oxedyne/fe2o3/fe2o3_file/src/office/odf/text.rs
28.6 KiB, 74 runs
created by r1870400018:22912, 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 | //! `.odt`: an OpenDocument text document, written from and read back into the neutral document tree. |
| 2 | //! |
| 3 | //! # A heading says its own level |
| 4 | //! |
| 5 | //! `<text:h text:outline-level="2">` is a level two heading, full stop. There is no style to resolve |
| 6 | //! and no built-in name to recognise, which is the whole of what made the WordprocessingML reader |
| 7 | //! need `styles.xml` before it could tell a heading from a paragraph. This is the easier direction and |
| 8 | //! it is worth saying why: OpenDocument put the meaning in the element and Microsoft put it in a |
| 9 | //! style, and every consequence follows from that one choice. |
| 10 | //! |
| 11 | //! [Written with AI entirely](https://need2know.ai/entirely-ai/code)\ |
| 12 | //! Anthropic Claude |
| 13 | |
| 14 | use crate::office::edit::{ |
| 15 | Find, |
| 16 | Piece, |
| 17 | Tally, |
| 18 | apply, |
| 19 | }; |
| 20 | use crate::office::odf::{ |
| 21 | NS_FO, |
| 22 | NS_OFFICE, |
| 23 | NS_STYLE, |
| 24 | NS_TABLE, |
| 25 | NS_TEXT, |
| 26 | NS_XLINK, |
| 27 | pkg, |
| 28 | }; |
| 29 | use crate::zip::{ |
| 30 | Method, |
| 31 | Zip, |
| 32 | }; |
| 33 | |
| 34 | use oxedyne_fe2o3_core::prelude::*; |
| 35 | use oxedyne_fe2o3_text::doc::{ |
| 36 | Align, |
| 37 | Block, |
| 38 | Cell, |
| 39 | Doc, |
| 40 | Inline, |
| 41 | Row, |
| 42 | }; |
| 43 | use oxedyne_fe2o3_text::xml::{ |
| 44 | Elem, |
| 45 | Node, |
| 46 | Xml, |
| 47 | }; |
| 48 | use oxedyne_fe2o3_text::xml::write::{ |
| 49 | Out, |
| 50 | escape, |
| 51 | }; |
| 52 | |
| 53 | use std::collections::BTreeMap; |
| 54 | |
| 55 | // Declared in the package's first member, which is what names the file. |
| 56 | pub const MEDIA: &str = "application/vnd.oasis.opendocument.text"; |
| 57 | |
| 58 | // The most a single part is inflated to. An `.odt` is one `content.xml`, so this is the whole |
| 59 | // document rather than a piece of it. |
| 60 | pub const MAX_PART: u64 = 64 * 1024 * 1024; |
| 61 | |
| 62 | /// What a created document could not carry. |
| 63 | #[derive(Clone, Debug, Default, PartialEq)] |
| 64 | pub struct Left { |
| 65 | pub images: Vec<String>, // those whose bytes could not be reached, by source |
| 66 | } |
| 67 | |
| 68 | impl Left { |
| 69 | |
| 70 | /// Did everything in the tree reach the document? |
| 71 | pub fn is_empty(&self) -> bool { |
| 72 | self.images.is_empty() |
| 73 | } |
| 74 | } |
| 75 | |
| 76 | pub fn write(doc: &Doc) -> Outcome<(Vec<u8>, Left)> { |
| 77 | let mut left = Left::default(); |
| 78 | let mut out = Out::declared(); |
| 79 | out.open("office:document-content", &[ |
| 80 | ("xmlns:office", NS_OFFICE), |
| 81 | ("xmlns:text", NS_TEXT), |
| 82 | ("xmlns:table", NS_TABLE), |
| 83 | ("xmlns:style", NS_STYLE), |
| 84 | ("xmlns:fo", NS_FO), |
| 85 | ("xmlns:xlink", NS_XLINK), |
| 86 | ("office:version", pkg::VERSION), |
| 87 | ]); |
| 88 | // The list style every list refers to. One definition serves both kinds: whether a level is |
| 89 | // numbered is a property of the LEVEL here, so a bulleted and a numbered list cannot share one -- |
| 90 | // hence two. |
| 91 | out.open("office:automatic-styles", &[]); |
| 92 | for (name, ordered) in [("LB", false), ("LN", true)] { |
| 93 | out.open("text:list-style", &[("style:name", name)]); |
| 94 | for lvl in 1..=9 { |
| 95 | let n = fmt!("{}", lvl); |
| 96 | match ordered { |
| 97 | true => out.open("text:list-level-style-number", &[ |
| 98 | ("text:level", &n), ("style:num-suffix", "."), ("style:num-format", "1"), |
| 99 | ]), |
| 100 | false => out.open("text:list-level-style-bullet", &[ |
| 101 | ("text:level", &n), ("text:bullet-char", "\u{2022}"), |
| 102 | ]), |
| 103 | } |
| 104 | res!(out.close(match ordered { |
| 105 | true => "text:list-level-style-number", |
| 106 | false => "text:list-level-style-bullet", |
| 107 | })); |
| 108 | } |
| 109 | res!(out.close("text:list-style")); |
| 110 | } |
| 111 | res!(out.close("office:automatic-styles")); |
| 112 | out.open("office:body", &[]); |
| 113 | out.open("office:text", &[]); |
| 114 | res!(blocks(&mut out, &doc.blocks, &mut left)); |
| 115 | res!(out.close("office:text")); |
| 116 | res!(out.close("office:body")); |
| 117 | res!(out.close("office:document-content")); |
| 118 | |
| 119 | let mut zip = pkg::start(MEDIA); |
| 120 | zip.set("content.xml", res!(out.finish()).into_bytes(), Method::Deflate); |
| 121 | zip.set("styles.xml", res!(pkg::styles_for(MEDIA)).into_bytes(), Method::Deflate); |
| 122 | zip.set("meta.xml", res!(pkg::meta()).into_bytes(), Method::Deflate); |
| 123 | res!(pkg::finish(&mut zip, MEDIA)); |
| 124 | Ok((res!(zip.write()), left)) |
| 125 | } |
| 126 | |
| 127 | /// The parameter is `run` and not `blocks`: a parameter of the same name as the function shadows it, |
| 128 | /// and the recursive calls below then resolve to a slice rather than to this. |
| 129 | fn blocks(out: &mut Out, run: &[Block], left: &mut Left) -> Outcome<()> { |
| 130 | for block in run { |
| 131 | match block { |
| 132 | Block::Heading { level, content } => { |
| 133 | let lvl = fmt!("{}", (*level).clamp(1, 10)); |
| 134 | out.open("text:h", &[("text:outline-level", &lvl)]); |
| 135 | res!(inlines(out, content, left)); |
| 136 | res!(out.close("text:h")); |
| 137 | } |
| 138 | Block::Para(content) => { |
| 139 | out.open("text:p", &[]); |
| 140 | res!(inlines(out, content, left)); |
| 141 | res!(out.close("text:p")); |
| 142 | } |
| 143 | Block::Quote(inner) => { |
| 144 | // A quotation is a run of paragraphs in the quotation style, because OpenDocument has |
| 145 | // no quotation element. That is what every writer does with one. |
| 146 | for b in inner { |
| 147 | match b { |
| 148 | Block::Para(content) => { |
| 149 | out.open("text:p", &[("text:style-name", "Quotations")]); |
| 150 | res!(inlines(out, content, left)); |
| 151 | res!(out.close("text:p")); |
| 152 | } |
| 153 | other => res!(blocks(out, &[other.clone()], left)), |
| 154 | } |
| 155 | } |
| 156 | } |
| 157 | Block::Code { text, .. } => { |
| 158 | for line in text.lines() { |
| 159 | out.open("text:p", &[("text:style-name", "Preformatted_20_Text")]); |
| 160 | res!(inlines(out, &[Inline::Text(line.to_string())], left)); |
| 161 | res!(out.close("text:p")); |
| 162 | } |
| 163 | } |
| 164 | Block::List { ordered, items } => res!(list(out, *ordered, items, left, true)), |
| 165 | Block::Rule => { |
| 166 | // OpenDocument has no thematic break element either; a paragraph in a style with a |
| 167 | // bottom border is what a reader recognises as one. |
| 168 | out.open("text:p", &[("text:style-name", "Horizontal_20_Line")]); |
| 169 | res!(out.close("text:p")); |
| 170 | } |
| 171 | Block::Table { head, rows, cols } => res!(table(out, head, rows, cols, left)), |
| 172 | Block::Div { content, .. } => res!(blocks(out, content, left)), |
| 173 | } |
| 174 | } |
| 175 | Ok(()) |
| 176 | } |
| 177 | |
| 178 | /// Nests by putting a list inside an item. |
| 179 | fn list( |
| 180 | out: &mut Out, |
| 181 | ordered: bool, |
| 182 | items: &[Vec<Block>], |
| 183 | left: &mut Left, |
| 184 | top: bool, |
| 185 | ) |
| 186 | -> Outcome<()> |
| 187 | { |
| 188 | let style = match ordered { |
| 189 | true => "LN", |
| 190 | false => "LB", |
| 191 | }; |
| 192 | match top { |
| 193 | true => out.open("text:list", &[("text:style-name", style)]), |
| 194 | // A nested list carries no style name: it inherits the level from where it sits, which is |
| 195 | // how OpenDocument nests. Naming the style again restarts the numbering. |
| 196 | false => out.open("text:list", &[]), |
| 197 | } |
| 198 | for item in items { |
| 199 | out.open("text:list-item", &[]); |
| 200 | for b in item { |
| 201 | match b { |
| 202 | Block::List { ordered, items } => res!(list(out, *ordered, items, left, false)), |
| 203 | other => res!(blocks(out, &[other.clone()], left)), |
| 204 | } |
| 205 | } |
| 206 | res!(out.close("text:list-item")); |
| 207 | } |
| 208 | res!(out.close("text:list")); |
| 209 | Ok(()) |
| 210 | } |
| 211 | |
| 212 | fn table( |
| 213 | out: &mut Out, |
| 214 | head: &Option<Row>, |
| 215 | rows: &[Row], |
| 216 | cols: &[Align], |
| 217 | left: &mut Left, |
| 218 | ) |
| 219 | -> Outcome<()> |
| 220 | { |
| 221 | let n = head.iter().chain(rows).map(|r| r.0.len()).max().unwrap_or(0).max(cols.len()); |
| 222 | if n == 0 { |
| 223 | return Ok(()); |
| 224 | } |
| 225 | out.open("table:table", &[("table:name", "Table1")]); |
| 226 | out.empty("table:table-column", &[("table:number-columns-repeated", &fmt!("{}", n))]); |
| 227 | if let Some(head) = head { |
| 228 | out.open("table:table-header-rows", &[]); |
| 229 | res!(row_of(out, head, n, left)); |
| 230 | res!(out.close("table:table-header-rows")); |
| 231 | } |
| 232 | for r in rows { |
| 233 | res!(row_of(out, r, n, left)); |
| 234 | } |
| 235 | res!(out.close("table:table")); |
| 236 | Ok(()) |
| 237 | } |
| 238 | |
| 239 | /// Padded to the width of the widest row. |
| 240 | fn row_of(out: &mut Out, r: &Row, n: usize, left: &mut Left) -> Outcome<()> { |
| 241 | let empty = Cell::default(); |
| 242 | out.open("table:table-row", &[]); |
| 243 | for i in 0..n { |
| 244 | out.open("table:table-cell", &[("office:value-type", "string")]); |
| 245 | out.open("text:p", &[]); |
| 246 | res!(inlines(out, &r.0.get(i).unwrap_or(&empty).0, left)); |
| 247 | res!(out.close("text:p")); |
| 248 | res!(out.close("table:table-cell")); |
| 249 | } |
| 250 | res!(out.close("table:table-row")); |
| 251 | Ok(()) |
| 252 | } |
| 253 | |
| 254 | fn inlines(out: &mut Out, content: &[Inline], left: &mut Left) -> Outcome<()> { |
| 255 | for item in content { |
| 256 | match item { |
| 257 | Inline::Text(t) => res!(text_run(out, t)), |
| 258 | Inline::Code(t) => { |
| 259 | out.open("text:span", &[("text:style-name", "Source_20_Text")]); |
| 260 | res!(text_run(out, t)); |
| 261 | res!(out.close("text:span")); |
| 262 | } |
| 263 | Inline::Emph { strong, content } => { |
| 264 | let style = match strong { |
| 265 | true => "Strong_20_Emphasis", |
| 266 | false => "Emphasis", |
| 267 | }; |
| 268 | out.open("text:span", &[("text:style-name", style)]); |
| 269 | res!(inlines(out, content, left)); |
| 270 | res!(out.close("text:span")); |
| 271 | } |
| 272 | Inline::Link { to, content } => { |
| 273 | // No relationship part, no id: the destination is on the element. This is the whole |
| 274 | // of the difference the module note is about. |
| 275 | out.open("text:a", &[("xlink:href", to), ("xlink:type", "simple")]); |
| 276 | res!(inlines(out, content, left)); |
| 277 | res!(out.close("text:a")); |
| 278 | } |
| 279 | Inline::Image { src, alt } => { |
| 280 | left.images.push(src.clone()); |
| 281 | out.open("text:span", &[("text:style-name", "Emphasis")]); |
| 282 | res!(text_run(out, alt)); |
| 283 | res!(out.close("text:span")); |
| 284 | } |
| 285 | Inline::Span { content, .. } => res!(inlines(out, content, left)), |
| 286 | Inline::Break => out.empty("text:line-break", &[]), |
| 287 | } |
| 288 | } |
| 289 | Ok(()) |
| 290 | } |
| 291 | |
| 292 | /// Text, with runs of spaces and tabs written as the elements OpenDocument has for them. |
| 293 | /// |
| 294 | /// A reader collapses whitespace in `text:p` exactly as HTML does, so two consecutive spaces arrive |
| 295 | /// as one unless the second is a `<text:s/>`. A listing indented with spaces loses its indentation |
| 296 | /// entirely without this, which is the whole point of writing a listing out. |
| 297 | fn text_run(out: &mut Out, text: &str) -> Outcome<()> { |
| 298 | let mut run = String::new(); |
| 299 | let mut chars = text.chars().peekable(); |
| 300 | while let Some(c) = chars.next() { |
| 301 | match c { |
| 302 | '\t' => { |
| 303 | if !run.is_empty() { |
| 304 | out.text(&run); |
| 305 | run.clear(); |
| 306 | } |
| 307 | out.empty("text:tab", &[]); |
| 308 | } |
| 309 | ' ' => { |
| 310 | let mut n = 1; |
| 311 | while chars.peek() == Some(&' ') { |
| 312 | chars.next(); |
| 313 | n += 1; |
| 314 | } |
| 315 | // A LEADING space is dropped by every reader, so ALL of a leading run goes out as |
| 316 | // `<text:s/>`. Keeping the first as a literal is what lost one space off the front of |
| 317 | // every indented line of every listing -- four became three, and only an external |
| 318 | // reader showed it. |
| 319 | let leading = run.is_empty(); |
| 320 | if !leading { |
| 321 | run.push(' '); |
| 322 | n -= 1; |
| 323 | out.text(&run); |
| 324 | run.clear(); |
| 325 | } |
| 326 | match n { |
| 327 | 0 => {} |
| 328 | 1 => out.empty("text:s", &[]), |
| 329 | _ => out.empty("text:s", &[("text:c", &fmt!("{}", n))]), |
| 330 | } |
| 331 | } |
| 332 | _ => run.push(c), |
| 333 | } |
| 334 | } |
| 335 | if !run.is_empty() { |
| 336 | out.text(&run); |
| 337 | } |
| 338 | Ok(()) |
| 339 | } |
| 340 | |
| 341 | #[derive(Clone, Debug, Default)] |
| 342 | pub struct Reading { |
| 343 | pub doc: Doc, |
| 344 | pub images: usize, // pictures held, which this does not draw |
| 345 | pub macros: bool, // a macro project is present; said, never run |
| 346 | } |
| 347 | |
| 348 | /// What one style says about what wears it. |
| 349 | /// |
| 350 | /// **A foreign document names almost nothing the way this crate's writer does.** LibreOffice writes |
| 351 | /// `T1`, `P2`, `L3` -- automatic styles, generated per document -- and puts the meaning in their |
| 352 | /// PROPERTIES. A reader matching on the name finds `Strong_20_Emphasis` in its own output and |
| 353 | /// nothing at all in anybody else's, so every bold word in a real document arrives as plain text. |
| 354 | /// The foreign fixture is what found that, and it is the same mistake the WordprocessingML reader |
| 355 | /// would have made had it trusted style ids. |
| 356 | #[derive(Clone, Debug, Default)] |
| 357 | struct Style { |
| 358 | bold: bool, // where the style says so itself |
| 359 | italic: bool, |
| 360 | mono: bool, // a monospaced face |
| 361 | parent: Option<String>, // followed for the properties this one does not set |
| 362 | } |
| 363 | |
| 364 | /// Every style a document defines, by name, and which list styles are numbered. |
| 365 | #[derive(Clone, Debug, Default)] |
| 366 | struct Styles { |
| 367 | by_name: BTreeMap<String, Style>, |
| 368 | lists: BTreeMap<String, bool>, // whether the FIRST level is numbered |
| 369 | } |
| 370 | |
| 371 | impl Styles { |
| 372 | |
| 373 | /// Follows what the style is based on. |
| 374 | fn of(&self, name: &str) -> Style { |
| 375 | let mut out = Style::default(); |
| 376 | let mut at = name.to_string(); |
| 377 | for _ in 0..8 { |
| 378 | let s = match self.by_name.get(&at) { |
| 379 | Some(s) => s.clone(), |
| 380 | None => break, |
| 381 | }; |
| 382 | out.bold |= s.bold; |
| 383 | out.italic |= s.italic; |
| 384 | out.mono |= s.mono; |
| 385 | match s.parent { |
| 386 | Some(p) => at = p, |
| 387 | None => break, |
| 388 | } |
| 389 | } |
| 390 | // A style this document does not define may still be one of the well-known names, which is |
| 391 | // what this crate's own writer emits. |
| 392 | let low = name.to_ascii_lowercase(); |
| 393 | out.bold |= low.contains("strong") || low.contains("bold"); |
| 394 | out.italic |= low.contains("emphasis") && !low.contains("strong"); |
| 395 | out.mono |= low.contains("source") || low.contains("teletype"); |
| 396 | out |
| 397 | } |
| 398 | |
| 399 | /// Quotation and listing, by name anywhere up the chain the style is based on. |
| 400 | fn para_kind(&self, name: &str) -> (bool, bool) { |
| 401 | let mut at = name.to_string(); |
| 402 | for _ in 0..8 { |
| 403 | let low = at.to_ascii_lowercase(); |
| 404 | if low.starts_with("quotation") || low.contains("block_20_quotation") { |
| 405 | return (true, false); |
| 406 | } |
| 407 | if low.starts_with("preformatted") || low.starts_with("source_20_text") |
| 408 | || low.contains("plain_20_text") |
| 409 | { |
| 410 | return (false, true); |
| 411 | } |
| 412 | match self.by_name.get(&at).and_then(|s| s.parent.clone()) { |
| 413 | Some(p) => at = p, |
| 414 | None => break, |
| 415 | } |
| 416 | } |
| 417 | (false, false) |
| 418 | } |
| 419 | } |
| 420 | |
| 421 | /// Added to what is already known rather than replacing it. |
| 422 | fn gather_styles(xml: &Xml, into: &mut Styles) { |
| 423 | for s in xml.all("style:style") { |
| 424 | let name = match s.attr("style:name") { |
| 425 | Some(n) => n.to_string(), |
| 426 | None => continue, |
| 427 | }; |
| 428 | let props = s.child("style:text-properties"); |
| 429 | into.by_name.insert(name, Style { |
| 430 | bold: props.and_then(|p| p.attr("fo:font-weight")) |
| 431 | .map(|v| v == "bold" || v == "600" || v == "700" || v == "800" || v == "900") |
| 432 | .unwrap_or(false), |
| 433 | italic: props.and_then(|p| p.attr("fo:font-style")) |
| 434 | .map(|v| v == "italic" || v == "oblique") |
| 435 | .unwrap_or(false), |
| 436 | mono: props.and_then(|p| p.attr("style:font-name").or(p.attr("fo:font-family"))) |
| 437 | .map(|f| { |
| 438 | let f = f.to_ascii_lowercase(); |
| 439 | f.contains("mono") || f.contains("courier") || f.contains("consol") |
| 440 | }) |
| 441 | .unwrap_or(false), |
| 442 | parent: s.attr("style:parent-style-name").map(|v| v.to_string()), |
| 443 | }); |
| 444 | } |
| 445 | for l in xml.all("text:list-style") { |
| 446 | let name = match l.attr("style:name") { |
| 447 | Some(n) => n.to_string(), |
| 448 | None => continue, |
| 449 | }; |
| 450 | // Whether the list is numbered is a property of its FIRST LEVEL, and asking whether ANY level |
| 451 | // is numbered is wrong: LibreOffice writes ten levels for every list style and makes level |
| 452 | // TEN a number even in a pure bullet list. That one line turned every bulleted list in a |
| 453 | // foreign document into a numbered one, and only the foreign fixture could have shown it. |
| 454 | let ordered = l.elems() |
| 455 | .find(|e| e.attr("text:level") == Some("1")) |
| 456 | .map(|e| e.name.qname == "text:list-level-style-number") |
| 457 | .unwrap_or(false); |
| 458 | into.lists.insert(name, ordered); |
| 459 | } |
| 460 | } |
| 461 | |
| 462 | pub fn read(bytes: &[u8]) -> Outcome<Reading> { |
| 463 | let zip = res!(Zip::read(bytes.to_vec())); |
| 464 | let mut out = Reading::default(); |
| 465 | // An OpenDocument macro lives in `Basic/`, not in a `vbaProject.bin`. |
| 466 | out.macros = zip.names().iter().any(|n| n.starts_with("Basic/")); |
| 467 | let src = res!(String::from_utf8(res!(zip.content_capped("content.xml", MAX_PART))), |
| 468 | Decode, String); |
| 469 | let xml = res!(Xml::parse(&src)); |
| 470 | let body = res!(res!(xml.root()).find(&["office:body", "office:text"]).ok_or_else(|| err!( |
| 471 | "This package has no <office:text>, so it is not a text document."; Invalid, Input, Missing))); |
| 472 | // Both parts, because a document splits its styles between them: the named ones a person applies |
| 473 | // live in `styles.xml` and the generated ones a writer makes live beside the content. |
| 474 | let mut styles = Styles::default(); |
| 475 | if zip.has("styles.xml") { |
| 476 | if let Ok(b) = zip.content_capped("styles.xml", MAX_PART) { |
| 477 | if let Ok(t) = String::from_utf8(b) { |
| 478 | if let Ok(x) = Xml::parse(&t) { |
| 479 | gather_styles(&x, &mut styles); |
| 480 | } |
| 481 | } |
| 482 | } |
| 483 | } |
| 484 | gather_styles(&xml, &mut styles); |
| 485 | let mut blocks = Vec::new(); |
| 486 | read_blocks(&xml, body, &mut blocks, &mut out.images, &styles); |
| 487 | // A listing arrives as one paragraph per line, so consecutive ones are one block. Left apart, |
| 488 | // every line of a program becomes its own fenced block. |
| 489 | out.doc = Doc { blocks: merge_code(blocks) }; |
| 490 | Ok(out) |
| 491 | } |
| 492 | |
| 493 | fn read_blocks(xml: &Xml, at: &Elem, out: &mut Vec<Block>, images: &mut usize, st: &Styles) { |
| 494 | for kid in at.elems() { |
| 495 | match kid.name.qname.as_str() { |
| 496 | "text:h" => { |
| 497 | // The level is ON the element. Nothing has to be resolved. |
| 498 | let level = kid.attr("text:outline-level") |
| 499 | .and_then(|v| v.parse::<u8>().ok()) |
| 500 | .unwrap_or(1) |
| 501 | .clamp(1, 6); |
| 502 | let content = read_inlines(xml, kid, images, st); |
| 503 | if !content.is_empty() { |
| 504 | out.push(Block::Heading { level, content }); |
| 505 | } |
| 506 | } |
| 507 | "text:p" => { |
| 508 | let content = read_inlines(xml, kid, images, st); |
| 509 | if content.is_empty() { |
| 510 | continue; |
| 511 | } |
| 512 | let style = kid.attr("text:style-name").unwrap_or(""); |
| 513 | let (quote, code) = st.para_kind(style); |
| 514 | if quote { |
| 515 | out.push(Block::Quote(vec![Block::Para(content)])); |
| 516 | } else if code { |
| 517 | out.push(Block::Code { |
| 518 | lang: None, |
| 519 | text: oxedyne_fe2o3_text::doc::text_of(&content), |
| 520 | }); |
| 521 | } else { |
| 522 | out.push(Block::Para(content)); |
| 523 | } |
| 524 | } |
| 525 | "text:list" => { |
| 526 | if let Some(list) = read_list(xml, kid, images, st) { |
| 527 | out.push(list); |
| 528 | } |
| 529 | } |
| 530 | "table:table" => { |
| 531 | if let Some(t) = read_table(xml, kid, images, st) { |
| 532 | out.push(t); |
| 533 | } |
| 534 | } |
| 535 | // A section, a frame, a change region: contributes nothing itself, and what it holds is |
| 536 | // read where it stood. The same rule as everywhere else in this crate. |
| 537 | _ => read_blocks(xml, kid, out, images, st), |
| 538 | } |
| 539 | } |
| 540 | } |
| 541 | |
| 542 | /// Whether it is numbered is a property of its STYLE, which is one level of indirection rather than |
| 543 | /// the two a `.xlsx` needs. A style this cannot find is read as a bullet, which says less than a |
| 544 | /// wrong number would. |
| 545 | fn read_list(xml: &Xml, at: &Elem, images: &mut usize, st: &Styles) -> Option<Block> { |
| 546 | // From the STYLE DEFINITION, not from the name: a foreign document calls its list style `L2` |
| 547 | // and puts `text:list-level-style-number` inside it, so a reader matching on the name reads |
| 548 | // every numbered list in the world as a bulleted one. |
| 549 | let style = at.attr("text:style-name").unwrap_or(""); |
| 550 | let ordered = match st.lists.get(style) { |
| 551 | Some(o) => *o, |
| 552 | None => style.contains("LN") || style.to_ascii_lowercase().contains("number"), |
| 553 | }; |
| 554 | let mut items = Vec::new(); |
| 555 | for item in at.children("text:list-item") { |
| 556 | let mut blocks = Vec::new(); |
| 557 | read_blocks(xml, item, &mut blocks, images, st); |
| 558 | if !blocks.is_empty() { |
| 559 | items.push(blocks); |
| 560 | } |
| 561 | } |
| 562 | match items.is_empty() { |
| 563 | true => None, |
| 564 | false => Some(Block::List { ordered, items }), |
| 565 | } |
| 566 | } |
| 567 | |
| 568 | fn read_table(xml: &Xml, at: &Elem, images: &mut usize, st: &Styles) -> Option<Block> { |
| 569 | let mut head = None; |
| 570 | let mut rows = Vec::new(); |
| 571 | if let Some(hr) = at.child("table:table-header-rows") { |
| 572 | if let Some(first) = hr.children("table:table-row").first() { |
| 573 | head = Some(read_row(xml, first, images, st)); |
| 574 | } |
| 575 | } |
| 576 | for tr in at.children("table:table-row") { |
| 577 | rows.push(read_row(xml, tr, images, st)); |
| 578 | } |
| 579 | if head.is_none() && rows.is_empty() { |
| 580 | return None; |
| 581 | } |
| 582 | let n = head.iter().chain(rows.iter()).map(|r| r.0.len()).max().unwrap_or(0); |
| 583 | Some(Block::Table { head, rows, cols: vec![Align::None; n] }) |
| 584 | } |
| 585 | |
| 586 | fn read_row(xml: &Xml, tr: &Elem, images: &mut usize, st: &Styles) -> Row { |
| 587 | let mut cells = Vec::new(); |
| 588 | for tc in tr.elems() { |
| 589 | match tc.name.qname.as_str() { |
| 590 | "table:table-cell" => { |
| 591 | let mut content = Vec::new(); |
| 592 | for (i, p) in tc.children("text:p").into_iter().enumerate() { |
| 593 | if i > 0 { |
| 594 | content.push(Inline::Break); |
| 595 | } |
| 596 | content.extend(read_inlines(xml, p, images, st)); |
| 597 | } |
| 598 | // A cell may say it repeats, which is how a run of empty cells is written. |
| 599 | let n = tc.attr("table:number-columns-repeated") |
| 600 | .and_then(|v| v.parse::<usize>().ok()) |
| 601 | .unwrap_or(1) |
| 602 | .min(1024); |
| 603 | for _ in 0..n { |
| 604 | cells.push(Cell(content.clone())); |
| 605 | } |
| 606 | } |
| 607 | "table:covered-table-cell" => cells.push(Cell::default()), |
| 608 | _ => {} |
| 609 | } |
| 610 | } |
| 611 | Row(cells) |
| 612 | } |
| 613 | |
| 614 | fn read_inlines(xml: &Xml, at: &Elem, images: &mut usize, st: &Styles) -> Vec<Inline> { |
| 615 | let mut out = Vec::new(); |
| 616 | for node in &at.kids { |
| 617 | match node { |
| 618 | oxedyne_fe2o3_text::xml::Node::Text(span) => { |
| 619 | let t = xml.text(span); |
| 620 | if !t.is_empty() { |
| 621 | out.push(Inline::Text(t)); |
| 622 | } |
| 623 | } |
| 624 | oxedyne_fe2o3_text::xml::Node::Elem(e) => match e.name.qname.as_str() { |
| 625 | "text:s" => { |
| 626 | let n = e.attr("text:c").and_then(|v| v.parse::<usize>().ok()).unwrap_or(1); |
| 627 | out.push(Inline::Text(" ".repeat(n.min(256)))); |
| 628 | } |
| 629 | "text:tab" => out.push(Inline::Text("\t".to_string())), |
| 630 | "text:line-break" => out.push(Inline::Break), |
| 631 | "draw:frame" | "draw:image" => { |
| 632 | *images += 1; |
| 633 | // A frame may hold a caption, which is prose and is kept. |
| 634 | out.extend(read_inlines(xml, e, images, st)); |
| 635 | } |
| 636 | "text:a" => { |
| 637 | let to = e.attr("xlink:href").unwrap_or("").to_string(); |
| 638 | let content = read_inlines(xml, e, images, st); |
| 639 | match content.is_empty() { |
| 640 | true => {} |
| 641 | false => out.push(Inline::Link { to, content }), |
| 642 | } |
| 643 | } |
| 644 | "text:span" => { |
| 645 | let style = st.of(e.attr("text:style-name").unwrap_or("")); |
| 646 | let content = read_inlines(xml, e, images, st); |
| 647 | if content.is_empty() { |
| 648 | continue; |
| 649 | } |
| 650 | let mut content = match style.mono { |
| 651 | true => vec![Inline::Code(oxedyne_fe2o3_text::doc::text_of(&content))], |
| 652 | false => content, |
| 653 | }; |
| 654 | if style.italic { |
| 655 | content = vec![Inline::Emph { strong: false, content }]; |
| 656 | } |
| 657 | if style.bold { |
| 658 | content = vec![Inline::Emph { strong: true, content }]; |
| 659 | } |
| 660 | out.extend(content); |
| 661 | } |
| 662 | // A bookmark, a note anchor, a soft page break: no words of their own. |
| 663 | "text:bookmark" | "text:bookmark-start" | "text:bookmark-end" |
| 664 | | "text:soft-page-break" | "text:tracked-changes" => {} |
| 665 | _ => out.extend(read_inlines(xml, e, images, st)), |
| 666 | }, |
| 667 | _ => {} |
| 668 | } |
| 669 | } |
| 670 | coalesce(out) |
| 671 | } |
| 672 | |
| 673 | /// A listing is one paragraph per line in this format, so a program arrives as a run of one-line |
| 674 | /// blocks. Left apart, every line of it renders as its own fenced block. |
| 675 | fn merge_code(blocks: Vec<Block>) -> Vec<Block> { |
| 676 | let mut out: Vec<Block> = Vec::with_capacity(blocks.len()); |
| 677 | for b in blocks { |
| 678 | match (out.last_mut(), b) { |
| 679 | (Some(Block::Code { text: a, .. }), Block::Code { text: b, .. }) => { |
| 680 | a.push('\n'); |
| 681 | a.push_str(&b); |
| 682 | } |
| 683 | (_, b) => out.push(b), |
| 684 | } |
| 685 | } |
| 686 | out |
| 687 | } |
| 688 | |
| 689 | /// Joins adjacent inlines that are marked alike. |
| 690 | fn coalesce(items: Vec<Inline>) -> Vec<Inline> { |
| 691 | let mut out: Vec<Inline> = Vec::with_capacity(items.len()); |
| 692 | for item in items { |
| 693 | match (out.last_mut(), item) { |
| 694 | (Some(Inline::Text(a)), Inline::Text(b)) => a.push_str(&b), |
| 695 | (Some(Inline::Code(a)), Inline::Code(b)) => a.push_str(&b), |
| 696 | ( |
| 697 | Some(Inline::Emph { strong: sa, content: ca }), |
| 698 | Inline::Emph { strong: sb, content: cb }, |
| 699 | ) if *sa == sb => { |
| 700 | let mut joined = std::mem::take(ca); |
| 701 | joined.extend(cb); |
| 702 | *ca = coalesce(joined); |
| 703 | } |
| 704 | (_, item) => out.push(item), |
| 705 | } |
| 706 | } |
| 707 | out.retain(|i| !matches!(i, Inline::Text(t) if t.is_empty())); |
| 708 | out |
| 709 | } |
| 710 | |
| 711 | // --------------------------------------------------------------------------- |
| 712 | // Editing an `.odt` in place |
| 713 | // --------------------------------------------------------------------------- |
| 714 | |
| 715 | /// What an edit of an `.odt` produced. |
| 716 | #[derive(Clone, Debug, Default)] |
| 717 | pub struct Edited { |
| 718 | pub bytes: Vec<u8>, |
| 719 | pub tallies: Vec<Tally>, // one per edit asked for, in order |
| 720 | pub runs: usize, // runs of text rewritten |
| 721 | } |
| 722 | |
| 723 | /// Replaces text in an `.odt`, leaving every other byte of the package as it arrived. |
| 724 | /// |
| 725 | /// The counterpart of [`crate::office::docx::edit::edit`] and the same design: the text of a paragraph |
| 726 | /// is the concatenation of its runs, a match is found there and pushed back down onto the runs it |
| 727 | /// covered, and [`crate::zip`] copies every member nobody touched. `styles.xml`, `meta.xml`, the |
| 728 | /// manifest, the macros and the pictures are never opened. |
| 729 | /// |
| 730 | /// Only `content.xml` is searched, so a phrase in a header or a footer -- which live in `styles.xml` -- |
| 731 | /// reports as absent rather than being changed in one of two places. |
| 732 | pub fn edit(bytes: &[u8], edits: &[Find]) -> Outcome<Edited> { |
| 733 | if edits.is_empty() { |
| 734 | return Err(err!("An edit of a document was asked for with no edits in it."; Invalid, Input)); |
| 735 | } |
| 736 | let mut zip = res!(Zip::read(bytes.to_vec())); |
| 737 | let src = res!(String::from_utf8(res!(zip.content_capped("content.xml", MAX_PART))), |
| 738 | Decode, String); |
| 739 | let mut xml = res!(Xml::parse(&src)); |
| 740 | let body = res!(res!(xml.root()).find(&["office:body", "office:text"]).ok_or_else(|| err!( |
| 741 | "This package has no <office:text>, so it is not a text document."; Invalid, Input, Missing))); |
| 742 | let mut groups = Vec::new(); |
| 743 | edit_walk(&xml, body, &mut groups); |
| 744 | let (changes, tallies) = res!(apply(&groups, edits)); |
| 745 | let runs = changes.len(); |
| 746 | for c in &changes { |
| 747 | res!(xml.splice(c.piece.span.clone(), content_markup(&c.text))); |
| 748 | } |
| 749 | zip.set("content.xml", xml.render().into_bytes(), Method::Deflate); |
| 750 | Ok(Edited { bytes: res!(zip.write()), tallies, runs }) |
| 751 | } |
| 752 | |
| 753 | /// Every paragraph and heading at or below an element, as a group of text pieces each. |
| 754 | /// |
| 755 | /// A nested paragraph -- a frame's caption inside a paragraph -- gets its own group and its text is not |
| 756 | /// also in the enclosing one, for the reason `docx::edit` gives: two splices over one span is a |
| 757 | /// refusal, and rightly. |
| 758 | fn edit_walk(xml: &Xml, at: &Elem, out: &mut Vec<Vec<Piece>>) { |
| 759 | match at.name.qname.as_str() { |
| 760 | "text:p" | "text:h" => { |
| 761 | let slot = out.len(); |
| 762 | out.push(Vec::new()); |
| 763 | let mut group = Vec::new(); |
| 764 | edit_gather(xml, at, &mut group, out); |
| 765 | out[slot] = group; |
| 766 | } |
| 767 | _ => { |
| 768 | for kid in at.elems() { |
| 769 | edit_walk(xml, kid, out); |
| 770 | } |
| 771 | } |
| 772 | } |
| 773 | } |
| 774 | |
| 775 | /// One paragraph's own text, run by run. |
| 776 | /// |
| 777 | /// Character data, and the three elements that ARE text: `<text:s>` is a run of spaces, because |
| 778 | /// OpenDocument collapses literal ones; `<text:tab>` is a tab; `<text:line-break>` is a newline. A |
| 779 | /// reader that skipped them would match `Q1 2026` against a document holding `Q1<text:s/>2026` and |
| 780 | /// report the phrase absent -- and the phrase is there, it is what a person typed. |
| 781 | fn edit_gather(xml: &Xml, at: &Elem, group: &mut Vec<Piece>, out: &mut Vec<Vec<Piece>>) { |
| 782 | for kid in &at.kids { |
| 783 | match kid { |
| 784 | Node::Text(span) => group.push(Piece::new(span.clone(), xml.text(span))), |
| 785 | Node::Elem(e) => match e.name.qname.as_str() { |
| 786 | "text:p" | "text:h" => edit_walk(xml, e, out), |
| 787 | "text:s" => { |
| 788 | let n = e.attr("text:c").and_then(|v| v.parse::<usize>().ok()).unwrap_or(1); |
| 789 | group.push(Piece::new(e.span.clone(), " ".repeat(n.min(4096)))); |
| 790 | } |
| 791 | "text:tab" => group.push(Piece::new(e.span.clone(), "\t")), |
| 792 | "text:line-break" => group.push(Piece::new(e.span.clone(), "\n")), |
| 793 | // A footnote's body, a comment's body and an index mark hold text that is not the |
| 794 | // paragraph's own, and replacing in them would edit two places for one phrase. |
| 795 | "text:note" | "office:annotation" => {} |
| 796 | _ => edit_gather(xml, e, group, out), |
| 797 | }, |
| 798 | _ => {} |
| 799 | } |
| 800 | } |
| 801 | } |
| 802 | |
| 803 | /// Text as OpenDocument paragraph content: the markup that means exactly these characters. |
| 804 | /// |
| 805 | /// Three characters cannot be written literally and survive. A run of two or more spaces is collapsed |
| 806 | /// to one by every reader, so it becomes `<text:s text:c="n"/>`; a space at either end of the run is |
| 807 | /// collapsed for the same reason and becomes `<text:s/>`; a tab and a newline have elements of their |
| 808 | /// own. Writing them literally produces a file that opens and says something slightly different from |
| 809 | /// what the edit asked for, which is the worst of the available outcomes. |
| 810 | pub fn content_markup(text: &str) -> String { |
| 811 | let mut out = String::with_capacity(text.len() + 16); |
| 812 | let chars: Vec<char> = text.chars().collect(); |
| 813 | let mut i = 0; |
| 814 | while i < chars.len() { |
| 815 | match chars[i] { |
| 816 | ' ' => { |
| 817 | let mut n = 0; |
| 818 | while i + n < chars.len() && chars[i + n] == ' ' { |
| 819 | n += 1; |
| 820 | } |
| 821 | // A single space with a character either side of it is safe as itself, and leaving it |
| 822 | // alone keeps the markup readable. Anywhere else it is spelled out. |
| 823 | let interior = i > 0 && i + n < chars.len(); |
| 824 | match n == 1 && interior { |
| 825 | true => out.push(' '), |
| 826 | false => match n { |
| 827 | 1 => out.push_str("<text:s/>"), |
| 828 | _ => out.push_str(&fmt!("<text:s text:c=\"{}\"/>", n)), |
| 829 | }, |
| 830 | } |
| 831 | i += n; |
| 832 | } |
| 833 | '\t' => { |
| 834 | out.push_str("<text:tab/>"); |
| 835 | i += 1; |
| 836 | } |
| 837 | '\n' | '\r' => { |
| 838 | out.push_str("<text:line-break/>"); |
| 839 | i += 1; |
| 840 | } |
| 841 | c => { |
| 842 | out.push_str(&escape(&c.to_string())); |
| 843 | i += 1; |
| 844 | } |
| 845 | } |
| 846 | } |
| 847 | out |
| 848 | } |
| 849 | |
| 850 | /// The text of the body, paragraph by paragraph, as an edit sees it. |
| 851 | /// |
| 852 | /// The strings a `find` is matched against, which is not the same as the document as prose: this is |
| 853 | /// where a run split by a style, or a space written as `<text:s/>`, shows up. |
| 854 | pub fn body_text(bytes: &[u8]) -> Outcome<Vec<String>> { |
| 855 | let zip = res!(Zip::read(bytes.to_vec())); |
| 856 | let src = res!(String::from_utf8(res!(zip.content_capped("content.xml", MAX_PART))), |
| 857 | Decode, String); |
| 858 | let xml = res!(Xml::parse(&src)); |
| 859 | let body = res!(res!(xml.root()).find(&["office:body", "office:text"]).ok_or_else(|| err!( |
| 860 | "This package has no <office:text>, so it is not a text document."; Invalid, Input, Missing))); |
| 861 | let mut groups = Vec::new(); |
| 862 | edit_walk(&xml, body, &mut groups); |
| 863 | Ok(groups.iter() |
| 864 | .map(|g| g.iter().map(|p| p.text.as_str()).collect::<Vec<_>>().concat()) |
| 865 | .collect()) |
| 866 | } |