devup-editor-html 1.0.11

//! HTML → [`CopiedBlocks`] parsing.
//!
//! Mirrors the behaviour of the React `htmlToBlocks` implementation
//! byte-for-byte on the inputs our editor and external apps (Word,
//! Notion, Google Docs) produce. The parser is tolerant: malformed
//! input is coerced through `html5ever` into a best-effort tree rather
//! than rejected.

use std::collections::HashMap;

use devup_editor_core::{Block, BlockId, IdGenerator, TextSpan, normalize_spans};
use html5ever::driver::{ParseOpts, parse_document};
use html5ever::tendril::TendrilSink;
use markup5ever_rcdom::{Handle, NodeData, RcDom};
use serde_json::{Map, Value};

use crate::clipboard::{CopiedBlocks, clean_html};

mod dom;
use dom::{
    MarkSet, attr_value, attrs_contains, build_synthetic_parent, clone_node_without_checkboxes,
    collect_inline_into, collect_raw_text, collect_table_rows, decode_props_from_element,
    detect_any_checkbox, detect_direct_checkbox, direct_children_of_tag_any, element_attrs,
    element_tag, extract_cell_props, extract_colgroup_widths, extract_row_props, extract_spans,
    extract_spans_from_li, find_body, find_descendant_with_any_class, has_class,
    has_descendant_with_class, is_all_whitespace, is_notion_v3_toggle, parse_inline_style,
    strip_nested_blocks,
};

// ── Public entry points ──────────────────────────────────────────

/// Parse an arbitrary HTML string (typically from a clipboard paste or
/// external rich-text export) into a flat [`Document`]. Prefer
/// [`html_to_copied_blocks`] for clipboard flows — it preserves the
/// table / toggle child structure that a Document's root-only view
/// drops.
pub(crate) fn parse_html(input: &str, id_gen: &mut dyn IdGenerator) -> CopiedBlocks {
    html_to_copied_blocks(input, id_gen)
}

/// Parse HTML into the clipboard-shaped [`CopiedBlocks`] subtree used
/// by the React paste flow. Block IDs come from `id_gen` so the caller
/// controls determinism.
pub fn html_to_copied_blocks(input: &str, id_gen: &mut dyn IdGenerator) -> CopiedBlocks {
    let cleaned = clean_html(input.trim());
    if cleaned.is_empty() {
        return CopiedBlocks {
            roots: Vec::new(),
            by_id: HashMap::new(),
        };
    }

    let dom = parse_document(RcDom::default(), ParseOpts::default()).one(cleaned);
    let body = find_body(&dom.document).unwrap_or_else(|| dom.document.clone());

    let mut ctx = Context::new(id_gen);
    ctx.process_children_with_indent(&body, 0);
    let roots = ctx.finalize_roots();
    CopiedBlocks {
        roots,
        by_id: ctx.by_id,
    }
}

// ── Parse context ────────────────────────────────────────────────

/// Mutable state shared by the recursive parser. Collects root blocks
/// (in document order) and every descendant block, keyed by id, so the
/// caller can rebuild the tree on paste without additional lookups.
struct Context<'a> {
    id_gen: &'a mut dyn IdGenerator,
    roots_order: Vec<BlockId>,
    by_id: HashMap<BlockId, Block>,
}

impl<'a> Context<'a> {
    fn new(id_gen: &'a mut dyn IdGenerator) -> Self {
        Self {
            id_gen,
            roots_order: Vec::new(),
            by_id: HashMap::new(),
        }
    }

    fn next_id(&mut self) -> BlockId {
        self.id_gen.next_id()
    }

    /// Insert a block into `by_id` and mark it as a root in document
    /// order.
    fn push_root(&mut self, block: Block) {
        self.roots_order.push(block.id.clone());
        self.by_id.insert(block.id.clone(), block);
    }

    /// Insert a descendant block (cell, row, toggle child…). Returns
    /// the id so the caller can reference it from its parent's
    /// `children` field.
    fn insert(&mut self, block: Block) -> BlockId {
        let id = block.id.clone();
        self.by_id.insert(id.clone(), block);
        id
    }

    fn finalize_roots(&mut self) -> Vec<Block> {
        let mut out = Vec::with_capacity(self.roots_order.len());
        for id in &self.roots_order {
            if let Some(b) = self.by_id.get(id) {
                out.push(b.clone());
            }
        }
        out
    }

    // ── Block-level recursion ────────────────────────────────────

    /// Walk `node`'s children, emitting root-level blocks and carrying
    /// `indent` through nested lists / toggles.
    fn process_children_with_indent(&mut self, node: &Handle, indent: i64) {
        // Buffer of inline DOM children that haven't yet hit a
        // block-level boundary — flushed as a `<p>` when we do.
        let mut inline_buf: Vec<Handle> = Vec::new();

        for child in node.children.borrow().iter() {
            match &child.data {
                NodeData::Text { contents } => {
                    if !contents.borrow().trim().is_empty() {
                        inline_buf.push(child.clone());
                    }
                    continue;
                }
                NodeData::Comment { .. } | NodeData::Doctype { .. } => continue,
                NodeData::Element { .. } => {}
                _ => continue,
            }

            let tag = element_tag(child).unwrap_or_default();

            if !BLOCK_TAGS.contains(&tag.as_str()) && tag != "details" {
                inline_buf.push(child.clone());
                continue;
            }

            self.flush_inline(&mut inline_buf, indent);

            if matches!(tag.as_str(), "h1" | "h2" | "h3" | "h4" | "h5" | "h6") {
                let level_digit = tag.chars().nth(1).and_then(|c| c.to_digit(10)).unwrap_or(1);
                let level = u64::from(level_digit);
                let spans = extract_spans(child);
                if !is_all_whitespace(&spans) {
                    let mut props = Map::new();
                    props.insert("level".into(), Value::from(level));
                    if indent > 0 {
                        props.insert("indent".into(), Value::from(indent));
                    }
                    let id = self.next_id();
                    let mut b = Block::with_props(id, "heading", props);
                    b.content = spans;
                    self.push_root(b);
                }
                continue;
            }

            if tag == "blockquote" {
                let spans = extract_spans(child);
                if !is_all_whitespace(&spans) {
                    let id = self.next_id();
                    let mut b = new_block(id, "quote", indent);
                    b.content = spans;
                    self.push_root(b);
                }
                continue;
            }

            if tag == "pre" {
                // Detect fenced code via inner `<code class="language-xxx">`.
                // Falls back to paragraph when no language class is present
                // (matches TS clipboard behaviour for generic `<pre>`).
                let code_child = child.children.borrow().iter().find_map(|c| {
                    if element_tag(c).as_deref() == Some("code") {
                        Some(c.clone())
                    } else {
                        None
                    }
                });
                let language = code_child
                    .as_ref()
                    .and_then(|code| attr_value(code, "class"))
                    .and_then(|cls| {
                        cls.split_whitespace()
                            .find_map(|c| c.strip_prefix("language-").map(String::from))
                    });
                let text = match code_child.as_ref() {
                    Some(code) => collect_raw_text(code),
                    None => collect_raw_text(child),
                };
                let text = text.strip_prefix('\n').unwrap_or(&text).to_string();
                if !text.is_empty() {
                    let id = self.next_id();
                    let ty = if language.is_some() {
                        "code"
                    } else {
                        "paragraph"
                    };
                    let mut props = Map::new();
                    if indent > 0 {
                        props.insert("indent".into(), Value::from(indent));
                    }
                    if let Some(lang) = language {
                        props.insert("language".into(), Value::String(lang));
                    }
                    let mut b = if props.is_empty() {
                        Block::new(id, ty)
                    } else {
                        Block::with_props(id, ty, props)
                    };
                    b.content = vec![TextSpan::plain(text)];
                    self.push_root(b);
                }
                continue;
            }

            if tag == "details" {
                self.process_toggle_details(child, indent);
                continue;
            }

            if tag == "ul" || tag == "ol" {
                let attrs = element_attrs(child);
                // data-type="todo" list marker
                if attrs_contains(&attrs, "data-devup-type", "todo")
                    || has_class(&attrs, "to-do-list")
                {
                    self.process_notion_todo_list(child);
                    continue;
                }
                if tag == "ul" && has_class(&attrs, "toggle") {
                    self.process_notion_toggle_list(child, indent);
                    continue;
                }
                self.process_list(child, tag == "ol", indent);
                continue;
            }

            if tag == "table" {
                self.process_table(child, indent);
                continue;
            }

            if tag == "hr" {
                let id = self.next_id();
                self.push_root(new_block(id, "divider", indent));
                continue;
            }

            if tag == "li" {
                // Bare <li> outside a list — salvage as paragraph so
                // text survives.
                let spans = extract_spans(child);
                if !is_all_whitespace(&spans) {
                    let id = self.next_id();
                    let mut b = new_block(id, "paragraph", indent);
                    b.content = spans;
                    self.push_root(b);
                }
                continue;
            }

            if tag == "p" {
                // Detect the `<p data-type="todo" data-checked="…">`
                // serialisation emitted by `write_block_html`. Parsed as
                // a todo block so copy→paste within devup survives.
                let p_attrs = element_attrs(child);
                if attrs_contains(&p_attrs, "data-type", "todo") {
                    let checked = attrs_contains(&p_attrs, "data-checked", "true");
                    let spans = extract_spans(child);
                    if !is_all_whitespace(&spans) {
                        let id = self.next_id();
                        let mut props = Map::new();
                        props.insert("checked".into(), Value::Bool(checked));
                        if indent > 0 {
                            props.insert("indent".into(), Value::from(indent));
                        }
                        let mut b = Block::with_props(id, "todo", props);
                        b.content = spans;
                        self.push_root(b);
                    }
                    continue;
                }
                let spans = extract_spans(child);
                if !is_all_whitespace(&spans) {
                    let id = self.next_id();
                    let mut b = new_block(id, "paragraph", indent);
                    b.content = spans;
                    self.push_root(b);
                }
                continue;
            }

            // Generic block wrapper (div/section/article/main/header/
            // footer/nav/aside/figure…) — descend if it has block
            // children, otherwise flatten to a paragraph.
            let has_block_child = child.children.borrow().iter().any(|c| {
                if let Some(t) = element_tag(c) {
                    BLOCK_TAGS.contains(&t.as_str()) || t == "details"
                } else {
                    false
                }
            });
            if has_block_child {
                self.process_children_with_indent(child, indent);
            } else {
                let spans = extract_spans(child);
                if !is_all_whitespace(&spans) {
                    let id = self.next_id();
                    let mut b = new_block(id, "paragraph", indent);
                    b.content = spans;
                    self.push_root(b);
                }
            }
        }

        self.flush_inline(&mut inline_buf, indent);
    }

    fn flush_inline(&mut self, buf: &mut Vec<Handle>, indent: i64) {
        if buf.is_empty() {
            return;
        }
        let mut spans: Vec<TextSpan> = Vec::new();
        for n in buf.iter() {
            let mark_set = MarkSet::empty();
            collect_inline_into(n, &mut spans, &mark_set);
        }
        buf.clear();
        normalize_spans(&mut spans);
        if is_all_whitespace(&spans) {
            return;
        }
        let id = self.next_id();
        let mut b = new_block(id, "paragraph", indent);
        b.content = spans;
        self.push_root(b);
    }

    // ── List / toggle / todo handlers ────────────────────────────

    fn process_toggle_details(&mut self, details: &Handle, indent: i64) {
        // Summary inline spans → toggle title.
        let summary_node = details.children.borrow().iter().find_map(|c| {
            if element_tag(c).as_deref() == Some("summary") {
                Some(c.clone())
            } else {
                None
            }
        });
        let title_spans = summary_node.as_ref().map(extract_spans).unwrap_or_default();

        let id = self.next_id();
        let mut props = Map::new();
        props.insert("collapsed".into(), Value::Bool(false));
        if indent > 0 {
            props.insert("indent".into(), Value::from(indent));
        }
        let mut toggle = Block::with_props(id, "toggle", props);
        toggle.content = title_spans;
        self.push_root(toggle);

        // Everything after <summary> becomes child blocks at indent+1.
        // We temporarily rehome them onto a synthetic container by
        // iterating the children list directly.
        let child_handle = details.clone();
        let original_children: Vec<Handle> = child_handle
            .children
            .borrow()
            .iter()
            .filter(|c| element_tag(c).as_deref() != Some("summary"))
            .cloned()
            .collect();
        self.process_handles_with_indent(&original_children, indent + 1);
    }

    fn process_notion_toggle_list(&mut self, ul: &Handle, indent: i64) {
        for li in ul.children.borrow().iter() {
            if element_tag(li).as_deref() != Some("li") {
                continue;
            }
            let details = li.children.borrow().iter().find_map(|c| {
                if element_tag(c).as_deref() == Some("details") {
                    Some(c.clone())
                } else {
                    None
                }
            });
            if let Some(det) = details {
                self.process_toggle_details(&det, indent);
            } else {
                let spans = extract_spans_from_li(li);
                if !is_all_whitespace(&spans) {
                    let id = self.next_id();
                    let mut props = Map::new();
                    props.insert("style".into(), Value::String("unordered".into()));
                    if indent > 0 {
                        props.insert("indent".into(), Value::from(indent));
                    }
                    let mut b = Block::with_props(id, "list", props);
                    b.content = spans;
                    self.push_root(b);
                }
            }
        }
    }

    fn process_list(&mut self, list_el: &Handle, ordered: bool, indent: i64) {
        let style = if ordered { "ordered" } else { "unordered" };
        for li in list_el.children.borrow().iter() {
            if element_tag(li).as_deref() != Some("li") {
                continue;
            }

            // Generic checkbox heuristic: `<li>[<div>]<input type=checkbox>…</li>`
            if let Some(checked) = detect_direct_checkbox(li) {
                let clone_without_cb = clone_node_without_checkboxes(li);
                let spans = extract_spans(&clone_without_cb);
                if !is_all_whitespace(&spans) {
                    let id = self.next_id();
                    let mut props = Map::new();
                    props.insert("checked".into(), Value::Bool(checked));
                    if indent > 0 {
                        props.insert("indent".into(), Value::from(indent));
                    }
                    let mut b = Block::with_props(id, "todo", props);
                    b.content = spans;
                    self.push_root(b);
                }
                // Recurse nested lists regardless.
                self.recurse_nested_lists(li, indent);
                continue;
            }

            // Notion v3 toggle heuristic
            if !ordered && is_notion_v3_toggle(li) {
                let block_children: Vec<Handle> = li
                    .children
                    .borrow()
                    .iter()
                    .filter(|c| {
                        if let Some(t) = element_tag(c) {
                            matches!(
                                t.as_str(),
                                "p" | "div"
                                    | "ul"
                                    | "ol"
                                    | "blockquote"
                                    | "pre"
                                    | "table"
                                    | "details"
                            ) || is_heading_tag(&t)
                        } else {
                            false
                        }
                    })
                    .cloned()
                    .collect();
                let title_el = &block_children[0];
                let title_spans = extract_spans(title_el);
                let id = self.next_id();
                let mut props = Map::new();
                props.insert("collapsed".into(), Value::Bool(false));
                if indent > 0 {
                    props.insert("indent".into(), Value::from(indent));
                }
                let mut toggle = Block::with_props(id, "toggle", props);
                toggle.content = title_spans;
                self.push_root(toggle);
                self.process_handles_with_indent(&block_children[1..], indent + 1);
                continue;
            }

            // Plain list item
            let spans = extract_spans_from_li(li);
            if !is_all_whitespace(&spans) {
                let id = self.next_id();
                let mut props = Map::new();
                props.insert("style".into(), Value::String(style.into()));
                if indent > 0 {
                    props.insert("indent".into(), Value::from(indent));
                }
                let mut b = Block::with_props(id, "list", props);
                b.content = spans;
                self.push_root(b);
            }

            self.recurse_nested_lists(li, indent);
        }
    }

    fn recurse_nested_lists(&mut self, li: &Handle, indent: i64) {
        for nested in li.children.borrow().iter() {
            let Some(t) = element_tag(nested) else {
                continue;
            };
            if t == "ul" {
                let attrs = element_attrs(nested);
                if has_class(&attrs, "toggle") {
                    self.process_notion_toggle_list(nested, indent + 1);
                } else {
                    self.process_list(nested, false, indent + 1);
                }
            } else if t == "ol" {
                self.process_list(nested, true, indent + 1);
            } else if t == "details" {
                self.process_toggle_details(nested, indent + 1);
            }
        }
    }

    fn process_notion_todo_list(&mut self, ul: &Handle) {
        for li in ul.children.borrow().iter() {
            if element_tag(li).as_deref() != Some("li") {
                continue;
            }

            // `<ul data-devup-type="todo">` items often carry a label +
            // checkbox inside — detect and reuse the direct-checkbox
            // path.
            let attrs = element_attrs(li);
            let marker_checked = attrs
                .iter()
                .find(|a| a.name.local.as_ref() == "data-checked")
                .map(|a| a.value.as_ref().eq_ignore_ascii_case("true"));
            let checkbox = detect_any_checkbox(li);
            let notion_checked = has_descendant_with_class(li, "checkbox-on");

            let checked = marker_checked
                .or(checkbox)
                .or(Some(notion_checked))
                .unwrap_or(false);

            // Text extraction: prefer Notion's wrappers when present,
            // fall back to the whole <li> minus checkboxes + nested
            // lists.
            let notion_wrapper = find_descendant_with_any_class(
                li,
                &["to-do-children-checked", "to-do-children-unchecked"],
            );
            let spans = if let Some(w) = notion_wrapper {
                extract_spans(&w)
            } else {
                let clone = clone_node_without_checkboxes(li);
                let clone = strip_nested_blocks(&clone);
                extract_spans(&clone)
            };
            if !is_all_whitespace(&spans) {
                let id = self.next_id();
                let mut props = Map::new();
                props.insert("checked".into(), Value::Bool(checked));
                let mut b = Block::with_props(id, "todo", props);
                b.content = spans;
                self.push_root(b);
            }
        }
    }

    // ── Toggle children (from either <details> or v3 heuristic) ──

    fn process_handles_with_indent(&mut self, handles: &[Handle], indent: i64) {
        let synthetic = build_synthetic_parent(handles);
        self.process_children_with_indent(&synthetic, indent);
    }

    // ── Table handler ────────────────────────────────────────────

    fn process_table(&mut self, table_el: &Handle, indent: i64) {
        // Flatten rows from tbody / thead / tfoot / bare <tr>.
        let row_els: Vec<Handle> = collect_table_rows(table_el);
        if row_els.is_empty() {
            return;
        }
        let max_cols = row_els
            .iter()
            .map(|r| direct_children_of_tag_any(r, &["td", "th"]).len())
            .max()
            .unwrap_or(0);
        if max_cols == 0 {
            return;
        }

        // Build row & cell descendants first.
        let mut row_ids: Vec<BlockId> = Vec::new();
        for tr in &row_els {
            let cells = direct_children_of_tag_any(tr, &["td", "th"]);
            let mut cell_ids: Vec<BlockId> = Vec::with_capacity(max_cols);
            for c in 0..max_cols {
                let spans = cells.get(c).map(extract_spans).unwrap_or_default();
                let mut props = cells
                    .get(c)
                    .and_then(extract_cell_props)
                    .unwrap_or_default();
                // Remove colspan/rowspan redundancy if cell only has
                // them and no other props? Keep everything — mirror TS.
                let cell_id = self.next_id();
                let mut cell_block = if props.is_empty() {
                    Block::new(cell_id.clone(), "table_cell")
                } else {
                    // Normalize colspan/rowspan number types to u64.
                    normalize_span_numbers(&mut props);
                    Block::with_props(cell_id.clone(), "table_cell", props)
                };
                cell_block.content = spans;
                cell_ids.push(self.insert(cell_block));
            }
            let row_props = extract_row_props(tr).unwrap_or_default();
            let row_id = self.next_id();
            let mut row_block = if row_props.is_empty() {
                Block::new(row_id.clone(), "table_row")
            } else {
                Block::with_props(row_id.clone(), "table_row", row_props)
            };
            row_block.children.clone_from(&cell_ids);
            // Backfill `parent`
            for cid in &cell_ids {
                if let Some(c) = self.by_id.get_mut(cid) {
                    c.parent = Some(row_id.clone());
                }
            }
            row_ids.push(self.insert(row_block));
        }

        // Table props
        let mut table_props = decode_props_from_element(table_el).unwrap_or_default();
        if let Some(style_attr) = attr_value(table_el, "style") {
            let decl = parse_inline_style(&style_attr);
            if !table_props.contains_key("backgroundColor")
                && let Some(v) = decl.background_color
            {
                table_props.insert("backgroundColor".into(), Value::String(v));
            }
            if !table_props.contains_key("borderColor")
                && let Some(v) = decl.border_color
            {
                table_props.insert("borderColor".into(), Value::String(v));
            }
            if !table_props.contains_key("borderWidth")
                && let Some(v) = decl.border_width
            {
                table_props.insert("borderWidth".into(), Value::String(v));
            }
            if !table_props.contains_key("borderStyle")
                && let Some(v) = decl.border_style
            {
                table_props.insert("borderStyle".into(), Value::String(v));
            }
            if !table_props.contains_key("verticalAlign")
                && let Some(v) = decl.vertical_align
            {
                table_props.insert("verticalAlign".into(), Value::String(v));
            }
            if !table_props.contains_key("padding")
                && let Some(v) = decl.padding
            {
                table_props.insert("padding".into(), Value::String(v));
            }
        }
        // Columns
        if !table_props.contains_key("columns") {
            let widths = extract_colgroup_widths(table_el, max_cols);
            let cols: Vec<Value> = match widths {
                Some(ws) => ws
                    .into_iter()
                    .map(|w| {
                        let mut m = Map::new();
                        m.insert("width".into(), Value::from(w));
                        Value::Object(m)
                    })
                    .collect(),
                None => (0..max_cols)
                    .map(|_| {
                        let mut m = Map::new();
                        m.insert("width".into(), Value::from(120u64));
                        Value::Object(m)
                    })
                    .collect(),
            };
            table_props.insert("columns".into(), Value::Array(cols));
        }
        if indent > 0 {
            table_props.insert("indent".into(), Value::from(indent));
        }
        let table_id = self.next_id();
        let mut table_block = Block::with_props(table_id.clone(), "table", table_props);
        table_block.children.clone_from(&row_ids);
        // Backfill parent on rows
        for rid in &row_ids {
            if let Some(r) = self.by_id.get_mut(rid) {
                r.parent = Some(table_id.clone());
            }
        }
        self.push_root(table_block);
    }
}

fn normalize_span_numbers(props: &mut Map<String, Value>) {
    for key in ["colspan", "rowspan"] {
        if let Some(v) = props.get(key) {
            let n = match v {
                Value::Number(n) => n.as_u64(),
                Value::String(s) => s.parse::<u64>().ok(),
                _ => None,
            };
            if let Some(n) = n {
                props.insert(key.into(), Value::from(n));
            }
        }
    }
}

fn new_block(id: BlockId, ty: &str, indent: i64) -> Block {
    if indent > 0 {
        let mut props = Map::new();
        props.insert("indent".into(), Value::from(indent));
        Block::with_props(id, ty, props)
    } else {
        Block::new(id, ty)
    }
}

// ── Helpers: tag recognition ──────────────────────────────────────

pub(super) fn is_heading_tag(tag: &str) -> bool {
    matches!(tag, "h1" | "h2" | "h3" | "h4" | "h5" | "h6")
}

static BLOCK_TAGS: &[&str] = &[
    "address",
    "article",
    "aside",
    "blockquote",
    "div",
    "dd",
    "dl",
    "dt",
    "figcaption",
    "figure",
    "footer",
    "h1",
    "h2",
    "h3",
    "h4",
    "h5",
    "h6",
    "header",
    "hr",
    "li",
    "main",
    "nav",
    "ol",
    "p",
    "pre",
    "section",
    "table",
    "tbody",
    "td",
    "th",
    "thead",
    "tfoot",
    "tr",
    "ul",
];

pub(super) static TABLE_STRUCTURE_TAGS: &[&str] = &[
    "table", "thead", "tbody", "tfoot", "tr", "td", "th", "col", "colgroup", "caption",
];