max / quasi
| 1 | //! The immediate-mode renderer for [`quasi_router`]: a described screen in, an |
| 2 | //! egui frame out. |
| 3 | //! |
| 4 | //! <!-- wiki: quasi-overview --> |
| 5 | //! |
| 6 | //! The third renderer, and the one the stack was missing. `quasi-webview` and |
| 7 | //! `quasi-tui` have consumed a [`Screen`] since the beginning; nothing did in |
| 8 | //! egui, so an egui app describing a screen had nowhere to send it. What looked |
| 9 | //! like the egui renderer was `makeover-immediate`, which draws described |
| 10 | //! *nodes* and takes no dependency on `quasi-router` at all: it is the peer of |
| 11 | //! `makeover-webview` and `makeover-tui`, one layer below this. |
| 12 | //! |
| 13 | //! | Layer | webview | terminal | egui | |
| 14 | //! |---|---|---|---| |
| 15 | //! | nodes, from `makeover-layout` | `makeover-webview` | `makeover-tui` | `makeover-immediate` | |
| 16 | //! | screens, from `quasi-router` | `quasi-webview` | `quasi-tui` | **this crate** | |
| 17 | //! |
| 18 | //! Named for the mode and not the library, the way `makeover-immediate` is: what |
| 19 | //! separates this renderer from the other two is that there is no retained tree |
| 20 | //! and no cascade, and egui is the backend it is written against. |
| 21 | //! |
| 22 | //! # What immediate mode gives this renderer for free |
| 23 | //! |
| 24 | //! `quasi-tui` holds five things a browser provides quietly: what is typed, what |
| 25 | //! has focus, how far a pane is scrolled, what is ticked, and where back goes. |
| 26 | //! egui provides three of them, so this crate is smaller than the terminal's |
| 27 | //! rather than larger: |
| 28 | //! |
| 29 | //! - **Reach and focus are egui's**, entirely. Its own id stack decides what is |
| 30 | //! reachable and its own state decides what holds the keyboard, which is the |
| 31 | //! rule `makeover-immediate`'s header already states for the ring. There is no |
| 32 | //! `focus.rs` here and there should not be one. |
| 33 | //! - **Scroll is egui's**, through `ScrollArea`. |
| 34 | //! - **What is typed and what is ticked are not.** A described field is built |
| 35 | //! from the description every frame, so the buffer behind it has to outlive |
| 36 | //! the frame and belongs to the app. That is [`View`], and it is the same |
| 37 | //! discovery the terminal made for the same reason. |
| 38 | //! |
| 39 | //! # Where a description stops being enough |
| 40 | //! |
| 41 | //! The three findings `quasi-tui`'s `region` module records apply here unchanged, |
| 42 | //! because they are about the description rather than about terminals: a tabbed |
| 43 | //! arrangement does not say which tab is showing, a tab has no label, and |
| 44 | //! nothing says a region's share beyond [`Arrangement::share`]. Nothing new is |
| 45 | //! invented here to paper over them; the same guesses are made and named. |
| 46 | |
| 47 | |
| 48 | |
| 49 | |
| 50 | |
| 51 | |
| 52 | |
| 53 | |
| 54 | |
| 55 | |
| 56 | |
| 57 | |
| 58 | |
| 59 | |
| 60 | pub use ; |
| 61 | pub use ; |
| 62 | pub use ; |
| 63 | pub use View; |
| 64 | |
| 65 | /// How often this renderer looks again at a [`Slot::live`] region. |
| 66 | /// |
| 67 | /// The description says the contents move and never says how often to look; |
| 68 | /// the rate is picked here, once, so every live region this crate draws moves |
| 69 | /// at one speed. The webview's own number, so a screen described once and |
| 70 | /// drawn twice does not go stale at two different rates. |
| 71 | /// |
| 72 | /// Longer than a frame on purpose. egui repaints when something asks it to, |
| 73 | /// and audiofiles' answer to that was to re-read every frame; this is the rate |
| 74 | /// that replaces it. [`Runtime::show`] asks for the next repaint itself, so a |
| 75 | /// host draws a live screen without owning a clock. |
| 76 | /// |
| 77 | /// [`Slot::live`]: quasi_router::Slot::live |
| 78 | pub const CADENCE: Duration = from_secs; |
| 79 | |
| 80 | /// This renderer's class, and why it may ignore |
| 81 | /// [`Destination::Local`](quasi_router::Destination::Local). |
| 82 | /// |
| 83 | /// And this renderer is the one the ruling was reasoned from: egui redraws |
| 84 | /// every frame from memory, so an immediate-mode host has no need to model |
| 85 | /// "happens without a request" as a separate kind of thing. Making it carry |
| 86 | /// the distinction anyway would be work spent satisfying a model rather than a |
| 87 | /// need, with audiofiles paying for it. |
| 88 | /// |
| 89 | /// The one place the mark is read anyway is [`Runtime::send`], which declines a |
| 90 | /// local action instead of treating it as an address to hand the host. |
| 91 | /// |
| 92 | /// [`Runtime::send`]: Runtime |
| 93 | pub const CLASS: Renderer = Client; |
| 94 | |
| 95 | use HashMap; |
| 96 | use Arc; |
| 97 | |
| 98 | use Ui; |
| 99 | use TableStyle; |
| 100 | use WidgetStyle; |
| 101 | use ; |
| 102 | use |
| 103 | Act, Action, Band, Chrome, Frame, Message, Node, Params, Place, Role, Screen, layout, |
| 104 | ; |
| 105 | |
| 106 | /// A banner this renderer raises about itself. |
| 107 | /// |
| 108 | /// A description bug reported to the user rather than swallowed: a fragment |
| 109 | /// naming a region that is not there would otherwise look like a control that |
| 110 | /// does nothing. |
| 111 | pub |
| 112 | Message |
| 113 | kind: Banner, |
| 114 | tone: Danger, |
| 115 | text, |
| 116 | undo: None, |
| 117 | |
| 118 | |
| 119 | |
| 120 | /// What a host draws inside a bespoke region. |
| 121 | /// |
| 122 | /// egui's counterpart to `Webview::fills` and to [`quasi_tui::Fill`], in this |
| 123 | /// host's currency: a browser takes markup, a terminal takes cells, and this |
| 124 | /// takes a closure that draws into the `Ui` where the region is. Decision 4 is |
| 125 | /// the same on all three sides -- the renderer hands the space over and never |
| 126 | /// looks at what went in it. |
| 127 | /// |
| 128 | /// One method and no measure, which is where this differs from the terminal's |
| 129 | /// trait rather than an oversight: egui lays out in the order it is told and |
| 130 | /// nothing here asks a region how tall it is, so a fill that draws is a fill |
| 131 | /// that has said everything the layout needs. |
| 132 | /// |
| 133 | /// Handed the renderer so a fill can paint in the palette the screen around it |
| 134 | /// is painted in. Never handed the [`View`], which holds what the user has done |
| 135 | /// to the *described* screen: a fill's state is the host's, the same way an |
| 136 | /// island's is in a browser. |
| 137 | /// |
| 138 | /// [`quasi_tui::Fill`]: https://docs.rs/quasi-tui |
| 139 | pub type Fill = ; |
| 140 | |
| 141 | /// An egui renderer for a described screen. |
| 142 | /// |
| 143 | /// Holds what a drawing needs and no screen state: the resolved palette, the |
| 144 | /// styles derived from it, and what to put inside a bespoke region. A host |
| 145 | /// makes one and keeps it, the same way it keeps a `Tui`. |
| 146 | |
| 147 | |
| 148 | palette: Palette, |
| 149 | frame: FrameStyle, |
| 150 | field: FieldStyle, |
| 151 | widget: WidgetStyle, |
| 152 | table: TableStyle, |
| 153 | fills: , |
| 154 | reduced_motion: bool, |
| 155 | |
| 156 | |
| 157 | /// The fills by name, since a drawing is not printable. |
| 158 | |
| 159 | |
| 160 | f.debug_struct |
| 161 | .field |
| 162 | .field |
| 163 | .field |
| 164 | .field |
| 165 | .field |
| 166 | .field |
| 167 | .field |
| 168 | .finish |
| 169 | |
| 170 | |
| 171 | |
| 172 | |
| 173 | /// A renderer drawing in this palette, with default styling. |
| 174 | |
| 175 | |
| 176 | Self |
| 177 | palette, |
| 178 | frame: default, |
| 179 | field: default, |
| 180 | widget: default, |
| 181 | table: default, |
| 182 | fills: new, |
| 183 | reduced_motion: false, |
| 184 | |
| 185 | |
| 186 | |
| 187 | /// Draw for a reader who has asked for less motion, chaining. |
| 188 | /// |
| 189 | /// egui has no `prefers-reduced-motion` to read, which is the whole reason |
| 190 | /// this is a setting and not a media query: the preference is the host's to |
| 191 | /// obtain from its own platform, and this is where it lands so that a |
| 192 | /// drawing can honour it. |
| 193 | /// |
| 194 | /// One flag rather than one per animated thing. The activity mark is the |
| 195 | /// only thing here that moves today, and a second flag beside it would be |
| 196 | /// two ways to answer one question. |
| 197 | /// |
| 198 | /// **It stills the mark rather than removing it.** A reader asking for less |
| 199 | /// motion has asked for the movement to stop, not for the information to go |
| 200 | /// away; see `makeover_timing::activity_blink`. |
| 201 | |
| 202 | pub const |
| 203 | self.reduced_motion = reduced; |
| 204 | self |
| 205 | |
| 206 | |
| 207 | /// Whether this renderer is drawing for a reader who asked for less motion. |
| 208 | |
| 209 | pub const |
| 210 | self.reduced_motion |
| 211 | |
| 212 | |
| 213 | /// Fill the bespoke region with this slot id, chaining. |
| 214 | /// |
| 215 | /// Keyed by [`Slot::id`] and not by the bespoke name, for the reason the |
| 216 | /// other two renderers are: a screen of N rows each carrying a fill shares |
| 217 | /// one name and has N ids. |
| 218 | /// |
| 219 | /// Repeated calls for one id replace. A slot with no entry draws the blocks |
| 220 | /// the description put in it and nothing else; an entry naming an id the |
| 221 | /// screen does not have is ignored rather than drawn anywhere. |
| 222 | /// |
| 223 | /// [`Slot::id`]: quasi_router::Slot |
| 224 | |
| 225 | |
| 226 | mut self, |
| 227 | slot_id: impl , |
| 228 | fill: impl Fn + Send + Sync + 'static, |
| 229 | |
| 230 | self.fills.insert; |
| 231 | self |
| 232 | |
| 233 | |
| 234 | /// What this renderer puts inside the bespoke region with this id. |
| 235 | |
| 236 | |
| 237 | self.fills.get |
| 238 | |
| 239 | |
| 240 | /// The colours this renderer draws in. |
| 241 | |
| 242 | pub const |
| 243 | &self.palette |
| 244 | |
| 245 | |
| 246 | /// Use these frame, field, widget and table styles, chaining. |
| 247 | |
| 248 | |
| 249 | mut self, |
| 250 | frame: FrameStyle, |
| 251 | field: FieldStyle, |
| 252 | widget: WidgetStyle, |
| 253 | table: TableStyle, |
| 254 | |
| 255 | self.frame = frame; |
| 256 | self.field = field; |
| 257 | self.widget = widget; |
| 258 | self.table = table; |
| 259 | self |
| 260 | |
| 261 | |
| 262 | /// Draw a whole screen, and answer what the user did to it. |
| 263 | /// |
| 264 | /// The title is not drawn, for the reason `quasi-tui` gives: a window title |
| 265 | /// is the host's to set, the same way a webview host puts it in `<title>` |
| 266 | /// rather than in the document. [`Screen::discovery`] is declined outright, |
| 267 | /// since nothing crawls a desktop window. |
| 268 | /// |
| 269 | /// [`Row::address`] is declined too, and this says so rather than leaving |
| 270 | /// it to be inferred: a member read by one renderer and silently ignored by |
| 271 | /// two is how the row types drifted apart in the first place. An |
| 272 | /// address is where a document holds a row, which is a fact about having a |
| 273 | /// document and an address bar; a window has neither. The identity this |
| 274 | /// host does answer with is [`Row::value`], which `row_at` uses to say |
| 275 | /// which row the pointer is over. |
| 276 | /// |
| 277 | /// [`Row::address`]: quasi_router::Row::address |
| 278 | /// [`Row::value`]: quasi_router::Row::value |
| 279 | /// |
| 280 | /// `None` is the ordinary frame. egui redraws continuously, so most frames |
| 281 | /// are a user doing nothing, and a renderer that answered a request per |
| 282 | /// frame would call the router sixty times a second. |
| 283 | |
| 284 | self.framed |
| 285 | |
| 286 | |
| 287 | /// Draw a screen inside the frame a mount put around it. |
| 288 | /// |
| 289 | /// # Where the frame goes |
| 290 | /// |
| 291 | /// The bottom, and the verbs below the status line, which is the reading |
| 292 | /// order the measured window has: what happened, then what to do next. egui |
| 293 | /// has no stylesheet to defer the question to, so this is the renderer |
| 294 | /// deciding, the same way it decides that notices go at the top. |
| 295 | |
| 296 | &self, |
| 297 | ui: &mut Ui, |
| 298 | screen: &Screen, |
| 299 | frame: &Frame, |
| 300 | view: &mut View, |
| 301 | |
| 302 | self.chromed |
| 303 | |
| 304 | |
| 305 | /// Draw a screen inside its frame, under the panel the app keeps on screen. |
| 306 | /// |
| 307 | /// # Where the panel goes |
| 308 | /// |
| 309 | /// Last, under the frame. The lifetimes stack that way -- the screen is |
| 310 | /// replaced by every navigation, the frame outlives the screen inside it, |
| 311 | /// and the panel outlives both -- and an immediate-mode host lays out in |
| 312 | /// the order it is told, so "under" is what drawing it last means. egui has |
| 313 | /// no stylesheet to defer the question to, which is why this renderer |
| 314 | /// answers it at all. |
| 315 | |
| 316 | &self, |
| 317 | ui: &mut Ui, |
| 318 | screen: &Screen, |
| 319 | frame: &Frame, |
| 320 | chrome: &Chrome, |
| 321 | view: &mut View, |
| 322 | |
| 323 | let hidden = hidden; |
| 324 | let mut pass = Pass |
| 325 | immediate: self, |
| 326 | view, |
| 327 | hidden: &hidden, |
| 328 | fired: None, |
| 329 | stirred: new, |
| 330 | ; |
| 331 | |
| 332 | // The band above everything, notices included. It is where the user is |
| 333 | // and what the app is called, and a window read top to bottom puts that |
| 334 | // first: a header that appeared under a toast would move when the toast |
| 335 | // went. |
| 336 | // |
| 337 | // Only when the app declared one. `Chrome::nav` on its own is not drawn |
| 338 | // by this renderer and was not before the band existed, so an app that |
| 339 | // declares no band gets the window it always got. |
| 340 | // |
| 341 | // `Disclose` is ignored, which is what the vocabulary says a renderer |
| 342 | // with no notion of "not enough room" does. egui has a width and could |
| 343 | // in principle answer it; hiding the places behind a control would be |
| 344 | // this renderer inventing a gesture on a window the reader can resize. |
| 345 | if let Some = &chrome.band |
| 346 | band_ui; |
| 347 | |
| 348 | |
| 349 | // Notices first and at the top, because a notice belongs to the screen |
| 350 | // rather than to a place in it. A webview leaves where they land to the |
| 351 | // stylesheet; egui has none, so this is the renderer deciding, and the |
| 352 | // top is the one place a message about the whole screen can go without |
| 353 | // claiming a region. |
| 354 | // |
| 355 | // Unless the mount says it has a place for one to rest, which is what a |
| 356 | // status line is. `Frame::holds` is the rule and it is the router's, so |
| 357 | // the three renderers cannot each decide it. |
| 358 | for notice in screen.notices.iter.filter |
| 359 | draw; |
| 360 | |
| 361 | |
| 362 | screen_regions; |
| 363 | |
| 364 | // After the screen rather than beside it. An immediate-mode host lays |
| 365 | // out in the order it is told, so "under the screen" is what drawing it |
| 366 | // last means, and there is no rect to reserve the way a terminal has to. |
| 367 | if !frame.bare |
| 368 | for notice in screen.notices.iter.filter |
| 369 | draw; |
| 370 | |
| 371 | for verb in &frame.verbs |
| 372 | draw; |
| 373 | |
| 374 | |
| 375 | |
| 376 | // Activity before status, so the app's condition is the last thing in |
| 377 | // the column. An immediate-mode host lays out in the order it is told, |
| 378 | // so this is the whole of the placement decision. Within a role it is |
| 379 | // declaration order, which is what the description gave. |
| 380 | for role in |
| 381 | for panel in chrome.panels.iter.filter |
| 382 | draw; |
| 383 | |
| 384 | |
| 385 | |
| 386 | pass.fired |
| 387 | |
| 388 | |
| 389 | |
| 390 | /// The header band: the brand and the places on one row, the search under it. |
| 391 | /// |
| 392 | /// One row for the brand and the places because that is what a header is, and a |
| 393 | /// second for the box because a search field given a share of a row is a box |
| 394 | /// too narrow to read what is in it. |
| 395 | /// |
| 396 | /// A [`Place`] is drawn as the control it is. The description says these are |
| 397 | /// addresses with names, and this renderer has no tab widget to draw them in; |
| 398 | /// what it has is the same button every other act is drawn as, which at least |
| 399 | /// cannot disagree with the rest of the window about what a control looks like. |
| 400 | /// |
| 401 | /// # Which place is showing is not marked here, and the reason is a hole |
| 402 | /// |
| 403 | /// [`Screen::place`] says which one it is and the other two renderers draw it: |
| 404 | /// a webview writes `aria-current` and a terminal reverses the tab. There is no |
| 405 | /// third spelling here, because `makeover_layout::State` carries one member and |
| 406 | /// it is `Disabled`. Marking it with a [`Tone`](quasi_router::layout::Tone) |
| 407 | /// would be saying "this one is a success", which is a different claim in the |
| 408 | /// same slot. So it goes unmarked until the vocabulary can say it, rather than |
| 409 | /// being said wrong. |
| 410 | |
| 411 | ui.horizontal |
| 412 | if let Some = &band.brand |
| 413 | // The whole name, mark included. egui draws one run of text with |
| 414 | // one style, and a renderer that dropped the mark would be drawing |
| 415 | // a name the app is not called. |
| 416 | draw |
| 417 | pass, |
| 418 | ui, |
| 419 | &Act, |
| 420 | ; |
| 421 | |
| 422 | for place in places |
| 423 | draw |
| 424 | pass, |
| 425 | ui, |
| 426 | &Act, |
| 427 | ; |
| 428 | |
| 429 | ; |
| 430 | // The sub-places of the one the reader is in, and nobody else's: a window |
| 431 | // holding every sub-place of every tab is a window that is mostly |
| 432 | // navigation. The same rule quasi-tui states for its second row. |
| 433 | if let Some = places |
| 434 | .iter |
| 435 | .find |
| 436 | .filter |
| 437 | |
| 438 | ui.horizontal |
| 439 | for inner in &open.within |
| 440 | draw |
| 441 | pass, |
| 442 | ui, |
| 443 | &Act, |
| 444 | ; |
| 445 | |
| 446 | ; |
| 447 | |
| 448 | if let Some = &band.search |
| 449 | // The same field emitter every question goes through. A search box in |
| 450 | // the header is not a second kind of box, and this crate has no second |
| 451 | // field drawer to make it one. |
| 452 | draw; |
| 453 | |
| 454 | ui.separator; |
| 455 | |
| 456 | |
| 457 | /// What the user set off this frame. |
| 458 | /// |
| 459 | /// An action and everything gathered to send with it. Separate from |
| 460 | /// [`Step`](runtime::Step) because a drawing does not know whether an action |
| 461 | /// needs asking about first: that is [`Act::confirm`](quasi_router::Act), and |
| 462 | /// the runtime is what holds the question while it is answered. |
| 463 | |
| 464 | |
| 465 | /// What to call. |
| 466 | pub action: Action, |
| 467 | /// What to send with it: a form's values, a selection's members. |
| 468 | pub payload: Params, |
| 469 | /// What to ask before doing it, if the description said to ask. |
| 470 | pub confirm: , |
| 471 | |
| 472 | |
| 473 | /// One frame's drawing, and what came out of it. |
| 474 | /// |
| 475 | /// Threaded through the walk rather than returned up it: a press can happen at |
| 476 | /// any depth, and every level returning an `Option` would make each node's |
| 477 | /// drawing responsible for propagating one. |
| 478 | pub |
| 479 | pub(crate) immediate: &'a Immediate, |
| 480 | pub(crate) view: &'a mut View, |
| 481 | /// The regions and questions that do not apply right now. |
| 482 | /// |
| 483 | /// Answered once, before the drawing starts, because the view is held |
| 484 | /// mutably here and a question asked mid-draw could not read it. |
| 485 | pub(crate) hidden: &'a , |
| 486 | pub(crate) fired: , |
| 487 | /// The questions whose value moved during this frame, by name. |
| 488 | /// |
| 489 | /// What a [`Slot::consults`](quasi_router::Slot::consults) needs and an |
| 490 | /// immediate-mode frame has nowhere else to keep: a region asks when a |
| 491 | /// dial *inside it* moves, the dials are drawn before the region's drawing |
| 492 | /// has finished, and a field cannot know which regions contain it. So the |
| 493 | /// fields say what moved and each region reads the set once its body is |
| 494 | /// drawn. |
| 495 | /// |
| 496 | /// One frame's worth. A `Pass` is one frame, which is the whole of why this |
| 497 | /// is here and the deadline it starts is on the view. |
| 498 | pub(crate) stirred: , |
| 499 | |
| 500 | |
| 501 | |
| 502 | /// Record what the user set off. |
| 503 | /// |
| 504 | /// First press wins. Two controls cannot be activated in one frame by a |
| 505 | /// user, so a second one here is a description that put two `Response`s |
| 506 | /// under one click, and taking the first is the same rule |
| 507 | /// `Chrome::bound` uses for a key claimed twice. |
| 508 | pub |
| 509 | if self.fired.is_none |
| 510 | self.fired = Some |
| 511 | action: action.clone, |
| 512 | payload, |
| 513 | confirm: confirm.map, |
| 514 | ; |
| 515 | |
| 516 | |
| 517 | |
| 518 |