bezel-markdown 0.2.24

//! The document model.
//!
//! A [`Doc`] is a **flat** list of [`Block`]s with an indent level, not a
//! nested tree. That is Notion's model rather than CommonMark's, and it is the
//! decision the rest of this crate hangs off: editing a flat list means Enter
//! splits, Backspace merges, and Tab indents — all list operations. On a
//! nested tree "the previous block" is a traversal and every edit is a
//! restructure.
//!
//! The trade is that arbitrarily nested CommonMark does not survive a round
//! trip: a quote inside a quote flattens to one. Notion has the same
//! limitation. What is guaranteed is [`crate::serialize`]'s fixed point —
//! parse, serialize, parse again, and the document is unchanged — so an
//! edit/save cycle never drifts.

use std::{
    collections::{HashMap, HashSet},
    ops::Range,
};

/// A markdown document: blocks in document order.
#[derive(Debug, Clone, Default, PartialEq, Eq)]
pub struct Doc {
    pub blocks: Vec<Block>,
}

impl Doc {
    /// Make each run of ordered items consecutive.
    ///
    /// Markdown honours only the *first* number in a list — `1.` followed by `9.`
    /// renders as 1, 2. Two source lists that used different delimiters (`1.` then
    /// `9)`) are separate lists, but a flat document has no list identity to
    /// preserve, so they serialize as one and the second item's number would move
    /// on the next read. Deciding it here means the document already holds what the
    /// next parse would produce.
    pub(crate) fn renumber(&mut self) {
        // The number owed to the next ordered item at each indent level. A run
        // survives blocks nested under it and ends at anything else.
        let mut expected: Vec<Option<u64>> = Vec::new();
        let mut runs = Runs::default();
        let mut above: Option<Run> = None;
        for block in &mut self.blocks {
            let indent = block.indent as usize;
            let run = runs.next(block.quote, block.indent);
            // A list on the other side of a quote's `>` is another list.
            if run.is_none_or(|run| run.first) {
                let edge = [above, run]
                    .into_iter()
                    .flatten()
                    .map(|run| run.level as usize)
                    .min();
                if let Some(edge) = edge {
                    expected.truncate(edge);
                }
            }
            above = run;
            expected.truncate(indent + 1);
            expected.resize(indent + 1, None);

            if let BlockKind::Ordered { number, .. } = &mut block.kind {
                if let Some(next) = expected[indent] {
                    *number = next;
                }
                expected[indent] = Some(number.saturating_add(1));
            } else {
                expected[indent] = None;
            }
        }
    }

    /// Each heading's anchor and its block, in document order. A slug already
    /// taken by an earlier heading gets `-1`, `-2`, … as on GitHub.
    pub fn anchors(&self) -> Vec<(String, usize)> {
        let mut taken = HashSet::new();
        // The first suffix not yet tried for each slug, so a title repeated n
        // times costs n probes rather than n².
        let mut next: HashMap<String, usize> = HashMap::new();
        self.blocks
            .iter()
            .enumerate()
            .filter_map(|(ix, block)| match &block.kind {
                BlockKind::Heading { text, .. } => Some((slug(&text.text), ix)),
                _ => None,
            })
            .map(|(base, ix)| {
                let from = next.get(&base).copied().unwrap_or(0);
                let (n, anchor) = (from..)
                    .map(|n| match n {
                        0 => (n, base.clone()),
                        n => (n, format!("{base}-{n}")),
                    })
                    .find(|(_, anchor)| !taken.contains(anchor))
                    .expect("an unbounded run of suffixes");
                next.insert(base, n + 1);
                taken.insert(anchor.clone());
                (anchor, ix)
            })
            .collect()
    }
}

/// The anchor GitHub gives a heading: lowercased, punctuation dropped but for
/// `-` and `_`, each space a `-`.
pub fn slug(text: &str) -> String {
    text.chars()
        .flat_map(char::to_lowercase)
        .filter_map(|c| match c {
            ' ' => Some('-'),
            '-' | '_' => Some(c),
            c if c.is_alphanumeric() => Some(c),
            _ => None,
        })
        .collect()
}

/// One block, and how deeply it is nested.
///
/// `indent` obeys one invariant, established by the parser and relied on by
/// the serializer: the first block is at 0, and no block is more than one
/// level deeper than the block before it. A document that satisfies it always
/// serializes to markdown that parses back to the same indents.
#[derive(Debug, Clone, PartialEq, Eq, Hash)]
pub struct Block {
    pub kind: BlockKind,
    pub indent: u8,
    /// The blockquote the block sits in, or `None` outside one. Adjacent
    /// quoted blocks form one blockquote — see [`quote_runs`].
    pub quote: Option<Quoted>,
}

/// A block's place in a blockquote.
#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash, Default)]
pub struct Quoted {
    /// The GFM alert the blockquote opens with — see [`QuoteKind`].
    pub alert: Option<QuoteKind>,
}

/// Where a quoted block falls in its blockquote.
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub(crate) struct Run {
    /// The indent the `>` is written at: the indent of the run's first block.
    pub level: u8,
    pub first: bool,
}

/// Each block's place in a blockquote, or `None` for an unquoted block.
///
/// A quoted block continues the run before it when it carries the same
/// [`Quoted`] and sits no shallower than the run's level. Anything else starts
/// a new run at its own indent.
pub(crate) fn quote_runs(
    blocks: impl IntoIterator<Item = (Option<Quoted>, u8)>,
) -> Vec<Option<Run>> {
    let mut runs = Runs::default();
    blocks
        .into_iter()
        .map(|(quote, indent)| runs.next(quote, indent))
        .collect()
}

/// [`quote_runs`] one block at a time.
#[derive(Default)]
pub(crate) struct Runs {
    open: Option<(Quoted, u8)>,
}

impl Runs {
    pub fn next(&mut self, quote: Option<Quoted>, indent: u8) -> Option<Run> {
        let Some(quote) = quote else {
            self.open = None;
            return None;
        };
        match self.open {
            Some((current, level)) if current == quote && indent >= level => Some(Run {
                level,
                first: false,
            }),
            _ => {
                self.open = Some((quote, indent));
                Some(Run {
                    level: indent,
                    first: true,
                })
            }
        }
    }
}

impl From<BlockKind> for Block {
    fn from(kind: BlockKind) -> Self {
        Self::new(kind)
    }
}

impl Block {
    pub fn new(kind: BlockKind) -> Self {
        Self::at(kind, 0)
    }

    pub fn at(kind: BlockKind, indent: u8) -> Self {
        Self {
            kind,
            indent,
            quote: None,
        }
    }

    pub fn quoted(self, quote: Quoted) -> Self {
        Self {
            quote: Some(quote),
            ..self
        }
    }

    /// One of the block's editable texts. `None` when the block has no such
    /// part — every block is atomic to some [`Part`], and bookmarks and rules
    /// are atomic to all of them.
    pub fn text_at(&self, part: Part) -> Option<&Text> {
        match (&self.kind, part) {
            (
                BlockKind::Paragraph(text)
                | BlockKind::Heading { text, .. }
                | BlockKind::Bullet(text)
                | BlockKind::Ordered { text, .. }
                | BlockKind::Task { text, .. },
                Part::Body,
            ) => Some(text),
            (BlockKind::Code { code, .. }, Part::Code) => Some(code),
            (BlockKind::Image { alt, .. }, Part::Caption) => Some(alt),
            (BlockKind::Table { header, .. }, Part::Cell { row: 0, column }) => header.get(column),
            (BlockKind::Table { rows, .. }, Part::Cell { row, column }) => {
                rows.get(row - 1)?.get(column)
            }
            _ => None,
        }
    }

    pub fn text_at_mut(&mut self, part: Part) -> Option<&mut Text> {
        match (&mut self.kind, part) {
            (
                BlockKind::Paragraph(text)
                | BlockKind::Heading { text, .. }
                | BlockKind::Bullet(text)
                | BlockKind::Ordered { text, .. }
                | BlockKind::Task { text, .. },
                Part::Body,
            ) => Some(text),
            (BlockKind::Code { code, .. }, Part::Code) => Some(code),
            (BlockKind::Image { alt, .. }, Part::Caption) => Some(alt),
            (BlockKind::Table { header, .. }, Part::Cell { row: 0, column }) => {
                header.get_mut(column)
            }
            (BlockKind::Table { rows, .. }, Part::Cell { row, column }) => {
                rows.get_mut(row - 1)?.get_mut(column)
            }
            _ => None,
        }
    }

    /// Every part a caret can sit in, in document order.
    pub fn parts(&self) -> Vec<Part> {
        match &self.kind {
            BlockKind::Paragraph(_)
            | BlockKind::Heading { .. }
            | BlockKind::Bullet(_)
            | BlockKind::Ordered { .. }
            | BlockKind::Task { .. } => vec![Part::Body],
            BlockKind::Code { .. } => vec![Part::Code],
            BlockKind::Image { .. } => vec![Part::Caption],
            BlockKind::Table { header, rows, .. } => {
                let mut parts = Vec::new();
                if !header.is_empty() {
                    parts.extend((0..header.len()).map(|column| Part::Cell { row: 0, column }));
                }
                for (ix, row) in rows.iter().enumerate() {
                    parts.extend((0..row.len()).map(|column| Part::Cell {
                        row: ix + 1,
                        column,
                    }));
                }
                parts
            }
            BlockKind::Bookmark { .. } | BlockKind::Rule => Vec::new(),
        }
    }

    /// Whether the block is a blank paragraph: the empty line an editor leaves
    /// behind, which markdown has no way to write down. The empty body of an
    /// alert is not one: `> [!TIP]` writes down and reads back.
    pub(crate) fn blank(&self) -> bool {
        matches!(&self.kind, BlockKind::Paragraph(text) if text.is_empty() && text.marks.is_empty())
            && !self.quote.is_some_and(|quote| quote.alert.is_some())
    }

    /// Whether what the block paints past its parts holds no caret — a picture,
    /// a card, a line. What a selection has to wash for itself, since there is
    /// no text under it to carry the highlight.
    pub fn opaque(&self) -> bool {
        matches!(
            self.kind,
            BlockKind::Image { .. } | BlockKind::Bookmark { .. } | BlockKind::Rule
        )
    }
}

/// Which of a block's texts a caret sits in.
///
/// A block has one kind of part and never a mix — prose blocks have a body, a
/// code block has its code, a table has cells — so this is a coordinate rather
/// than a path, and the model stays flat. The ordering is document order, which
/// is what makes a [`crate::Cursor`] comparable and therefore what makes a
/// selection a range.
#[derive(Debug, Clone, Copy, PartialEq, Eq, PartialOrd, Ord, Default)]
pub enum Part {
    #[default]
    Body,
    Code,
    /// An image's caption, which is also its alt text.
    Caption,
    /// Row 0 is the header row; row `n` is `rows[n - 1]`.
    Cell {
        row: usize,
        column: usize,
    },
}

/// The block vocabulary. Closed by design — a consumer that needs a block of
/// its own is a reason to widen this enum rather than to grow an extension
/// system.
#[derive(Debug, Clone, PartialEq, Eq, Hash)]
pub enum BlockKind {
    Paragraph(Text),
    Heading {
        /// 1–6.
        level: u8,
        text: Text,
    },
    Bullet(Text),
    Ordered {
        /// The rendered number. Stored rather than derived so a list starting
        /// at 3 survives the round trip.
        number: u64,
        text: Text,
    },
    Task {
        checked: bool,
        text: Text,
    },
    /// The code carries a [`Text`] like every other editable region, so one
    /// accessor and one edit path cover the whole document. Its marks are
    /// unreachable rather than forbidden: nothing that writes here creates one.
    Code {
        language: Option<String>,
        code: Text,
        /// The height a painted fence stands at, in whole pixels, from a drag
        /// off its handle in `bezel-editor`. `None` is the block's own height.
        /// Spelled as the info string's second word: ` ```mermaid 360 `.
        height: Option<u32>,
    },
    /// The caption is the alt text — markdown has one slot, and a reader that
    /// cannot see the picture reads the same words. Like [`BlockKind::Code`]'s,
    /// its marks are unreachable rather than forbidden.
    Image {
        url: String,
        alt: Text,
        /// A drag off the handle in `bezel-editor`, in whole pixels — `None` is
        /// the natural width. `u32` rather than a float: this derives `Eq`, and
        /// a `f32` cannot because of NaN. Spelled `![alt|480](url)`, which
        /// leaves the title slot to say what a title says.
        width: Option<u32>,
    },
    /// A link with a block to itself, painted richly. Atomic on purpose:
    /// everything it shows past the URL comes from [`crate::app::preview`], so there
    /// is nothing here for a caret to edit.
    ///
    /// [`Form`] picks a card or a card with its picture across the width;
    /// [`Form::Chip`] paints as the card. A chip is a [`Mark::Mention`], on a
    /// line of its own or not.
    Bookmark {
        url: String,
        form: Form,
    },
    Table {
        align: Vec<Align>,
        header: Vec<Text>,
        rows: Vec<Vec<Text>>,
    },
    Rule,
}

/// A GFM alert's kind — the `[!NOTE]` marker a blockquote opens with.
///
/// Only recognised when the marker is alone on the quote's first line and
/// names one of these five; anything else stays the text it was written as.
#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
pub enum QuoteKind {
    Note,
    Tip,
    Important,
    Warning,
    Caution,
}

impl QuoteKind {
    /// The marker line, bracket to bracket.
    pub fn marker(self) -> &'static str {
        match self {
            Self::Note => "[!NOTE]",
            Self::Tip => "[!TIP]",
            Self::Important => "[!IMPORTANT]",
            Self::Warning => "[!WARNING]",
            Self::Caution => "[!CAUTION]",
        }
    }

    /// What the alert calls itself where it is painted.
    pub fn label(self) -> &'static str {
        match self {
            Self::Note => "Note",
            Self::Tip => "Tip",
            Self::Important => "Important",
            Self::Warning => "Warning",
            Self::Caution => "Caution",
        }
    }
}

/// GFM column alignment.
#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash, Default)]
pub enum Align {
    #[default]
    Left,
    Center,
    Right,
}

/// Inline content: a string, plus marks over byte ranges of it.
///
/// Marks are a separate list rather than flags on a run because an editor has
/// to *map* them through insertions and deletions, and because run flags lose
/// nesting order — under flags `**_x_**` and `_**x**_` are the same value.
/// Here they differ by the order of the two spans, and both survive a round
/// trip.
///
/// A newline in `text` is a line break within the block (markdown's soft or
/// hard break, which this model does not distinguish — neither does Notion).
/// Whether it paints as a break or a space is a rendering decision.
#[derive(Debug, Clone, Default, PartialEq, Eq, Hash)]
pub struct Text {
    pub text: String,
    /// Outermost first. Ranges may overlap and may be identical.
    pub marks: Vec<MarkSpan>,
}

impl From<&str> for Text {
    fn from(text: &str) -> Self {
        Self::plain(text)
    }
}

impl From<String> for Text {
    fn from(text: String) -> Self {
        Self::plain(text)
    }
}

impl Text {
    /// Unmarked text.
    pub fn plain(text: impl Into<String>) -> Self {
        Self {
            text: text.into(),
            marks: Vec::new(),
        }
    }

    /// A URL that links to itself — what a pasted link is, and what a bookmark
    /// hands back when it turns into prose.
    pub fn link(url: &str) -> Self {
        Self {
            text: url.to_string(),
            marks: vec![MarkSpan {
                range: 0..url.len(),
                mark: Mark::Link(url.to_string()),
            }],
        }
    }

    pub fn is_empty(&self) -> bool {
        self.text.is_empty()
    }

    /// The mention `offset` falls strictly inside. A caret has no place there:
    /// the mention shows its preview, not its text.
    pub(crate) fn mention_around(&self, offset: usize) -> Option<Range<usize>> {
        self.marks
            .iter()
            .find(|span| {
                matches!(span.mark, Mark::Mention { .. })
                    && span.range.start < offset
                    && offset < span.range.end
            })
            .map(|span| span.range.clone())
    }

    /// Whether no other mark overlaps the one at `ix`.
    ///
    /// A mark written whole — a code span, a mention — leaves no room inside
    /// itself for another mark's boundary, so this is what decides whether it
    /// can be spelled its own way at all.
    pub(crate) fn alone(&self, ix: usize) -> bool {
        let span = &self.marks[ix].range;
        self.marks.iter().enumerate().all(|(other, mark)| {
            other == ix || mark.range.end <= span.start || mark.range.start >= span.end
        })
    }
}

#[derive(Debug, Clone, PartialEq, Eq, Hash)]
pub struct MarkSpan {
    pub range: Range<usize>,
    pub mark: Mark,
}

/// How a [`Mark::Mention`] was written down, which is also how it paints.
///
/// `Auto` is the shorthand `<https://x>`: a chip in a sentence, a card on a
/// line of its own. It is a variant rather than a resolved form because the
/// spelling is what has to survive the round trip — resolve it at parse and
/// every `<url>` grows brackets the first time the file is saved.
///
/// The other two are CommonMark's title slot, `[url](url "chip")`, which is
/// core, ignored by every other renderer, and the only place left to say what
/// the shorthand cannot: a chip alone on a line, and the bigger card.
/// [`Form::Chip`] stays inline wherever it is.
#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
pub enum Form {
    Auto,
    Chip,
    /// The bigger card, at a height in whole pixels from a drag off its
    /// handle in `bezel-editor`, or its own height for `None`. Spelled
    /// `"embed 360"`.
    Embed(Option<u32>),
}

impl Form {
    /// The title that spells this form, and `None` for the shorthand.
    pub(crate) fn title(self) -> Option<String> {
        match self {
            Self::Auto => None,
            Self::Chip => Some("chip".to_owned()),
            Self::Embed(None) => Some("embed".to_owned()),
            Self::Embed(Some(height)) => Some(format!("embed {height}")),
        }
    }

    pub(crate) fn from_title(title: &str) -> Option<Self> {
        let (name, height) = crate::parse::split_size(title);
        match (name, height) {
            ("chip", None) => Some(Self::Chip),
            ("embed", height) => Some(Self::Embed(height)),
            _ => None,
        }
    }
}

#[derive(Debug, Clone, PartialEq, Eq, Hash)]
pub enum Mark {
    Bold,
    Italic,
    Strike,
    Code,
    Link(String),
    /// A link painted richly rather than as underlined text — a chip inline, a
    /// [`BlockKind::Bookmark`] with a block to itself.
    ///
    /// The chip shows a favicon and the [`crate::app::preview`] title in place of
    /// its text, and opens the card on hover. Its text is atomic: a caret sits
    /// at either end, never inside, and typing at its end falls outside it.
    Mention {
        url: String,
        form: Form,
    },
    /// A mark the app spells itself — underline, a highlight, a colour. The
    /// name is [`crate::Marks`]'s, and the delimiter that writes it comes from
    /// the same registry.
    Custom(String),
    /// An image among text. [`BlockKind::Image`] is the shape an editor offers;
    /// this is what keeps `see ![x](u) here` from silently becoming a link when
    /// the document is saved. No width: one among text has no box of its own to
    /// resize, and the `|480` that would say so stays the ordinary text it is.
    Image(String),
}