vectorcraft/crates/engine/src/cmd/spelling.rs

//! Edit → Spelling: the words of the document's type that its dictionaries don't know
//! (`text.spelling.check`) and the words they suggest instead (`text.spelling.suggest`), checked
//! by craft-proof in each run's language
//! ([Check and correct spelling](https://helpx.adobe.com/illustrator/desktop/design-with-text/edit-format-text/check-and-correct-spelling.html)).
//! A word is changed with `text.editRange` over the bytes the check gives. Add and Edit Custom
//! Dictionary keep words in the custom dictionary (`text.spelling.dictionary`, `addWord`,
//! `removeWord`, `changeWord`), which every document's check takes as right; Ignore All
//! (`text.spelling.ignoreAll`) skips a word in the active document only.

use std::collections::BTreeSet;
use std::ops::Range;
use std::sync::atomic::{AtomicU64, Ordering};

use craft_proof::{IssueKind, Options, ProofLang};
use serde_json::{Value, json};
use vectorcraft_doc::{Document, NodeId, NodeKind, TextObject};

use super::menucmds::whole_or;
use super::typemenu::{all_texts, text_ids};
use super::{CommandSpec, Result, Session, always, bad, checked_ids_param, cmd, has_doc, str_param};

/// Most unknown words one check lists; past it the result says `more`.
pub const MAX_WORDS: usize = 10_000;

/// Most suggestions one word gets.
pub const MAX_SUGGESTIONS: u64 = 20;

/// Longest word suggestions are looked for, in bytes: the search costs more with the length.
const MAX_WORD_LEN: usize = 128;

/// The Edit › Spelling commands.
pub fn specs() -> Vec<CommandSpec> {
    vec![
        cmd!(
            query "text.spelling.check",
            "Check Spelling",
            [],
            None,
            "{ids?: [id…] (type objects, or groups holding them; default: every visible, unlocked type object)} the words the dictionaries don't know, each in its run's language (English when it has none; none in a language without a dictionary), each paragraph on its own, in paint order (given ids too) → {words: [{id, start: byte, end: byte, word, lang}…], more: bool (more than 10000 words, the rest left out)}. text.editRange {id, start, end, insert} changes one",
            has_doc,
            check
        ),
        cmd!(
            query "text.spelling.suggest",
            "Spelling Suggestions",
            [],
            None,
            "{word, max?: 1..20 (default 5), lang?: language tag (default English)} the words the dictionary of lang suggests instead of word, best first → {suggestions: [word…]}",
            has_doc,
            suggest
        ),
        cmd!(
            query "text.spelling.dictionary",
            "Edit Custom Dictionary",
            [],
            None,
            "{} the words of the custom dictionary, sorted (lower case, as they are kept) → {words: [word…]}",
            always,
            dictionary
        ),
        cmd!(
            query "text.spelling.addWord",
            "Add to Custom Dictionary",
            [],
            None,
            "{word} adds word (one word, at most 64 characters; kept in lower case) to the custom dictionary, which every language's check takes as right, and saves it → {word, added: bool (false: it was there already)}",
            always,
            add_word
        ),
        cmd!(
            query "text.spelling.removeWord",
            "Delete from Custom Dictionary",
            [],
            None,
            "{word} deletes word from the custom dictionary and saves it (an error when it isn't there) → {word}",
            always,
            remove_word
        ),
        cmd!(
            query "text.spelling.changeWord",
            "Change in Custom Dictionary",
            [],
            None,
            "{word, to} replaces word in the custom dictionary with to and saves it (an error when word isn't there) → {word: to}",
            always,
            change_word
        ),
        cmd!(
            query "text.spelling.editDictionary",
            "Edit Custom Dictionary",
            [],
            None,
            "{remove?: [word…], add?: [word…] (each one word, at most 64 characters; at most 10000 of each)} deletes the words of remove that are there and adds those of add, then saves once: all of it, or none of it when the save fails (Edit Custom Dictionary's Done) → {removed: [word…], added: [word…]}",
            always,
            edit_dictionary
        ),
        cmd!(
            query "text.spelling.ignoreAll",
            "Ignore All",
            [],
            None,
            "{word} the active document's spelling checks skip word from now on, until it closes (not saved, not undoable) → {word}",
            has_doc,
            ignore_all
        ),
    ]
}

/// `text.spelling.check`: the unknown words of the given type objects (or every visible, unlocked
/// one), in paint order, each paragraph checked on its own, at most [`MAX_WORDS`] of them.
fn check(s: &mut Session, p: &Value) -> Result<Value> {
    const C: &str = "text.spelling.check";
    // Given `ids` are checked like Find and Replace's, so a malformed one is an error rather than
    // a check of the whole document.
    let ids = match p.get("ids").filter(|v| !v.is_null()) {
        Some(_) => in_paint_order(&s.doc()?.doc, text_ids(&s.doc()?.doc, &checked_ids_param(s, p, "ids", C)?)),
        None => all_texts(&s.doc()?.doc),
    };
    let st = s.doc()?;
    let mut words = vec![];
    let mut more = false;
    for id in ids {
        let Some(NodeKind::Text(t)) = st.doc.node(id).map(|n| &n.kind) else { continue };
        let text = t.plain_text();
        let (found, cut) = misspellings(t, &st.spelling_ignored, MAX_WORDS - words.len());
        for m in found {
            let Some(word) = text.get(m.range.clone()) else { continue };
            words.push(json!({ "id": id.0, "start": m.range.start, "end": m.range.end, "word": word, "lang": m.lang.tag() }));
        }
        if cut {
            more = true;
            break;
        }
    }
    Ok(json!({ "words": words, "more": more }))
}

/// One word of a type object its dictionaries don't know ([`misspellings`]).
#[derive(Clone, Debug, PartialEq)]
pub struct Misspelling {
    /// The word's bytes in the object's [`TextObject::plain_text`].
    pub range: Range<usize>,
    /// The language it was checked in.
    pub lang: ProofLang,
}

/// The words of `t` its dictionaries don't know, in text order: each run checked in its
/// language, each paragraph on its own, leaving out the words in `ignored` (a document's Ignore
/// All words). At most `max` of them; the flag says the list was cut short there. What
/// `text.spelling.check` lists for one object, and what Auto Spell Check underlines.
pub fn misspellings(t: &TextObject, ignored: &BTreeSet<String>, max: usize) -> (Vec<Misspelling>, bool) {
    let text = t.plain_text();
    // Each run in its own language, at its bytes of the plain text.
    let mut spans: Vec<(Range<usize>, ProofLang)> = Vec::with_capacity(t.runs.len());
    let mut at = 0;
    for r in &t.runs {
        let end = at + r.text.len();
        spans.push((at..end, ProofLang::from_tag(r.style.language.as_deref())));
        at = end;
    }
    let opts = Options::default();
    let mut found = vec![];
    // craft-proof checks one paragraph at a time (its language detection reads the paragraph),
    // and the check stops at the paragraph where the list fills up rather than proofing the rest
    // of the text.
    let mut para_start = 0;
    for para in text.split('\n') {
        let para_end = para_start + para.len();
        let para_spans: Vec<(Range<usize>, ProofLang)> = spans
            .iter()
            .filter_map(|(r, lang)| {
                let (a, b) = (r.start.max(para_start), r.end.min(para_end));
                (a < b).then(|| (a - para_start..b - para_start, *lang))
            })
            .collect();
        for issue in craft_proof::check_text_with(para, &para_spans, &opts, false, false) {
            if issue.kind != IssueKind::Spelling {
                continue;
            }
            let (start, end) = (para_start + issue.start, para_start + issue.end);
            let Some(word) = text.get(start..end) else { continue };
            if craft_proof::custom_word(word).is_some_and(|w| ignored.contains(&w)) {
                continue;
            }
            if found.len() == max {
                return (found, true);
            }
            found.push(Misspelling { range: start..end, lang: issue.lang });
        }
        // Past the paragraph and its `\n`.
        para_start = para_end + 1;
    }
    (found, false)
}

/// Goes up each time the custom dictionary may have changed: an add, delete or change (saved or
/// not, since a save takes in the words another window saved), and a dictionary file attached.
static DICTIONARY_GENERATION: AtomicU64 = AtomicU64::new(0);

/// The custom dictionary's generation: what [`misspellings`] found before it last changed may be
/// out of date (Auto Spell Check checks the type again).
pub fn dictionary_generation() -> u64 {
    DICTIONARY_GENERATION.load(Ordering::Relaxed)
}

fn dictionary_changed() {
    DICTIONARY_GENERATION.fetch_add(1, Ordering::Relaxed);
}

/// `ids` sorted back to front, as the document paints them: by their index paths, which are in
/// paint order.
fn in_paint_order(d: &Document, mut ids: Vec<NodeId>) -> Vec<NodeId> {
    let paths = d.index_paths(ids.iter().copied());
    ids.sort_by(|a, b| paths.get(a).cmp(&paths.get(b)));
    ids
}

/// `text.spelling.suggest`: up to `max` of the words the dictionary of `lang` suggests for `word`.
fn suggest(_s: &mut Session, p: &Value) -> Result<Value> {
    const C: &str = "text.spelling.suggest";
    let word = str_param(p, "word").map(str::trim).filter(|w| !w.is_empty()).ok_or_else(|| bad(C, "missing `word`"))?;
    let max = whole_or(p, "max", 5, C)?;
    if !(1..=MAX_SUGGESTIONS).contains(&max) {
        return Err(bad(C, format!("`max` must be 1 to {MAX_SUGGESTIONS}")));
    }
    if word.len() > MAX_WORD_LEN {
        return Ok(json!({ "suggestions": [] }));
    }
    let lang = ProofLang::from_tag(str_param(p, "lang"));
    // `max` is at most MAX_SUGGESTIONS, so it fits.
    let suggestions = craft_proof::suggest_in(word, usize::try_from(max).unwrap_or(5), lang);
    Ok(json!({ "suggestions": suggestions }))
}

/// `w` as the custom dictionary keeps it (trimmed, lower case), or None when it isn't one word of at
/// most `craft_proof::MAX_CUSTOM_WORD_CHARS` characters: what `text.spelling.addWord` takes.
pub fn custom_word(w: &str) -> Option<String> {
    craft_proof::custom_word(w)
}

/// `p`'s `key` as a custom-dictionary word (trimmed, lower case), or an error naming what it must be.
fn word_param(p: &Value, key: &str, command: &str) -> Result<String> {
    let w = str_param(p, key).ok_or_else(|| bad(command, format!("missing `{key}`")))?;
    craft_proof::custom_word(w)
        .ok_or_else(|| bad(command, format!("`{key}` must be one word of at most {} characters", craft_proof::MAX_CUSTOM_WORD_CHARS)))
}

/// Largest custom dictionary file taken in, in bytes: craft-proof reads no more of one (its
/// `MAX_CUSTOM_FILE_BYTES`), so a larger file would lose its other words when written back.
pub const MAX_DICTIONARY_FILE_BYTES: u64 = 1_000_000;

/// Whether the custom dictionary file at `path` can be written over without losing words: it is
/// missing (a first save creates it), or a file that can be read whole. Otherwise why not.
fn dictionary_file_readable(path: &std::path::Path) -> std::result::Result<(), String> {
    match std::fs::metadata(path) {
        Err(e) if e.kind() == std::io::ErrorKind::NotFound => Ok(()),
        Err(e) => Err(e.to_string()),
        Ok(m) if !m.is_file() => Err("it isn't a file".into()),
        Ok(m) if m.len() > MAX_DICTIONARY_FILE_BYTES => Err(format!("it is larger than {MAX_DICTIONARY_FILE_BYTES} bytes")),
        Ok(_) => {
            // Read it whole, as craft-proof will: a file that opens but then can't be read would
            // read as no words there.
            use std::io::Read;
            let mut bytes = Vec::new();
            let file = std::fs::File::open(path).map_err(|e| e.to_string())?;
            file.take(MAX_DICTIONARY_FILE_BYTES + 1).read_to_end(&mut bytes).map_err(|e| e.to_string())?;
            if bytes.len() as u64 > MAX_DICTIONARY_FILE_BYTES {
                return Err(format!("it is larger than {MAX_DICTIONARY_FILE_BYTES} bytes"));
            }
            Ok(())
        }
    }
}

/// After a failed save took a command's change back: take in the words the file holds again, so
/// a word another window saved meanwhile, which the change had added too, stays.
fn retake_file_words(s: &Session) {
    if let Some(path) = s.dictionary_file.as_deref().filter(|p| dictionary_file_readable(p).is_ok()) {
        craft_proof::load_user_dictionary(path);
    }
}

/// Write the custom dictionary to its file when the session has one, leaving out `removed`. A file
/// that is there but can't be read whole is left alone rather than written over with only the
/// words read. On an error the caller takes its own change back, so memory and file agree.
fn save_dictionary(s: &Session, removed: &[String], command: &str) -> Result<()> {
    let Some(path) = &s.dictionary_file else { return Ok(()) };
    if let Err(why) = dictionary_file_readable(path) {
        return Err(bad(command, format!("the custom dictionary {} can't be read ({why}), so it is left as it is", path.display())));
    }
    craft_proof::save_user_dictionary(path, removed)
        .map_err(|e| bad(command, format!("the custom dictionary couldn't be saved to {}: {e}", path.display())))
}

/// `text.spelling.dictionary`: the custom dictionary's words.
fn dictionary(_s: &mut Session, _p: &Value) -> Result<Value> {
    Ok(json!({ "words": craft_proof::user_dictionary() }))
}

/// `text.spelling.addWord`: Add, and Edit Custom Dictionary's Add.
fn add_word(s: &mut Session, p: &Value) -> Result<Value> {
    const C: &str = "text.spelling.addWord";
    let word = word_param(p, "word", C)?;
    let added = !craft_proof::user_dictionary().contains(&word);
    craft_proof::add_word(&word);
    let saved = save_dictionary(s, &[], C);
    dictionary_changed();
    saved.inspect_err(|_| {
        // Only this command's word: words the save took in from the file stay.
        if added {
            craft_proof::remove_word(&word);
            retake_file_words(s);
        }
    })?;
    Ok(json!({ "word": word, "added": added }))
}

/// `text.spelling.removeWord`: Edit Custom Dictionary's Delete.
fn remove_word(s: &mut Session, p: &Value) -> Result<Value> {
    const C: &str = "text.spelling.removeWord";
    let word = word_param(p, "word", C)?;
    if !craft_proof::remove_word(&word) {
        return Err(bad(C, format!("`{word}` isn't in the custom dictionary")));
    }
    let saved = save_dictionary(s, std::slice::from_ref(&word), C);
    dictionary_changed();
    saved.inspect_err(|_| craft_proof::add_word(&word))?;
    Ok(json!({ "word": word }))
}

/// `text.spelling.changeWord`: Edit Custom Dictionary's Change.
fn change_word(s: &mut Session, p: &Value) -> Result<Value> {
    const C: &str = "text.spelling.changeWord";
    let word = word_param(p, "word", C)?;
    let to = word_param(p, "to", C)?;
    if !craft_proof::user_dictionary().contains(&word) {
        return Err(bad(C, format!("`{word}` isn't in the custom dictionary")));
    }
    if word == to {
        let saved = save_dictionary(s, &[], C);
        dictionary_changed();
        saved?;
        return Ok(json!({ "word": to }));
    }
    let to_is_new = !craft_proof::user_dictionary().contains(&to);
    craft_proof::remove_word(&word);
    craft_proof::add_word(&to);
    let saved = save_dictionary(s, std::slice::from_ref(&word), C);
    dictionary_changed();
    saved.inspect_err(|_| {
        if to_is_new {
            craft_proof::remove_word(&to);
            retake_file_words(s);
        }
        craft_proof::add_word(&word);
    })?;
    Ok(json!({ "word": to }))
}

/// Most words `text.spelling.editDictionary` takes in each of its lists.
const MAX_EDIT_WORDS: usize = 10_000;

/// `p`'s `key`, a list of custom-dictionary words (none when it is missing), deduplicated.
fn word_list(p: &Value, key: &str, command: &str) -> Result<BTreeSet<String>> {
    let Some(v) = p.get(key) else { return Ok(BTreeSet::new()) };
    let words = v.as_array().ok_or_else(|| bad(command, format!("`{key}` must be a list of words")))?;
    if words.len() > MAX_EDIT_WORDS {
        return Err(bad(command, format!("`{key}` holds more than {MAX_EDIT_WORDS} words")));
    }
    words
        .iter()
        .map(|w| {
            w.as_str()
                .and_then(craft_proof::custom_word)
                .ok_or_else(|| bad(command, format!("`{key}` must hold single words of at most {} characters", craft_proof::MAX_CUSTOM_WORD_CHARS)))
        })
        .collect()
}

/// `text.spelling.editDictionary`: Edit Custom Dictionary's Done. Every delete and add is made,
/// then saved once; when the save fails, all of them are taken back, so a failed Done leaves the
/// dictionary as it was rather than half changed.
fn edit_dictionary(s: &mut Session, p: &Value) -> Result<Value> {
    const C: &str = "text.spelling.editDictionary";
    let (remove, add) = (word_list(p, "remove", C)?, word_list(p, "add", C)?);
    let before: BTreeSet<String> = craft_proof::user_dictionary().into_iter().collect();
    // A word in both lists stays; a word to delete that another window deleted already is skipped.
    let removed: Vec<String> = remove.iter().filter(|w| !add.contains(*w) && before.contains(*w)).cloned().collect();
    let added: Vec<String> = add.iter().filter(|w| !before.contains(*w)).cloned().collect();
    if removed.is_empty() && added.is_empty() {
        return Ok(json!({ "removed": removed, "added": added }));
    }
    for w in &removed {
        craft_proof::remove_word(w);
    }
    for w in &added {
        craft_proof::add_word(w);
    }
    let saved = save_dictionary(s, &removed, C);
    dictionary_changed();
    saved.inspect_err(|_| {
        for w in &added {
            craft_proof::remove_word(w);
        }
        for w in &removed {
            craft_proof::add_word(w);
        }
        // Words the save took in from the file stay, as for one word.
        if !added.is_empty() {
            retake_file_words(s);
        }
    })?;
    Ok(json!({ "removed": removed, "added": added }))
}

/// `text.spelling.ignoreAll`: Ignore All, for the active document.
fn ignore_all(s: &mut Session, p: &Value) -> Result<Value> {
    const C: &str = "text.spelling.ignoreAll";
    let word = word_param(p, "word", C)?;
    // Not a change to the document, so its revision stays: a tool edit computed meanwhile still
    // applies (`command_sweep`'s check of commands that don't wait for one).
    s.doc_mut()?.spelling_ignored.insert(word.clone());
    Ok(json!({ "word": word }))
}

impl Session {
    /// Keep the custom dictionary in the file at `path` (the desktop app's, beside its
    /// preferences), and take in the words already there. A missing file reads as no words and is
    /// created by the first save. A file that is there but can't be read whole is logged and left
    /// as it is: changes to the dictionary are refused rather than written over it.
    pub fn attach_custom_dictionary(&mut self, path: std::path::PathBuf) {
        match dictionary_file_readable(&path) {
            Ok(()) => {
                craft_proof::load_user_dictionary(&path);
            }
            Err(why) => log::warn!("not reading the custom dictionary {}: {why}; it is left as it is", path.display()),
        }
        self.dictionary_file = Some(path);
        dictionary_changed();
    }
}