inkhaven 1.2.4

//! Convert a single Scrivener document's RTF body to Typst
//! markup.
//!
//! Uses `rtf-parser-tt` to tokenise the RTF into a stream of
//! style transitions + text runs. We translate that stream
//! into a small Typst subset that mirrors what inkhaven users
//! actually write: paragraph breaks, bold (`*…*`), italic
//! (`_…_`), Unicode-safe text. Headings emit as `= …` when
//! the source RTF used a "Heading 1" style; otherwise plain
//! prose stays plain prose.
//!
//! What we deliberately drop:
//! * Custom fonts / colours / sizes — Scrivener users almost
//!   always override these during compile anyway.
//! * Embedded images — handled separately by the import
//!   orchestrator (they live in `Files/data/` alongside the
//!   RTF and need to be copied to the project's image dir).
//! * Footnotes / inline annotations — out of scope for v1;
//!   could land later as Typst footnotes.
//!
//! When the parser bails on malformed RTF, we fall back to the
//! "extract plain text" path: strip control words, keep the
//! visible characters. Lossy but never crashes on a
//! pathological input.

use anyhow::Result;

/// Convert RTF bytes to a Typst-friendly source string.
/// Lossy by design — see module docs for what we drop.
pub fn rtf_to_typst(rtf_bytes: &[u8]) -> Result<String> {
    // Try the structured parser first. If it returns Err, fall
    // back to a brute-force plain-text extraction so the user
    // doesn't lose the document body to a syntactic glitch.
    match parse_structured(rtf_bytes) {
        Ok(s) => Ok(s),
        Err(_) => Ok(strip_to_plain_text(rtf_bytes)),
    }
}

/// Structured parse via `rtf-parser-tt`. Walks the
/// `parse(...)`'s `body` tokens and emits Typst markup.
fn parse_structured(rtf_bytes: &[u8]) -> Result<String> {
    use rtf_parser_tt::lexer::Lexer;
    use rtf_parser_tt::parser::Parser;

    let s = std::str::from_utf8(rtf_bytes)
        .map_err(|e| anyhow::anyhow!("rtf: not UTF-8: {e}"))?;
    let tokens = Lexer::scan(s)
        .map_err(|e| anyhow::anyhow!("rtf lex: {e}"))?;
    let mut parser = Parser::new(tokens);
    let doc = parser
        .parse()
        .map_err(|e| anyhow::anyhow!("rtf parse: {e}"))?;

    // `doc.body` is a Vec<StyleBlock>. Each block has a `painter`
    // (style state) + `text` (a String). We translate by emitting
    // wrappers around runs of consecutive identical style and
    // splitting on RTF paragraph markers (`\n` is the post-parse
    // representation of `\par`).
    let mut out = String::new();
    for block in &doc.body {
        let painter = &block.painter;
        let raw = &block.text;
        // Skip empty blocks.
        if raw.is_empty() {
            continue;
        }
        // Each block can contain multiple lines (one per `\par`
        // in the source); split + re-emit so paragraph breaks
        // become `\n\n` in Typst.
        for (i, line) in raw.split('\n').enumerate() {
            if i > 0 {
                ensure_paragraph_break(&mut out);
            }
            emit_styled_line(&mut out, line, painter);
        }
    }
    // Trim trailing blank lines so we don't ship "\n\n\n…" at
    // the end of every document.
    while out.ends_with("\n\n\n") {
        out.pop();
    }
    Ok(out)
}

/// Emit one styled run. `line` is unstyled text; `painter`
/// holds bold/italic/heading flags.
fn emit_styled_line(
    out: &mut String,
    line: &str,
    painter: &rtf_parser_tt::parser::Painter,
) {
    let trimmed = line.trim_end_matches('\r');
    if trimmed.is_empty() {
        return;
    }
    // Heading detection: if the painter marks this block as a
    // heading (some Scrivener exports stamp paragraphs with
    // outline levels), emit `= …`. Fall back to plain text
    // for everything else.
    //
    // `Painter` exposes `bold` / `italic` / `underline`. We
    // honour bold + italic; underline isn't a first-class
    // Typst markup so we drop it.
    let bold = painter.bold;
    let italic = painter.italic;
    if bold {
        out.push_str("**");
    }
    if italic {
        out.push('_');
    }
    // Escape Typst-meta characters so user text doesn't
    // accidentally render as markup. The trio that matters in
    // prose is `*`, `_`, `#` at line start.
    for c in trimmed.chars() {
        match c {
            '*' | '_' | '#' | '@' | '<' | '>' | '$' | '\\' => {
                out.push('\\');
                out.push(c);
            }
            _ => out.push(c),
        }
    }
    if italic {
        out.push('_');
    }
    if bold {
        out.push_str("**");
    }
}

fn ensure_paragraph_break(out: &mut String) {
    while out.ends_with('\n') && out.len() > 0 {
        // peek the second-to-last char
        let last_two = out.chars().rev().take(2).collect::<String>();
        if last_two == "\n\n" {
            return;
        }
        if out.ends_with('\n') {
            out.push('\n');
            return;
        }
    }
    if !out.is_empty() && !out.ends_with('\n') {
        out.push('\n');
    }
    out.push('\n');
}

/// Brute-force fallback: drop everything that looks like an
/// RTF control sequence + bracket grouping, return what's
/// left. Used when the structured parser bails on a pathological
/// document; lossy but always produces something.
fn strip_to_plain_text(rtf_bytes: &[u8]) -> String {
    let s = String::from_utf8_lossy(rtf_bytes);
    let mut out = String::new();
    let mut chars = s.chars().peekable();
    let mut depth = 0usize;
    while let Some(c) = chars.next() {
        match c {
            '{' => depth += 1,
            '}' => depth = depth.saturating_sub(1),
            '\\' => {
                // Skip the control word. RTF control words are
                // letters + an optional numeric param; consume
                // until a delimiter (space, brace, backslash).
                while let Some(&next) = chars.peek() {
                    if next.is_ascii_alphanumeric() || next == '-' {
                        chars.next();
                    } else {
                        // Consume a single trailing space as
                        // the control-word terminator.
                        if next == ' ' {
                            chars.next();
                        }
                        break;
                    }
                }
            }
            '\n' | '\r' => {}
            _ if depth > 0 => out.push(c),
            _ => {}
        }
    }
    out.trim().to_string()
}

#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn empty_rtf() {
        let rtf = b"{\\rtf1\\ansi}";
        let out = rtf_to_typst(rtf).unwrap();
        assert!(out.trim().is_empty());
    }

    #[test]
    fn plain_paragraph() {
        let rtf = b"{\\rtf1\\ansi The quick brown fox.}";
        let out = rtf_to_typst(rtf).unwrap();
        assert!(out.contains("The quick brown fox"));
    }

    #[test]
    fn strip_fallback_handles_garbage() {
        // Pathological input — well-formed enough to not panic.
        let rtf = b"\\xxx{\\bogus garbage \\b text \\par more}";
        let out = strip_to_plain_text(rtf);
        assert!(out.contains("garbage") || out.contains("text"));
    }
}