oxedyne/fe2o3/fe2o3_text/src/doc/mod.rs
17.4 KiB, 51 runs
created by r1870400018:14266, which is this file's identity for as long as the history lasts, whatever it is later renamed to
download · who wrote it · its history
| 1 | //! A neutral document tree: what a piece of prose *is*, free of the syntax it was written in and the |
| 2 | //! form it will be rendered to. |
| 3 | //! |
| 4 | //! A document is a sequence of [`Block`]s, and a line of prose a sequence of [`Inline`]s. The |
| 5 | //! vocabulary is small and closed: a heading, a paragraph, a list, a quotation, a table, a run of |
| 6 | //! text, a link. That is the whole of it. |
| 7 | //! |
| 8 | //! # The tree names nothing at either end |
| 9 | //! |
| 10 | //! It names no input syntax -- there is no `Block::Asterisks`, only [`Inline::Emph`] -- so a second |
| 11 | //! front-end produces the same tree and every consumer of it keeps working. And it names no output |
| 12 | //! format, so a caller walks it and makes of it whatever it likes: HTML, a signed document, a |
| 13 | //! terminal rendering, an index. |
| 14 | //! |
| 15 | //! Both halves are load-bearing. A tree that admitted one syntax's spelling would make every consumer |
| 16 | //! learn that syntax; a tree that admitted one format's constructs -- a raw HTML node, say -- would |
| 17 | //! make every front-end learn that format. The tree is the narrow waist between the two, and it stays |
| 18 | //! narrow by carrying meaning rather than markup. |
| 19 | //! |
| 20 | //! # Attributes are names, not meanings |
| 21 | //! |
| 22 | //! A [`Block::Div`] and an [`Inline::Span`] carry [`Attrs`] -- an id, classes, key-value pairs. The |
| 23 | //! tree carries them and interprets none of them. `{.warning}` says a region is in the class |
| 24 | //! `warning`; it does not say what `warning` looks like. That is the whole of how the tree can carry a |
| 25 | //! named box or a styled span and still name no format: the name travels, and the meaning is supplied |
| 26 | //! where the tree is rendered -- a stylesheet for HTML, a style table for a signed document -- never |
| 27 | //! here. A tree that resolved `warning` to a colour would be an HTML tree, or an SBJ tree, and no |
| 28 | //! longer the narrow waist between them. |
| 29 | //! |
| 30 | //! # Front-ends |
| 31 | //! |
| 32 | //! - [`markdown`] -- reads Markdown, the form most existing prose is written in. |
| 33 | //! - [`djot`] -- reads Djot, which a prose author reaches for to name a box or a style the syntax of |
| 34 | //! Markdown cannot. |
| 35 | //! - [`html`] -- reads HTML, the form a typesetter exports prose to once it has resolved the author's |
| 36 | //! own macros. |
| 37 | //! |
| 38 | //! # Outputs |
| 39 | //! |
| 40 | //! - [`html`] -- writes the tree out as HTML, for a browser to read. |
| 41 | //! |
| 42 | //! # Usage |
| 43 | //! |
| 44 | //! ```ignore |
| 45 | //! use oxedyne_fe2o3_text::doc::markdown; |
| 46 | //! |
| 47 | //! let tree = res!(markdown::parse("# A heading\n\nA paragraph with *emphasis*.\n")); |
| 48 | //! for block in &tree.blocks { |
| 49 | //! // Walk the tree. |
| 50 | //! } |
| 51 | //! ``` |
| 52 | |
| 53 | pub mod djot; |
| 54 | pub mod html; |
| 55 | pub mod markdown; |
| 56 | pub mod policy; |
| 57 | |
| 58 | /// The attributes a [`Block::Div`] or an [`Inline::Span`] carries: an id, classes, and key-value |
| 59 | /// pairs. |
| 60 | /// |
| 61 | /// Opaque, on purpose. The tree holds the names and interprets none of them, which is what lets it |
| 62 | /// carry a named box or a styled span while still naming no output format. See the module's own |
| 63 | /// "Attributes are names, not meanings". `{#intro .warning .boxed k=v}` parses to an id `intro`, the |
| 64 | /// classes `warning` and `boxed`, and the pair `(k, v)`; what any of those *mean* is the business of |
| 65 | /// whatever renders the tree. |
| 66 | #[derive(Clone, Debug, Default, PartialEq)] |
| 67 | pub struct Attrs { |
| 68 | /// The id, where one was given. At most one; a second `{#...}` replaces the first, as Djot's does. |
| 69 | pub id: Option<String>, |
| 70 | /// The classes, in the order written. |
| 71 | pub classes: Vec<String>, |
| 72 | /// Key-value pairs, in the order written. |
| 73 | pub pairs: Vec<(String, String)>, |
| 74 | } |
| 75 | |
| 76 | impl Attrs { |
| 77 | |
| 78 | /// Whether these attributes name nothing at all -- no id, no class, no pair. |
| 79 | /// |
| 80 | /// A span or a div that carries empty attributes is one the syntax marked but named nothing on; |
| 81 | /// a consumer may treat it as the bare content, since there is nothing to render from an empty |
| 82 | /// set. |
| 83 | pub fn is_empty(&self) -> bool { |
| 84 | self.id.is_none() && self.classes.is_empty() && self.pairs.is_empty() |
| 85 | } |
| 86 | } |
| 87 | |
| 88 | /// A document: the blocks it is made of, in the order they were written. |
| 89 | #[derive(Clone, Debug, Default, PartialEq)] |
| 90 | pub struct Doc { |
| 91 | /// The document's blocks, in reading order. |
| 92 | pub blocks: Vec<Block>, |
| 93 | } |
| 94 | |
| 95 | impl Doc { |
| 96 | |
| 97 | /// An empty document. |
| 98 | pub fn new() -> Self { |
| 99 | Self { blocks: Vec::new() } |
| 100 | } |
| 101 | |
| 102 | /// The text of the document's first heading of the given level, if it has one. |
| 103 | /// |
| 104 | /// A convenience for the common case of a document whose title is its opening heading. Returns the |
| 105 | /// heading's text with every inline flattened, so emphasis inside a title does not lose its words. |
| 106 | pub fn first_heading(&self, level: u8) -> Option<String> { |
| 107 | for block in &self.blocks { |
| 108 | if let Block::Heading { level: l, content } = block { |
| 109 | if *l == level { |
| 110 | return Some(text_of(content)); |
| 111 | } |
| 112 | } |
| 113 | } |
| 114 | None |
| 115 | } |
| 116 | |
| 117 | /// The text of the document's most prominent heading: the first at the shallowest level it has. |
| 118 | /// |
| 119 | /// What a caller after the document's own idea of its title wants. Naming a level would ask the |
| 120 | /// wrong question, because which level a piece is headed by says where the prose came from rather |
| 121 | /// than what it says: an author writing Markdown heads a chapter with a level 1, and the same |
| 122 | /// chapter exported from Typst arrives headed by a level 2, since the exporter keeps level 1 for |
| 123 | /// the document it thinks it is making. A caller asking for level 1 finds no title at all in the |
| 124 | /// second, and titles the chapter after its file. |
| 125 | /// |
| 126 | /// Taking the first heading of any level would be wrong the other way, since a piece may carry a |
| 127 | /// lesser heading above its title. The shallowest level present is the prominent one, whatever |
| 128 | /// number it happens to wear, and among equals the first wins. |
| 129 | pub fn top_heading(&self) -> Option<String> { |
| 130 | let mut top: Option<(u8, &Vec<Inline>)> = None; |
| 131 | for block in &self.blocks { |
| 132 | if let Block::Heading { level, content } = block { |
| 133 | match top { |
| 134 | // Strictly shallower, so a later heading of an equal level never displaces an |
| 135 | // earlier one. |
| 136 | Some((best, _)) if *level >= best => {}, |
| 137 | _ => top = Some((*level, content)), |
| 138 | } |
| 139 | } |
| 140 | } |
| 141 | top.map(|(_, content)| text_of(content)) |
| 142 | } |
| 143 | |
| 144 | /// The number of words in the document's prose. |
| 145 | /// |
| 146 | /// Counts what a reader reads: headings, paragraphs, lists, quotations, divisions and the cells of |
| 147 | /// a table. A code block is left out, because a listing is scanned rather than read, and counting |
| 148 | /// one at prose speed puts minutes on a piece that no reader spends. |
| 149 | pub fn word_count(&self) -> usize { |
| 150 | count_blocks(&self.blocks) |
| 151 | } |
| 152 | } |
| 153 | |
| 154 | /// A block-level element: the things a document is a sequence of. |
| 155 | #[derive(Clone, Debug, PartialEq)] |
| 156 | pub enum Block { |
| 157 | /// A heading, of a level from 1 to 6. |
| 158 | Heading { |
| 159 | /// The heading's level, 1 being the most prominent. |
| 160 | level: u8, |
| 161 | /// The heading's inline content. |
| 162 | content: Vec<Inline>, |
| 163 | }, |
| 164 | /// A paragraph of inline content. |
| 165 | Para(Vec<Inline>), |
| 166 | /// An ordered or unordered list. |
| 167 | List { |
| 168 | /// Whether the list is numbered. |
| 169 | ordered: bool, |
| 170 | /// The items, each a sequence of blocks, so an item may hold a paragraph, a nested list, or more. |
| 171 | items: Vec<Vec<Block>>, |
| 172 | }, |
| 173 | /// A run of source code, preserved exactly as written. |
| 174 | Code { |
| 175 | /// The language the fence named, if it named one. |
| 176 | lang: Option<String>, |
| 177 | /// The code itself, its line structure intact. |
| 178 | text: String, |
| 179 | }, |
| 180 | /// A block quotation, itself a sequence of blocks. |
| 181 | Quote(Vec<Block>), |
| 182 | /// A table: a header row where there is one, the rows of the body, and the columns they are laid |
| 183 | /// out in. |
| 184 | /// |
| 185 | /// The table's words reach a summary or an index through its cells, each of which flattens with |
| 186 | /// [`Cell::text_of`] as any other run of inlines does. |
| 187 | Table { |
| 188 | /// The header row, where the table names its columns. A table need not: a grid of figures is |
| 189 | /// a table whether or not anything stands at the head of it. |
| 190 | head: Option<Row>, |
| 191 | /// The rows of the body, in reading order. |
| 192 | rows: Vec<Row>, |
| 193 | /// The columns, one entry to each, so a row's nth cell is aligned by the nth entry. |
| 194 | cols: Vec<Align>, |
| 195 | }, |
| 196 | /// A thematic break: a division between passages. |
| 197 | Rule, |
| 198 | /// A named or attributed division: a box the prose itself asked for. |
| 199 | /// |
| 200 | /// The construct a prose front-end reaches for to name a region -- an aside, a warning, a figure |
| 201 | /// -- without saying, or knowing, what that region looks like. Markdown cannot write one; Djot's |
| 202 | /// `:::` can. The [`Attrs`] name it and the content is a document in its own right, so a division |
| 203 | /// may hold paragraphs, lists, or further divisions. |
| 204 | Div { |
| 205 | /// What names the division: its id, classes and pairs. |
| 206 | attrs: Attrs, |
| 207 | /// The blocks the division holds. |
| 208 | content: Vec<Block>, |
| 209 | }, |
| 210 | } |
| 211 | |
| 212 | /// One row of a [`Block::Table`]: the cells it holds, in reading order. |
| 213 | #[derive(Clone, Debug, Default, PartialEq)] |
| 214 | pub struct Row(pub Vec<Cell>); |
| 215 | |
| 216 | impl Row { |
| 217 | |
| 218 | /// The row's plain text: every cell's words, a space between each. |
| 219 | pub fn text_of(&self) -> String { |
| 220 | let mut s = String::new(); |
| 221 | for (i, cell) in self.0.iter().enumerate() { |
| 222 | if i > 0 { |
| 223 | s.push(' '); |
| 224 | } |
| 225 | s.push_str(&cell.text_of()); |
| 226 | } |
| 227 | s |
| 228 | } |
| 229 | } |
| 230 | |
| 231 | /// One cell of a [`Row`]: the inline content it holds. |
| 232 | /// |
| 233 | /// A cell holds inlines and not blocks. A cell is a phrase -- a name, a figure, a link -- and a tree |
| 234 | /// that admitted a list or a quotation here would promise every consumer a cell it must lay out as a |
| 235 | /// document of its own. That is a promise no front-end this tree has can keep, and a tree should not |
| 236 | /// make one on their behalf. |
| 237 | #[derive(Clone, Debug, Default, PartialEq)] |
| 238 | pub struct Cell(pub Vec<Inline>); |
| 239 | |
| 240 | impl Cell { |
| 241 | |
| 242 | /// The cell's plain text, its every inline flattened. See [`text_of`]. |
| 243 | pub fn text_of(&self) -> String { |
| 244 | text_of(&self.0) |
| 245 | } |
| 246 | } |
| 247 | |
| 248 | /// How a column's cells sit within the width they are given. |
| 249 | /// |
| 250 | /// The sides are named `Start` and `End`, and are never named `Left` and `Right`. The tree does not |
| 251 | /// know left from right, because it does not know which way its text runs: this crate ships |
| 252 | /// [`bidi`](crate::unicode::bidi) precisely because the prose it carries may run right to left, and a |
| 253 | /// column aligned to the start of the line is then on the *right* of the page. `Start` is the side the |
| 254 | /// text begins on, whichever side that is, and the consumer -- which knows the direction it is laying |
| 255 | /// out in, and is the only thing that does -- is where the two meet. |
| 256 | /// |
| 257 | /// A tree that said `Left` would be wrong for half the world's prose, and would be wrong silently: the |
| 258 | /// table would lay out, and lay out backwards. This is worth leaving alone. |
| 259 | #[derive(Clone, Copy, Debug, Default, PartialEq)] |
| 260 | pub enum Align { |
| 261 | /// No alignment given, which is most columns: the consumer's own default stands. |
| 262 | #[default] |
| 263 | None, |
| 264 | /// Aligned to the side the text begins on. |
| 265 | Start, |
| 266 | /// Centred within the column. |
| 267 | Centre, |
| 268 | /// Aligned to the side the text ends on. |
| 269 | End, |
| 270 | } |
| 271 | |
| 272 | /// An inline element: the things a line of prose is a sequence of. |
| 273 | #[derive(Clone, Debug, PartialEq)] |
| 274 | pub enum Inline { |
| 275 | /// A run of literal text. |
| 276 | Text(String), |
| 277 | /// Emphasised content. |
| 278 | Emph { |
| 279 | /// Whether the emphasis is strong (bold) rather than ordinary (italic). |
| 280 | strong: bool, |
| 281 | /// The emphasised content. |
| 282 | content: Vec<Inline>, |
| 283 | }, |
| 284 | /// A link to a destination. |
| 285 | Link { |
| 286 | /// Where the link points, exactly as written: a URL, a path, or any other name. |
| 287 | to: String, |
| 288 | /// The link's own content, which is what a reader sees. |
| 289 | content: Vec<Inline>, |
| 290 | }, |
| 291 | /// An image, by its source and the text that stands for it. |
| 292 | Image { |
| 293 | /// Where the image is, exactly as written. |
| 294 | src: String, |
| 295 | /// The text that stands in for the image. |
| 296 | alt: String, |
| 297 | }, |
| 298 | /// A span of code within a line. |
| 299 | Code(String), |
| 300 | /// A run of inline content the prose named or attributed. |
| 301 | /// |
| 302 | /// The inline counterpart to [`Block::Div`]: `[text]{.highlight}` marks a span the way `:::` marks |
| 303 | /// a division. The [`Attrs`] name it and interpret nothing. |
| 304 | Span { |
| 305 | /// What names the span. |
| 306 | attrs: Attrs, |
| 307 | /// The span's content. |
| 308 | content: Vec<Inline>, |
| 309 | }, |
| 310 | /// A break the author asked for within a paragraph. |
| 311 | /// |
| 312 | /// Only ever a *hard* break. Where an author's editor wrapped a line is not a break the author |
| 313 | /// asked for, so a front-end resolves such a wrap to a space in the surrounding [`Inline::Text`] |
| 314 | /// and never emits this. A consumer may therefore honour this as a break unconditionally, and |
| 315 | /// needs no rule of its own about whitespace. |
| 316 | Break, |
| 317 | } |
| 318 | |
| 319 | /// The plain text of a run of inlines, with every element flattened to its words. |
| 320 | /// |
| 321 | /// Emphasis and links contribute their content, an image its alt text, a code span its code, and a |
| 322 | /// hard break a single space. What is left is what the passage says, with nothing of how it is marked |
| 323 | /// up -- which is what a title, a summary or an index wants. |
| 324 | pub fn text_of(content: &[Inline]) -> String { |
| 325 | let mut s = String::new(); |
| 326 | for item in content { |
| 327 | match item { |
| 328 | Inline::Text(t) => s.push_str(t), |
| 329 | Inline::Emph { content, .. } => s.push_str(&text_of(content)), |
| 330 | Inline::Link { content, .. } => s.push_str(&text_of(content)), |
| 331 | Inline::Image { alt, .. } => s.push_str(alt), |
| 332 | Inline::Code(c) => s.push_str(c), |
| 333 | Inline::Span { content, .. } => s.push_str(&text_of(content)), |
| 334 | Inline::Break => s.push(' '), |
| 335 | } |
| 336 | } |
| 337 | s |
| 338 | } |
| 339 | |
| 340 | /// The number of words in a run of blocks, descending into those that hold blocks of their own. |
| 341 | fn count_blocks(blocks: &[Block]) -> usize { |
| 342 | let mut n = 0; |
| 343 | for block in blocks { |
| 344 | n += match block { |
| 345 | Block::Heading { content, .. } => count_inlines(content), |
| 346 | Block::Para(content) => count_inlines(content), |
| 347 | Block::List { items, .. } => items.iter().map(|item| count_blocks(item)).sum(), |
| 348 | Block::Quote(content) => count_blocks(content), |
| 349 | Block::Div { content, .. } => count_blocks(content), |
| 350 | Block::Table { head, rows, .. } => head.iter().chain(rows) |
| 351 | .map(|row| row.0.iter().map(|cell| count_inlines(&cell.0)).sum::<usize>()) |
| 352 | .sum(), |
| 353 | // A listing is scanned rather than read, and a rule holds no words. See `Doc::word_count`. |
| 354 | Block::Code { .. } | Block::Rule => 0, |
| 355 | }; |
| 356 | } |
| 357 | n |
| 358 | } |
| 359 | |
| 360 | /// The number of words in a run of inlines, flattened. |
| 361 | /// |
| 362 | /// A word is a segment holding at least one alphanumeric character, which is what separates the words |
| 363 | /// from the runs of space and punctuation that [`words`](crate::unicode::segment::words) returns |
| 364 | /// between them. |
| 365 | fn count_inlines(content: &[Inline]) -> usize { |
| 366 | crate::unicode::segment::words(&text_of(content)) |
| 367 | .iter() |
| 368 | .filter(|w| w.chars().any(char::is_alphanumeric)) |
| 369 | .count() |
| 370 | } |
| 371 | |
| 372 | #[cfg(test)] |
| 373 | mod tests { |
| 374 | use super::*; |
| 375 | |
| 376 | use oxedyne_fe2o3_core::prelude::*; |
| 377 | |
| 378 | #[test] |
| 379 | fn test_the_plain_text_of_a_run_flattens_every_inline_00() -> Outcome<()> { |
| 380 | // Every inline contributes its words and none of its markup, so a title reads as it was written. |
| 381 | let content = vec![ |
| 382 | Inline::Text("A ".to_string()), |
| 383 | Inline::Emph { |
| 384 | strong: true, |
| 385 | content: vec![Inline::Text("loud".to_string())], |
| 386 | }, |
| 387 | Inline::Text(" ".to_string()), |
| 388 | Inline::Link { |
| 389 | to: "somewhere".to_string(), |
| 390 | content: vec![Inline::Text("link".to_string())], |
| 391 | }, |
| 392 | ]; |
| 393 | assert_eq!(text_of(&content), "A loud link"); |
| 394 | Ok(()) |
| 395 | } |
| 396 | |
| 397 | /// A table says what its cells say, so a summary or an index that walks the tree finds a table's |
| 398 | /// words where it finds every other block's. |
| 399 | #[test] |
| 400 | fn test_a_table_contributes_the_words_of_its_cells_01() -> Outcome<()> { |
| 401 | let head = Row(vec![ |
| 402 | Cell(vec![Inline::Text("Name".to_string())]), |
| 403 | Cell(vec![Inline::Text("Age".to_string())]), |
| 404 | ]); |
| 405 | let row = Row(vec![ |
| 406 | Cell(vec![ |
| 407 | Inline::Emph { |
| 408 | strong: true, |
| 409 | content: vec![Inline::Text("Alice".to_string())], |
| 410 | }, |
| 411 | ]), |
| 412 | Cell(vec![Inline::Text("30".to_string())]), |
| 413 | ]); |
| 414 | // A cell flattens as any other run of inlines does, and a row is its cells. |
| 415 | assert_eq!(head.0[0].text_of(), "Name"); |
| 416 | assert_eq!(head.text_of(), "Name Age"); |
| 417 | assert_eq!(row.text_of(), "Alice 30"); |
| 418 | let table = Block::Table { |
| 419 | head: Some(head), |
| 420 | rows: vec![row], |
| 421 | cols: vec![Align::Start, Align::End], |
| 422 | }; |
| 423 | match &table { |
| 424 | Block::Table { head, rows, cols } => { |
| 425 | match head { |
| 426 | Some(head) => assert_eq!(head.text_of(), "Name Age"), |
| 427 | None => panic!("the table lost its header row"), |
| 428 | } |
| 429 | assert_eq!(rows[0].text_of(), "Alice 30"); |
| 430 | assert_eq!(cols.len(), 2); |
| 431 | } |
| 432 | other => panic!("expected a table, got {:?}", other), |
| 433 | } |
| 434 | Ok(()) |
| 435 | } |
| 436 | |
| 437 | /// A column nobody aligned is aligned by nothing, which is what a consumer's own default is for. |
| 438 | #[test] |
| 439 | fn test_an_alignment_defaults_to_none_02() -> Outcome<()> { |
| 440 | assert_eq!(Align::default(), Align::None); |
| 441 | Ok(()) |
| 442 | } |
| 443 | |
| 444 | /// The count is of words and not of the punctuation and spaces between them, and it reaches the |
| 445 | /// words a nested block holds. |
| 446 | #[test] |
| 447 | fn test_a_document_counts_the_words_a_reader_reads_03() -> Outcome<()> { |
| 448 | let text = |s: &str| vec![Inline::Text(s.to_string())]; |
| 449 | let doc = Doc { |
| 450 | blocks: vec![ |
| 451 | Block::Heading { level: 1, content: text("A short title") }, // 3 |
| 452 | Block::Para(text("Four words, one comma.")), // 4 |
| 453 | Block::Quote(vec![Block::Para(text("Two words"))]), // 2 |
| 454 | Block::List { |
| 455 | ordered: false, |
| 456 | items: vec![ |
| 457 | vec![Block::Para(text("One"))], // 1 |
| 458 | vec![Block::Para(text("Another two"))], // 2 |
| 459 | ], |
| 460 | }, |
| 461 | Block::Rule, |
| 462 | ], |
| 463 | }; |
| 464 | assert_eq!(doc.word_count(), 12); |
| 465 | Ok(()) |
| 466 | } |
| 467 | |
| 468 | /// A listing is scanned rather than read, so it is not counted: a post whose bulk is code would |
| 469 | /// otherwise be given minutes no reader spends on it. |
| 470 | #[test] |
| 471 | fn test_a_code_block_is_not_counted_04() -> Outcome<()> { |
| 472 | let doc = Doc { |
| 473 | blocks: vec![ |
| 474 | Block::Para(vec![Inline::Text("Three words here".to_string())]), |
| 475 | Block::Code { |
| 476 | lang: Some("rust".to_string()), |
| 477 | text: "let a = 1; let b = 2; let c = 3;".to_string(), |
| 478 | }, |
| 479 | ], |
| 480 | }; |
| 481 | assert_eq!(doc.word_count(), 3); |
| 482 | Ok(()) |
| 483 | } |
| 484 | } |