220 escape_into(prefix, &mut id);
221 id.push('-');
222 }
223 escape_into(name, &mut id);
224 id
225 }
226 }
227
228 /// Encode the five characters that let a value stop being a value, into a
229 /// buffer the caller already has.
230 ///
231 /// The form the emitters use. [`escape`] is this with a `String` allocated
232 /// around it, and the allocation is the whole difference: a described screen
233 /// escapes once per attribute and once per run of text, so a function that
234 /// returns a `String` allocates a few thousand times to produce one page,
235 /// where a template engine writes its escaped bytes straight into the output
236 /// buffer.
237 ///
238 /// Sound in element text and in a double-quoted attribute alike, which is the
239 /// property `textContent`-based escaping cannot have. Both sinks are covered by
240 /// one function so that no call site has to choose, here or downstream.
241 ///
242 /// Copies in runs rather than per character. All five encoded characters are
243 /// ASCII, so a byte scan cannot land inside a multi-byte character and the
244 /// slice between two of them is always a valid `&str`. Text with nothing to
245 /// encode — which is most text — is one `push_str` of the whole thing.
246 pub fn escape_into(text: &str, out: &mut String) {
247 let mut start = 0;
248 for (index, byte) in text.bytes().enumerate() {
249 let encoded = match byte {
250 b'&' => "&amp;",
251 b'<' => "&lt;",
252 b'>' => "&gt;",
253 b'"' => "&quot;",
254 b'\'' => "&#39;",
255 _ => continue,
256 };
257 out.push_str(&text[start..index]);
258 out.push_str(encoded);
259 start = index + 1;
260 }
261 out.push_str(&text[start..]);
262 }
263
264 /// Encode the five characters that let a value stop being a value.
265 ///
266 /// [`escape_into`] with a buffer of its own, for the callers that want a value
267 /// rather than an append: a caller assembling an attribute out of several
268 /// pieces, and everything outside this crate that took this function before the
269 /// buffer-writing form existed. Emitting into a buffer you already hold is the
270 /// cheaper path and the one this crate's own emitters take.
271 #[must_use]
272 pub fn escape(text: &str) -> String {
273 let mut out = String::with_capacity(text.len());
274 escape_into(text, &mut out);
275 out
276 }
277
278 /// The `type` an input takes for a kind.
279 ///
280 /// [`FieldKind::Secret`] is `password`, which both apps already map by hand.
281 const fn input_type(kind: FieldKind) -> &'static str {
282 match kind {
283 FieldKind::Secret => "password",
284 FieldKind::Number => "number",
285 FieldKind::Checkbox => "checkbox",
286 FieldKind::File => "file",
287 FieldKind::Hidden => "hidden",
288 // Not decoration. Each of these changes the keyboard a touch device
289 // offers and turns on the platform's own validation, which is why the
290 // description names them apart from text rather than letting the app
291 // pass an HTML type through.
292 FieldKind::Email => "email",
293 FieldKind::Url => "url",
294 FieldKind::Tel => "tel",
295 // The same argument, and it buys more here than anywhere else in this
296 // list: a native picker as well as the keyboard and the validation.
297 // Both submit the format `makeover-layout` names, `DATE_FORMAT` and
298 // `DATETIME_FORMAT`, so honouring it costs this renderer nothing.
299 FieldKind::Date => "date",
300 FieldKind::DateTime => "datetime-local",
301 FieldKind::Radio => "radio",
302 // The clearest case in this list that a kind is not decoration: a
303 // number and a range submit the same value and are different controls,
304 // and the browser is the one drawing the difference.
305 FieldKind::Range => "range",
306 // Select and Textarea are not inputs at all; they never reach here.
307 // Radio is one, but it is emitted once per option by `radio_html` and
308 // so does not reach here either.
309 FieldKind::Text | FieldKind::Select | FieldKind::Textarea | FieldKind::Rich => "text",
310 // A kind added to the description since this renderer was built. Text
311 // accepts any value the others would, so it degrades rather than
312 // dropping the field.
313 _ => "text",
314 }
315 }
316
317 /// The attributes every visible control carries, error state included.
318 ///
319 /// `aria-invalid` is the whole reason the error state is readable at all: the
320 /// generated stylesheet keys the danger ring on `[aria-invalid="true"]` rather
321 /// than on a class, so a control rendered already-invalid without it is styled
322 /// as if nothing were wrong. goingson's runtime validation path sets the
323 /// attribute and its initial render does not, which is exactly the drift one
324 /// emitter removes.
325 /// `id` and `name` arrive separately because they are not the same fact. The
326 /// name is what submits and is fixed by the description; the id has to be
327 /// unique in the document and so carries [`Filling::id_prefix`] when a form
328 /// appears more than once.
329 /// The `accept` attribute, from the description's accept list.
330 ///
331 /// The list is comma-joined because that is the
332 /// attribute's own format, and each entry writes itself: a family is its
333 /// wildcard media type, a media type is itself, a suffix is itself with its
334 /// leading dot. Nothing is normalised on the way through -- `.tar.gz` is two
335 /// dots and the browser is fine with it.
336 ///
337 /// An empty list emits no attribute at all, which is the browser's own "any
338 /// file" and is what the description means by listing nothing. Emitting
339 /// `accept=""` instead would be a filter that matches nothing on some browsers
340 /// and everything on others.
341 ///
342 /// It is a filter and not a guarantee, on the browser's side as much as here:
343 /// the picker keeps an "All Files" escape and the user may take it. Whoever
344 /// validated still validates.
345 fn push_accept(out: &mut String, field: &Field<'_>) {
346 if field.accept.is_empty() {
347 return;
348 }
349 out.push_str(" accept=\"");
350 for (index, one) in field.accept.iter().enumerate() {
351 if index > 0 {
352 out.push(',');
353 }
354 escape_into(one.as_str(), out);
355 }
356 out.push('"');
357 }
358
359 /// The extent and the granularity, as the browser spells them.
360 ///
361 /// Its own function because an interval writes them onto both of its ends: they
362 /// describe the axis rather than either end of it, which is what
363 /// [`FieldKind::Interval`] says and what the six audiofiles filter axes are.
364 fn push_bounds(out: &mut String, field: &Field<'_>) {
365 if let Some(min) = field.min {
366 out.push_str(" min=\"");
367 escape_into(min, out);
368 out.push('"');
369 }
370 if let Some(max) = field.max {
371 out.push_str(" max=\"");
372 escape_into(max, out);
373 out.push('"');
374 }
375 // The browser's own default is `step="1"`, which turns a 0-to-1 threshold
376 // into a two-position control. That is the granularity the description
377 // means when it says nothing, so this is emitted only when an app has said
378 // otherwise rather than defaulted here.
379 //
380 // A range takes its granularity from its curve as of makeover-layout
381 // 0.32.0, and every other kind keeps `Field::step`. See the crate header on
382 // what this renderer can and cannot do with a curve.
383 let step = if field.kind == FieldKind::Range {
384 field.curve.step()
385 } else {
386 field.step
387 };
388 if let Some(step) = step {
389 out.push_str(" step=\"");
390 escape_into(step, out);
391 out.push('"');
392 }
393 }
394
395 fn push_control_attributes(
396 out: &mut String,
397 field: &Field<'_>,
398 filling: &Filling<'_>,
399 id: &str,
400 name: &str,
401 ) {
402 let _ = write!(out, " id=\"{id}\" name=\"");
403 escape_into(name, out);
404 out.push('"');
405 if field.required {
406 out.push_str(" required");
407 }
408 // makeover-layout 0.11.0's constraints. The description carries the rule and
409 // this emits the browser's idiom for it, which is the model `required` has
410 // been using since before the crate wrote down that it carried none.
411 // Enforcement is still whoever validated's, and arrives back as `error`.
412 if let Some(limit) = field.max_length {
413 let _ = write!(out, " maxlength=\"{limit}\"");
414 }
415 push_bounds(out, field);
416 // The description asks for the wall-clock value to be submitted as the
417 // moment it names, and in a browser that conversion is script's: `<input
418 // type="datetime-local">` submits what the user typed and nothing in HTML
419 // turns it into an instant. So this emits the mark and quasi-webview's
420 // `instant.js` does the converting -- the same division as `data-clock`,
421 // where the markup says what to do and the shipped script is what a browser
422 // knows that a description cannot.
423 //
424 // Only DateTime. A date and a time are each half a moment and cannot name
425 // one on their own, so the flag is ignored there rather than emitting a
426 // mark nothing can honour.
427 if field.as_instant && matches!(field.kind, FieldKind::DateTime) {
428 out.push_str(" data-instant=\"true\"");
429 }
430 if field.invalid() {
431 out.push_str(" aria-invalid=\"true\"");
432 }
433
434 push_described_by(out, field, id);
435
436 // Last, so that a host attaching a fact of its own can see everything this
437 // emitter decided and cannot be overwritten by it. Duplicate attributes are
438 // the caller's to avoid: HTML takes the first of a repeated pair, so an
439 // attribute spelled here as well as there keeps this crate's answer.
440 if let Some(Markup(attrs)) = filling.control_attrs {
441 out.push(' ');
442 out.push_str(attrs);
443 }
444 }
445
446 /// The `aria-describedby` naming whatever of the hint and the error exist.
447 ///
448 /// Both associations, in the order they are useful: the standing help, then
449 /// what is currently wrong. goingson's runtime path points describedby at the
450 /// error alone and drops the hint association it never made in the first place;
451 /// naming both here means the hint survives an error appearing.
452 ///
453 /// Its own function because a radio group carries it on the group rather than
454 /// on a control, and one reading of "what describes this field" is the point.
455 fn push_described_by(out: &mut String, field: &Field<'_>, id: &str) {
456 let unit = unit_of(field).is_some();
457 if field.hint.is_none() && field.error.is_none() && field.note.is_none() && !unit {
458 return;
459 }
460 let mut written = false;
461 out.push_str(" aria-describedby=\"");
462 if field.hint.is_some() {
463 let _ = write!(out, "{id}-hint");
464 written = true;
465 }
466 // The unit before the error and after the hint, which is the order they are
467 // useful in: what the number is measured in is standing context like the
468 // hint, and what is wrong with it now comes last.
469 if unit {
470 if written {
471 out.push(' ');
472 }
473 let _ = write!(out, "{id}-unit");
474 written = true;
475 }
476 // The note after the unit and before the error, matching the order the
477 // three are drawn in and the order they are useful in: what the answer
478 // costs is context, and what is wrong with it now still comes last.
479 if field.note.is_some() {
480 if written {
481 out.push(' ');
482 }
483 let _ = write!(out, "{id}-note");
484 written = true;
485 }
486 if field.error.is_some() {
487 if written {
488 out.push(' ');
489 }
490 let _ = write!(out, "{id}-error");
491 }
492 out.push('"');
493 }
494
495 /// The unit to draw beside this field's value, if there is one to draw.
496 ///
497 /// Two conditions rather than one: the field has to carry a unit and its kind
498 /// has to be one that means anything by it. `FieldKind::measurable` is the
499 /// description answering the second, so this renderer keeps no list of its own
500 /// of which kinds are quantities.
501 fn unit_of<'a>(field: &Field<'a>) -> Option<&'a str> {
502 field.unit.filter(|_| field.kind.measurable())
503 }
504
505 /// Whether the field's control is a set of elements rather than one.
506 ///
507 /// A DOM concern rather than a description one, which is why it is decided here
508 /// and not in `makeover-layout`: `for` and `id` are an HTML association and
509 /// egui has no counterpart to get wrong. A `<label for>` aimed at a radio group
510 /// points at nothing, because no single element carries the group's id, so the
511 /// association has to invert — the label takes an id and the group names itself
512 /// with `aria-labelledby`.
513 const fn is_group_control(kind: FieldKind) -> bool {
514 matches!(kind, FieldKind::Radio | FieldKind::Interval)
515 }
516
517 /// An interval: two number boxes inside one labelled group.
518 ///
519 /// The markup MNW's discover sidebar writes by hand -- a `role="group"` with
520 /// `aria-labelledby` pointing at the question, holding `min_price` and
521 /// `max_price` -- which is HTML saying by hand exactly what
522 /// [`FieldKind::Interval`] now says in the description. So this emits what that
523 /// page already proved is right, rather than inventing a shape.
524 ///
525 /// The group carries the error state and the descriptions, for
526 /// [`push_radio`]'s reason: what is wrong is the answer, and marking one box
527 /// invalid would name the wrong half of a fault that belongs to both ends.
528 ///
529 /// # Both boxes take the same extent
530 ///
531 /// [`Field::min`], [`Field::max`] and [`Field::step`] describe the axis rather
532 /// than either end, so [`push_bounds`] writes them onto both. The crossing rule
533 /// is not emitted, because the description does not carry it and the browser
534 /// has no attribute for it: an upper end below the lower one is a refusal
535 /// whoever validated hands back as [`Field::error`], which lands on the group.
536 ///
537 /// # Which end is which, in words
538 ///
539 /// `aria-label`, because the description states direction structurally -- the
540 /// lower end's name is [`Field::name`] and the upper one's is
541 /// [`Field::upper_name`] -- and never in words. Words for the ends are the
542 /// host's, the same way a slider's readout is, and a page with visible Min and
543 /// Max captions supplies them through [`Filling::trailing`] rather than having
544 /// this crate own two strings of English.
545 fn push_interval(out: &mut String, field: &Field<'_>, filling: &Filling<'_>, opts: &Emit) {
546 let id = filling.id_for(field.name);
547
548 out.push_str("<div class=\"");
549 push_class(out, "form-interval", opts);
550 let _ = write!(out, "\" role=\"group\" aria-labelledby=\"{id}-label\"");
551 if field.invalid() {
552 out.push_str(" aria-invalid=\"true\"");
553 }
554 push_described_by(out, field, &id);
555 out.push('>');
556
557 // An interval with no upper name has one end that can be submitted, which
558 // is what the description said and is drawn honestly rather than repaired:
559 // `Field::interval` is what makes it unsayable, and inventing a name here
560 // would submit a parameter no handler is reading.
561 let ends: [(&str, &str, &str); 2] = [
562 ("lower", field.name, filling.value.as_text()),
563 (
564 "upper",
565 field.upper_name.unwrap_or(""),
566 filling.value.upper_text(),
567 ),
568 ];
569 for (end, name, value) in ends {
570 if name.is_empty() {
571 continue;
572 }
573 out.push_str("<input type=\"number\" class=\"");
574 push_class(out, "field", opts);
575 let _ = write!(out, "\" id=\"{id}-{end}\" name=\"");
576 escape_into(name, out);
577 let _ = write!(out, "\" aria-label=\"{end}\"");
578 if field.required {
579 out.push_str(" required");
580 }
581 push_bounds(out, field);
582 if let Some(text) = field.placeholder {
583 out.push_str(" placeholder=\"");
584 escape_into(text, out);
585 out.push('"');
586 }
587 out.push_str(" value=\"");
588 escape_into(value, out);
589 out.push_str("\">");
590 }
591
592 out.push_str("</div>");
593 }
594
595 /// A radio group: the options as sibling inputs sharing one `name`.
596 ///
597 /// The group carries the error state and the descriptions, and the inputs carry
598 /// what submits. That split is [`Field::invalid`]'s reasoning applied one level
599 /// down: marking a single input invalid would say the wrong thing, since what
600 /// is wrong is the answer to the question and not one of the alternatives.
601 ///
602 /// Ids are numbered rather than built from the option values, which can hold
603 /// anything a `&str` can — spaces and quotes included — and would otherwise
604 /// have to be slugged into something unique by a rule this crate would then own.
605 ///
606 /// `required` lands on every input, which is how HTML says a group is
607 /// compulsory: the constraint is satisfied when any one of them is checked.
608 fn push_radio(out: &mut String, field: &Field<'_>, filling: &Filling<'_>, opts: &Emit) {
609 let id = filling.id_for(field.name);
610 let value = filling.value.as_text();
611 let name = escape(field.name);
612
613 out.push_str("<div class=\"");
614 push_class(out, "form-radio-group", opts);
615 let _ = write!(out, "\" role=\"radiogroup\" aria-labelledby=\"{id}-label\"");
616 if field.invalid() {
617 out.push_str(" aria-invalid=\"true\"");
618 }
619 push_described_by(out, field, &id);
620 out.push('>');
621
622 // A group described with no options emits an empty group, for the reason
623 // `Field::options` gives: an app whose option list has not loaded has
624 // exactly that, and an empty group says so on screen rather than in a log.
625 for (index, opt) in field.options.iter().enumerate() {
626 out.push_str("<label class=\"");
627 push_class(out, "form-radio-label", opts);
628 let _ = write!(
629 out,
630 "\"><input type=\"radio\" id=\"{id}-{index}\" name=\"{name}\" value=\""
631 );
632 escape_into(opt.value, out);
633 out.push('"');
634 if opt.value == value {
635 out.push_str(" checked");
636 }
637 if field.required {
638 out.push_str(" required");
639 }
640 // A radio group has room a `<select>` does not, so the reason gets its
641 // own element beside the label rather than being run into it. The class
642 // is what a stylesheet mutes; the text is there either way, which is
643 // the half that matters — the finding was a greyed control with its
644 // explanation behind a hover.
645 if opt.unavailable.is_some() {
646 out.push_str(" disabled");
647 }
648 out.push_str("><span>");
649 escape_into(opt.label, out);
650 out.push_str("</span>");
651 // What picking it means, on the line under the label. `5e21dcfc`, and
652 // the same treatment the reason gets one line down: a radio group has
653 // room, so the sentence sits in its own element rather than being run
654 // into the label the way a `<select>`'s has to be.
655 //
656 // Before the reason, which is the order the two read in: what this
657 // option *is* comes ahead of why it cannot be picked, and an option
658 // carrying both has said two things rather than one long one.
659 if let Some(detail) = opt.detail {
660 out.push_str("<span class=\"");
661 push_class(out, "form-option-detail", opts);
662 out.push_str("\">");
663 escape_into(detail, out);
664 out.push_str("</span>");
665 }
666 if let Some(reason) = opt.unavailable {
667 out.push_str("<span class=\"");
668 push_class(out, "form-option-reason", opts);
669 out.push_str("\">");
670 escape_into(reason, out);
671 out.push_str("</span>");
672 }
673 out.push_str("</label>");
674 }
675
676 out.push_str("</div>");
677 }
678
679 /// The options of a select: the unanswered instruction, an unmatched current
680 /// value carried as its own, then the options themselves.
681 ///
682 /// A select handed a value no option carries renders with nothing selected, the
683 /// browser falls back to the first option, and the next save writes a value
684 /// nobody chose. goingson hit exactly that with a backup-retention default of
685 /// 10 against a 1/3/7/14/0 list, and grew this stray-option fix locally; it is
686 /// here so the second app gets it without hitting the bug first.
687 fn push_options(out: &mut String, field: &Field<'_>, options: &[Choice<'_>], value: &str) {
688 // The unanswered state, which HTML has no attribute for: `placeholder` is
689 // not a `<select>` attribute, and the idiom is an empty option that cannot
690 // be chosen back. `disabled` is what stops it being re-selected once the
691 // user has answered, and `selected` is what puts it in the closed control
692 // while the value is empty; together they read as an instruction rather
693 // than as an option.
694 //
695 // `required` keeps working through it rather than around it: the option's
696 // value is empty, so a required select with this showing is invalid, which
697 // is the true report on a question nobody has answered.
698 //
699 // Emitted only while the value is empty, so it does not sit in the open
700 // list once the field is answered. A non-empty value no option carries is a
701 // wrong answer rather than an absent one and takes the stray-option path
702 // below.
703 if value.is_empty()
704 && let Some(text) = field.placeholder
705 {
706 out.push_str("<option value=\"\" disabled selected>");
707 escape_into(text, out);
708 out.push_str("</option>");
709 }
710 if !value.is_empty() && !options.iter().any(|opt| opt.value == value) {
711 // The one place an escaped value is worth keeping: it is written twice,
712 // as the option's value and as its text.
713 let escaped = escape(value);
714 let _ = write!(
715 out,
716 "<option value=\"{escaped}\" selected data-unmatched=\"true\">{escaped}</option>"
717 );
718 }
719 for opt in options {
720 out.push_str("<option value=\"");
721 escape_into(opt.value, out);
722 out.push('"');
723 if opt.value == value {
724 out.push_str(" selected");
725 }
726 // `disabled` is what the browser reads, and it says nothing about why.
727 // The reason goes in the option's own text, because a `<select>` gives
728 // its options no room for anything else: no title attribute the
729 // keyboard reaches, no second line, no element inside. So the row reads
730 // "Multi-sample: Drop a second sample onto the keyboard." and is the
731 // one place the precondition can be both attached to its option and
732 // read without a pointer.
733 if opt.unavailable.is_some() {
734 out.push_str(" disabled");
735 }
736 out.push('>');
737 escape_into(opt.label, out);
738 // Both extra strings run into the row's text, for the reason above:
739 // this is the one control with nowhere else to put either of them.
740 // `5e21dcfc` did not invent that rule, it met it.
741 if let Some(detail) = opt.detail {
742 out.push_str(": ");
743 escape_into(detail, out);
744 }
745 if let Some(reason) = opt.unavailable {
746 out.push_str(": ");
747 escape_into(reason, out);
748 }
749 out.push_str("</option>");
750 }
751 }
752
753 /// The themes, as one `<optgroup>` per variant with a contrast mark per row.
754 ///
755 /// # The grouping comes out of the order, not out of a group list
756 ///
757 /// [`makeover_layout::Field::themes`] arrives sorted by variant and then by
758 /// measured contrast, and the run of one variant is the group. So this walks
759 /// the list once and opens a new `<optgroup>` whenever the variant changes,
760 /// which is the whole of the grouping logic and cannot disagree with the order
761 /// the way a separately-carried group list could.
762 ///
763 /// A theme whose variant equals its predecessor's never opens a group, so a
764 /// list that arrived unsorted would emit repeated groups rather than silently
765 /// merging distant rows. That is the honest report on a description that broke
766 /// its own contract, and it is visible on screen rather than in a log.
767 ///
768 /// # The follow row is not in a group
769 ///
770 /// It names no theme and sits in no variant, so it is emitted first and bare.
771 /// Grouping it under a heading would be inventing a fourth variant for one row.
772 ///
773 /// # The badge is text, because a `<select>` has nowhere else to put it
774 ///
775 /// A `<select>`'s options take no elements, no second line and no title the
776 /// keyboard reaches, which is [`push_options`]' finding about
777 /// [`Choice::unavailable`] met a second time. So the tier rides in the option's
778 /// own text, in brackets after the name, and it is
779 /// [`makeover_layout::Contrast::badge`]'s spelling rather than one invented
780 /// here — three renderers picking their own is one picker reading three ways.
781 fn push_theme_options(out: &mut String, field: &Field<'_>, value: &str) {
782 if let Some(follow) = field.follows {
783 out.push_str("<option value=\"");
784 escape_into(follow.value, out);
785 out.push('"');
786 if follow.value == value {
787 out.push_str(" selected");
788 }
789 out.push('>');
790 escape_into(follow.label, out);
791 out.push_str("</option>");
792 }
793
794 // A stored id naming a theme that is no longer installed. `push_options`'
795 // reasoning applies unchanged: a value no row carries is a wrong answer
796 // rather than an absent one, and dropping it would silently show the user
797 // a different theme than the one their config names.
798 let known = field.themes.iter().any(|theme| theme.id == value)
799 || field.follows.is_some_and(|follow| follow.value == value);
800 if !value.is_empty() && !known {
801 let escaped = escape(value);
802 let _ = write!(
803 out,
804 "<option value=\"{escaped}\" selected data-unmatched=\"true\">{escaped}</option>"
805 );
806 }
807
808 let mut open: Option<ThemeVariant> = None;
809 for theme in field.themes {
810 if open != Some(theme.variant) {
811 if open.is_some() {
812 out.push_str("</optgroup>");
813 }
814 out.push_str("<optgroup label=\"");
815 escape_into(theme.variant.heading(), out);
816 out.push_str("\" data-variant=\"");
817 out.push_str(theme.variant.as_str());
818 out.push_str("\">");
819 open = Some(theme.variant);
820 }
821
822 out.push_str("<option value=\"");
823 escape_into(theme.id, out);
824 out.push_str("\" data-contrast=\"");
825 out.push_str(theme.contrast.as_str());
826 out.push('"');
827 if theme.id == value {
828 out.push_str(" selected");
829 }
830 out.push('>');
831 escape_into(theme.name, out);
832 out.push_str(" (");
833 out.push_str(theme.contrast.badge());
834 out.push(')');
835 out.push_str("</option>");
836 }
837 if open.is_some() {
838 out.push_str("</optgroup>");
839 }
840 }
841
842 /// The control itself, without its label, hint or error.
843 fn push_control(out: &mut String, field: &Field<'_>, filling: &Filling<'_>, opts: &Emit) {
844 // Emitted before anything else is computed: a radio group carries its
845 // descriptions on the group rather than on a control, so none of the
846 // attributes below belong to it.
847 if matches!(field.kind, FieldKind::Radio) {
848 push_radio(out, field, filling, opts);
849 return;
850 }
851 // The same split one kind along: an interval is two inputs and one
852 // question, so the group carries the error and the descriptions and the
853 // boxes carry what submits.
854 if matches!(field.kind, FieldKind::Interval) {
855 push_interval(out, field, filling, opts);
856 return;
857 }
858
859 let id = filling.id_for(field.name);
860 let placeholder = |out: &mut String| {
861 if let Some(text) = field.placeholder {
862 out.push_str(" placeholder=\"");
863 escape_into(text, out);
864 out.push('"');
865 }
866 };
867
868 match field.kind {
869 // Both multi-line kinds are a `<textarea>`, and the markdown one says so
870 // in an attribute rather than in a class: what the value *is* is not a
871 // styling hook, and a progressive enhancement looking for editors to
872 // upgrade needs a selector that survives `Emit`'s class prefixing.
873 // Without the mark, a described editor is a plain box and the four
874 // hand-written MNW editors have nothing to convert onto.
875 //
876 // `data-format` and not `data-value`: this names the shape of the
877 // value, and `facet` already spends `data-facet-value` on carrying an
878 // actual one. Two attributes a letter apart meaning opposite things is
879 // how a renderer's own vocabulary starts drifting.
880 kind if kind.multiline() => {
881 let rich = matches!(kind, FieldKind::Rich);
882 if rich {
883 push_editor_open(out, opts);
884 }
885 out.push_str("<textarea class=\"");
886 push_class(out, "field", opts);
887 out.push('"');
888 if rich {
889 out.push_str(" data-format=\"markdown\"");
890 }
891 push_control_attributes(out, field, filling, &id, field.name);
892 placeholder(out);
893 out.push('>');
894 escape_into(filling.value.as_text(), out);
895 out.push_str("</textarea>");
896 if rich {
897 push_editor_close(out, opts);
898 }
899 }
900 FieldKind::Select => {
901 out.push_str("<select class=\"");
902 push_class(out, "field", opts);
903 out.push('"');
904 push_control_attributes(out, field, filling, &id, field.name);
905 out.push('>');
906 // A select described with no options emits an empty select, which
907 // says so on screen rather than in a log. That is the description's
908 // own position on `Field::options`, not a fallback invented here.
909 push_options(out, field, field.options, filling.value.as_text());
910 out.push_str("</select>");
911 }
912 // The one place this renderer emits `<optgroup>`, and it emits it
913 // because the description finally says there is a group. The measured
914 // history is the argument: `optgroup` appears at one live site in the
915 // whole tree, and the two apps that had grouped theme pickers lost the
916 // grouping the moment they were described, because `Choice` is a value
917 // and a label and a group is neither.
918 FieldKind::Theme => {
919 out.push_str("<select class=\"");
920 push_class(out, "field", opts);
921 out.push('"');
922 push_control_attributes(out, field, filling, &id, field.name);
923 out.push('>');
924 push_theme_options(out, field, filling.value.as_text());
925 out.push_str("</select>");
926 }
927 FieldKind::Checkbox => {
928 out.push_str("<label class=\"");
929 push_class(out, "form-checkbox-label", opts);
930 out.push_str("\"><input type=\"checkbox\"");
931 push_control_attributes(out, field, filling, &id, field.name);
932 if matches!(filling.value, Value::On(true)) {
933 out.push_str(" checked");
934 }
935 out.push_str("><span>");
936 escape_into(field.label, out);
937 out.push_str("</span></label>");
938 }
939 // A secret never carries its value into the markup. `FieldKind::secret`
940 // is documented as a value that must not be round-tripped through
941 // anything that might persist it, and the DOM is such a thing: it is
942 // read by every extension on the page and is the first thing a crash
943 // reporter serialises. Neither app pre-fills one today, so this costs
944 // nothing and closes the door before something does.
945 FieldKind::Secret => {
946 out.push_str("<input type=\"password\" class=\"");
947 push_class(out, "field", opts);
948 out.push('"');
949 push_control_attributes(out, field, filling, &id, field.name);
950 placeholder(out);
951 out.push('>');
952 }
953 // A file input carries no value, and this is the browser's rule rather
954 // than a preference: setting one from markup is refused, because a page
955 // that could preselect a path could read a file the user never offered.
956 // Nothing upstream needs to know, which is why the exception is here.
957 FieldKind::File => {
958 out.push_str("<input type=\"file\" class=\"");
959 push_class(out, "field", opts);
960 out.push('"');
961 push_control_attributes(out, field, filling, &id, field.name);
962 push_accept(out, field);
963 if field.multiple {
964 out.push_str(" multiple");
965 }
966 out.push('>');
967 }
968 kind => {
969 let _ = write!(out, "<input type=\"{}\" class=\"", input_type(kind));
970 push_class(out, "field", opts);
971 out.push('"');
972 push_control_attributes(out, field, filling, &id, field.name);
973 placeholder(out);
974 out.push_str(" value=\"");
975 escape_into(filling.value.as_text(), out);
976 out.push_str("\">");
977 }
978 }
979 }
980
981 /// The chrome a markdown field gets and a plain textarea does not: the two
982 /// modes, and the pane a preview lands in.
983 ///
984 /// # Why this is the one field with markup around it
985 ///
986 /// [`FieldKind::Rich`]'s own doc says the mark buys a renderer permission to
987 /// offer a preview or a syntax pass, and that a renderer with neither draws a
988 /// textarea. A renderer taking the permission and emitting the same box as
989 /// [`FieldKind::Textarea`] leaves an app converting onto the member with less
990 /// than it had written by hand: MNW's `partial-item-text-editor.js` has a
991 /// Write/Preview pair and a pane behind it, and describing the field without
992 /// this would delete both. So the pair is here, on `facet`'s argument one
993 /// field down -- the markup it replaces is not markup an app is keeping.
994 ///
995 /// # Nothing here renders markdown, and that is where the sanitising stays
996 ///
997 /// The pane arrives empty and this crate never turns a value into markup.
998 /// Converting markdown is the host's, which is where the sanitiser already is:
999 /// MNW renders through `docengine` over ammonia and holds an allowlist beside
1000 /// it. A converter here would move that guarantee into a crate with no view of
1001 /// the host's content-security posture, and `Rich`'s doc is explicit that a
1002 /// host with its own sanitiser still owns it. What this emits is a hook, and
1003 /// whatever fills it fills it with markup it has already made safe.
1004 ///
1005 /// # The direction the enhancement runs
1006 ///
1007 /// [`crate::stylesheet`]'s rule for a showing region, and for its reason: a
1008 /// control rendered into a document with no script is a control that looks live
1009 /// and answers nothing. Nothing is hidden here and no control is shown until
1010 /// whatever binds the editor sets `data-ready` on the wrapper, so a reader with
1011 /// no script gets the textarea alone and a reader with script gets the modes. A bound editor says which mode it is in with
1012 /// `data-mode`, and [`editor_rules`] reads that.
1013 fn push_editor_open(out: &mut String, opts: &Emit) {
1014 // The mark sits on the wrapper as well as on the control, saying one thing
1015 // about two: this control's value is markdown, and this editor edits
1016 // markdown. The rules gate on the wrapper and they are attribute rules
1017 // rather than class rules for `data-format`'s own reason -- the gate has to
1018 // survive `Emit`'s class prefixing, because the enhancement selects on it
1019 // too.
1020 out.push_str("<div data-format=\"markdown\"><div class=\"");
1021 push_class(out, "form-editor-modes", opts);
1022 out.push_str("\">");
1023 push_mode(out, "write", "Write", true, opts);
1024 push_mode(out, "preview", "Preview", false, opts);
1025 out.push_str("</div>");
1026 }
1027
1028 /// One of the two modes, as a segment of the pair.
1029 ///
1030 /// [`crate::option_class`] for [`Selector::Segmented`] rather than a name of
1031 /// its own: a Write/Preview pair is a segmented control, and spelling it as one
1032 /// gets it the depth, the focus ring and the chosen state every described
1033 /// selector gets, from rules that already exist. The words are written here for
1034 /// the reason `facet`'s exclude button writes its own: a description carrying
1035 /// them would be choosing them for the terminal as well.
1036 fn push_mode(out: &mut String, mode: &str, label: &str, chosen: bool, opts: &Emit) {
1037 out.push_str("<button type=\"button\" class=\"");
1038 push_class(out, crate::option_class(Selector::Segmented), opts);
1039 if chosen {
1040 // The sheet keys the held-in segment on the class and a screen reader
1041 // reads the attribute. Both, because they are two readings of one fact,
1042 // which is the arrangement a facet value already has.
1043 out.push_str(" chosen");
1044 }
1045 let _ = write!(
1046 out,
1047 "\" data-editor-mode=\"{mode}\" aria-pressed=\"{chosen}\">{label}</button>"
1048 );
1049 }
1050
1051 /// The preview pane, and the wrapper closing over both halves.
1052 fn push_editor_close(out: &mut String, opts: &Emit) {
1053 out.push_str("<div class=\"");
1054 push_class(out, "form-editor-preview", opts);
1055 // `data-editor-preview` and not an id: a form appears twice in a document
1056 // often enough that `Filling::id_prefix` exists for it, and a binder holding
1057 // the control can reach this without either of them being unique.
1058 out.push_str("\" data-editor-preview></div></div>");
1059 }
1060
1061 /// The rules the markdown editor's chrome needs.
1062 ///
1063 /// The one place this module writes CSS. The class names [`field_html`] emits
1064 /// are goingson's and are deliberately unruled -- `.form-group`, `.form-label`,
1065 /// `.form-hint` and `.form-error` are the app's own, and phase A emits only what
1066 /// it can generate from the description -- but the two names here have no app
1067 /// counterpart to keep, because the chrome did not exist before the member did.
1068 ///
1069 /// Every rule is gated on `[data-format="markdown"]`, which is what keeps them
1070 /// off a plain textarea, and every rule that hides content is gated on
1071 /// `data-ready` as well, which is what keeps them out of a document with no
1072 /// script.
1073 pub(crate) fn editor_rules(opts: &Emit) -> String {
1074 let mut css = String::new();
1075 let modes = class("form-editor-modes", opts);
1076 let preview = class("form-editor-preview", opts);
1077 let field = class("field", opts);
1078
1079 // Hidden until something binds the editor, which is the whole argument in
1080 // `push_editor_open`.
1081 let _ = writeln!(
1082 css,
1083 "[data-format=\"markdown\"] > .{modes} {{\n display: none;\n}}"
1084 );
1085 // Block, and nothing about how the two segments sit in it. A button is
1086 // inline already, so they make a row without this crate saying so, and
1087 // saying so is where a gap would follow -- a magnitude, and
1088 // `makeover-geometry`'s.
1089 let _ = writeln!(
1090 css,
1091 "[data-format=\"markdown\"][data-ready] > .{modes} {{\n display: block;\n}}"
1092 );
1093
1094 // The pane is empty until the host fills it, so it is out of flow in every
1095 // state but the one where a bound editor is showing it. An empty box under
1096 // the control is chrome claiming a preview nobody rendered.
1097 let _ = writeln!(
1098 css,
1099 "[data-format=\"markdown\"] > .{preview} {{\n display: none;\n}}"
1100 );
1101 let _ = writeln!(
1102 css,
1103 "[data-format=\"markdown\"][data-ready][data-mode=\"preview\"] > .{preview} \
1104 {{\n display: block;\n}}"
1105 );
1106 // One at a time. The source and the preview are the same content read two
1107 // ways, and a field showing both answers its own question twice.
1108 let _ = writeln!(
1109 css,
1110 "[data-format=\"markdown\"][data-ready][data-mode=\"preview\"] > .{field} \
1111 {{\n display: none;\n}}"
1112 );
1113
1114 // The pane stands where the control stood, so it reads as the surface the
1115 // control was: `.field` is a well, and this is the well it stands in for.
1116 // Nothing about size -- how tall a preview is is the app's, the way the
1117 // height of a track is.
1118 let _ = write!(
1119 css,
1120 "[data-format=\"markdown\"] > .{preview} {{\n{}}}\n",
1121 crate::depth_declarations(Depth::Well)
1122 );
1123
1124 css
1125 }
1126
1127 /// The rule a field's unit needs.
1128 ///
1129 /// [`suggestion_rules`]' precedent and its argument: `.form-group`,
1130 /// `.form-label`, `.form-hint` and `.form-error` are the apps' own names and
1131 /// stay unruled here, and this one has no app counterpart to keep because
1132 /// nothing emitted it before `Field::unit` existed.
1133 ///
1134 /// One declaration, and it is the whole look. A unit is a fact about the number
1135 /// beside it rather than a second thing to read, so it takes the muted content
1136 /// intent -- the same reading `.figure-caption` and `.track-tick` take, and for
1137 /// the same reason.
1138 ///
1139 /// Nothing about placement or spacing. Where the span sits relative to the
1140 /// control is the app's layout, exactly as `.form-hint`'s is, and a margin
1141 /// asserted here would be this crate deciding a magnitude that belongs to
1142 /// `makeover-geometry`.
1143 /// The rules a field's note needs.
1144 ///
1145 /// [`unit_rules`]' precedent and its argument: `.form-hint` and `.form-error`
1146 /// are the apps' own names and stay unruled here, and this one has no app
1147 /// counterpart to keep because nothing emitted it before [`Field::note`]
1148 /// existed.
1149 ///
1150 /// Colour only, and the tones are the four a badge carries. The bare class is
1151 /// `content` rather than `content-muted`: a note is a consequence the user is
1152 /// meant to read before answering, so muting it by default would be this crate
1153 /// deciding it does not matter.
1154 pub(crate) fn note_rules(opts: &Emit) -> String {
1155 let note = class("form-note", opts);
1156 let mut css = String::new();
1157 let _ = writeln!(css, ".{note} {{\n color: var(--content);\n}}");
1158 for tone in [Tone::Info, Tone::Success, Tone::Warning, Tone::Danger] {
1159 let _ = writeln!(
1160 css,
1161 ".{note}[data-tone=\"{0}\"] {{\n color: var(--{0});\n}}",
1162 tone.token()
1163 );
1164 }
1165 css
1166 }
1167
1168 pub(crate) fn unit_rules(opts: &Emit) -> String {
1169 let unit = class("form-unit", opts);
1170 let mut css = String::new();
1171 let _ = writeln!(css, ".{unit} {{\n color: var(--content-muted);\n}}");
1172 css
1173 }
1174
1175 /// The rules an option's second line needs.
1176 ///
1177 /// [`unit_rules`]' argument: rule what has no app counterpart to keep. An
1178 /// unruled second line renders identically to the label it sits under, which is
1179 /// a worse default than the hand-written markup it replaces.
1180 ///
1181 /// Colour only, and muted, which is the same reading `.form-unit` and
1182 /// `.form-suggestion-detail` take: the line orients the label rather than
1183 /// competing with it. Nothing about placement or spacing, for `unit_rules`'
1184 /// reason — a magnitude asserted here belongs to `makeover-geometry`.
1185 pub(crate) fn option_detail_rules(opts: &Emit) -> String {
1186 let detail = class("form-option-detail", opts);
1187 let mut css = String::new();
1188 let _ = writeln!(css, ".{detail} {{\n color: var(--content-muted);\n}}");
1189 css
1190 }
1191
1192 /// The rules a field's suggestion list needs.
1193 ///
1194 /// [`editor_rules`]' precedent and its argument: the class names this module's
1195 /// markup emits are the apps' own and stay unruled, and these three have no app
1196 /// counterpart to keep because the list did not exist before the member did.
1197 /// The markup is `quasi-webview`'s rather than this crate's — a suggestion
1198 /// source is a route, which no description layer carries — and the look is
1199 /// still this crate's, because a renderer inventing how a list of candidates
1200 /// reads is the drift the vocabulary check exists to catch.
1201 ///
1202 /// # In flow, and not floating
1203 ///
1204 /// An absolutely positioned list needs a positioned ancestor, and the only
1205 /// candidate is `.form-group`, which is the app's class and deliberately
1206 /// unruled here. So the list stands under the control and moves what is below
1207 /// it. An app that wants it over the form positions the group itself, which is
1208 /// one declaration and is the app's call about its own layout.
1209 ///
1210 /// `:empty` is what takes it away, so a route that answers with no candidates
1211 /// leaves no box behind. It is a content question rather than a whitespace one
1212 /// only because the emitter writes no whitespace inside the container, which is
1213 /// stated in `quasi-webview`'s own test.
1214 ///
1215 /// # Nothing about size
1216 ///
1217 /// No height, no scroll ceiling, no padding. How tall a list of candidates gets
1218 /// to be before it scrolls is a magnitude, and magnitudes are
1219 /// `makeover-geometry`'s, exactly as the preview pane's height is.
1220 pub(crate) fn suggestion_rules(opts: &Emit) -> String {
1221 let list = class("form-suggestions", opts);
1222 let entry = class("form-suggestion", opts);
1223 let detail = class("form-suggestion-detail", opts);
1224 let mut css = String::new();
1225
1226 let _ = writeln!(css, ".{list}:empty {{\n display: none;\n}}");
1227 // Over what it covers, which is what a list of candidates is even in flow:
1228 // it is answering the box above it and goes away when the answer is taken.
1229 css.push_str(&crate::depth_rule(&list, Depth::Overlay));
1230 // An entry answers a click, so it gets every state one implies.
1231 css.push_str(&crate::interactive_rules(&entry, Depth::Flat, opts));
1232 // The keyboard's highlight and the pointer's are the same surface. They are
1233 // the same fact told two ways, and a list where arrowing and hovering look
1234 // different is a list that has two current entries.
1235 //
1236 // Keyed on `aria-selected` rather than on a class, for the reason
1237 // `aria-invalid` carries the error state: it is what a screen reader hears,
1238 // so a look keyed on it cannot drift from what is announced. A `.current`
1239 // class would also be a name apps already spell for their own reasons --
1240 // the MNW server has one -- and unlayered app CSS beats this layer in
1241 // silence.
1242 let _ = writeln!(
1243 css,
1244 ".{entry}[aria-selected=\"true\"] {{\n background: var(--hover-surface);\n}}"
1245 );
1246 // The second line, muted rather than disabled. `1fcf2e9b` replaced the
1247 // unavailable reason this rule used to draw: a candidate carries no
1248 // `unavailable`, and what sits beside the label now is what tells one row
1249 // from another that reads the same. Disabled would say the row cannot be
1250 // picked, which is the opposite of what the detail is for.
1251 let _ = writeln!(css, ".{detail} {{\n color: var(--content-muted);\n}}");
1252
1253 css
1254 }
1255
1256 /// One field, as the group the app drops into its form.
1257 ///
1258 /// The shape is goingson's, down to the class names, so adoption there deletes
1259 /// `renderFormField` rather than restyling anything. That is also why the class
1260 /// names are not emitted by [`crate::stylesheet`]: `.form-group`, `.form-label`,
1261 /// `.form-hint` and `.form-error` are the apps' own, and phase A deliberately
1262 /// emits only what it can generate from the description. Whether they should
1263 /// move into the description is the next question this raises, not one it
1264 /// answers.
1265 ///
1266 /// A [`FieldKind::Hidden`] field is the input alone: no group, no label, and
1267 /// nothing drawn, which is what [`FieldKind::visible`] means.
1268 ///
1269 /// The error marks the group as well as the control. That is
1270 /// [`Field::invalid`]'s own reasoning: a renderer with no descendant selectors
1271 /// cannot find the group from the message, so the group has to be told.
1272 ///
1273 /// ```
1274 /// use makeover_layout::{Field, FieldKind};
1275 /// use makeover_webview::{Emit, form::{Filling, Value, field_html}};
1276 ///
1277 /// let field = Field::new(FieldKind::Text, "title", "Title");
1278 /// let html = field_html(&field, &Filling::of(Value::Text("Ship it")), &Emit::default());
1279 ///
1280 /// assert!(html.contains(r#"<label class="form-label" for="title">Title</label>"#));
1281 /// assert!(html.contains(r#"value="Ship it""#));
1282 /// ```
1283 #[must_use]
1284 pub fn field_html(field: &Field<'_>, filling: &Filling<'_>, opts: &Emit) -> String {
1285 let mut html = String::new();
1286 field_html_into(field, filling, opts, &mut html);
1287 html
1288 }
1289
1290 /// One field, written into a buffer the caller already has.
1291 ///
1292 /// [`field_html`]'s streaming form, byte-identical to it. A form is a run of
1293 /// these, so a host building one should hold a single buffer and append each
1294 /// field into it rather than take a `String` per field and concatenate.
1295 pub fn field_html_into(field: &Field<'_>, filling: &Filling<'_>, opts: &Emit, out: &mut String) {
1296 let id = filling.id_for(field.name);
1297
1298 if !field.kind.visible() {
1299 // Name only, no id: a hidden field is never pointed at by a label or a
1300 // description, so the one attribute it needs is the one that submits.
1301 out.push_str("<input type=\"hidden\" name=\"");
1302 escape_into(field.name, out);
1303 out.push_str("\" value=\"");
1304 escape_into(filling.value.as_text(), out);
1305 out.push_str("\">");
1306 return;
1307 }
1308
1309 out.push_str("<div class=\"");
1310 push_class(out, "form-group", opts);
1311 if field.invalid() {
1312 out.push_str(" has-error");
1313 }
1314 if field.extended {
1315 // The disclosure that hides these is a property of the form, not of the
1316 // field, so the field is marked and the app opens or closes the group.
1317 out.push_str("\" data-extended=\"true");
1318 }
1319 out.push_str("\">");
1320
1321 // A checkbox labels itself, on the right of the box. Both apps special-case
1322 // this inline today, which is the tell that it belongs in the description;
1323 // `FieldKind::labels_itself` is where it went.
1324 if !field.kind.labels_itself() {
1325 out.push_str("<label class=\"");
1326 push_class(out, "form-label", opts);
1327 // A group control is named *by* its label rather than pointing at it,
1328 // so the two carry opposite halves of the association. See
1329 // `is_group_control`.
1330 if is_group_control(field.kind) {
1331 let _ = write!(out, "\" id=\"{id}-label\">");
1332 } else {
1333 let _ = write!(out, "\" for=\"{id}\">");
1334 }
1335 escape_into(field.label, out);
1336 out.push_str("</label>");
1337 }
1338
1339 push_control(out, field, filling, opts);
1340
1341 // Adjacent text, because HTML has no unit attribute and inventing one would
1342 // be markup nothing reads. Pointed at by `aria-describedby` so it is not
1343 // decoration a screen reader skips: the number and what it is measured in
1344 // are one fact, and reading the first without the second is reading it
1345 // wrong.
1346 if let Some(unit) = unit_of(field) {
1347 out.push_str("<span class=\"");
1348 push_class(out, "form-unit", opts);
1349 let _ = write!(out, "\" id=\"{id}-unit\">");
1350 escape_into(unit, out);
1351 out.push_str("</span>");
1352 }
1353
1354 if let Some(hint) = field.hint {
1355 out.push_str("<div class=\"");
1356 push_class(out, "form-hint", opts);
1357 let _ = write!(out, "\" id=\"{id}-hint\">");
1358 escape_into(hint, out);
1359 out.push_str("</div>");
1360 }
1361 // A consequence of the answer, between the standing help and the failure.
1362 // The tone rides on `data-tone` -- the same attribute every other toned
1363 // thing in this crate takes -- and it also picks the live region: Warning
1364 // and Danger are assertive, which is quasi-webview's own reading at
1365 // `node.rs:1403` and is honoured here rather than restated differently.
1366 if let Some((tone, note)) = field.note {
1367 out.push_str("<div class=\"");
1368 push_class(out, "form-note", opts);
1369 let assertive = matches!(tone, Tone::Warning | Tone::Danger);
1370 let _ = write!(
1371 out,
1372 "\" id=\"{id}-note\" role=\"{}\"",
1373 if assertive { "alert" } else { "status" }
1374 );
1375 // Neutral is the bare class rather than a variant, matching every
1376 // other toned component here: it is the absence of a status.
1377 if tone != Tone::Neutral {
1378 let _ = write!(out, " data-tone=\"{}\"", tone.token());
1379 }
1380 out.push('>');
1381 escape_into(note, out);
1382 out.push_str("</div>");
1383 }
1384 if let Some(Markup(markup)) = filling.trailing {
1385 out.push_str(markup);
1386 }
1387 if let Some(error) = field.error {
1388 out.push_str("<div class=\"");
1389 push_class(out, "form-error", opts);
1390 let _ = write!(out, " visible\" id=\"{id}-error\" role=\"alert\">");
1391 escape_into(error, out);
1392 out.push_str("</div>");
1393 }
1394
1395 out.push_str("</div>");
1396 }
1397
1398 #[cfg(test)]
1399 mod tests;
1400