devup-editor-html 1.0.4

//! DOM query / walk / filter helpers used by the HTML importer.
//!
//! Split out of `import.rs` so no file exceeds 1000 lines. Callers in
//! `super::mod` reach these via `use super::dom::*;`.

use devup_editor_core::{Mark, TextSpan, normalize_spans};
use markup5ever::interface::Attribute;
use markup5ever_rcdom::{Handle, NodeData};
use serde_json::{Map, Value};

use super::{TABLE_STRUCTURE_TAGS, is_heading_tag};
use crate::clipboard::decode_props;

// ── Helpers: DOM queries ──────────────────────────────────────────

pub(super) fn element_tag(node: &Handle) -> Option<String> {
    if let NodeData::Element { name, .. } = &node.data {
        Some(name.local.as_ref().to_ascii_lowercase())
    } else {
        None
    }
}

pub(super) fn element_attrs(node: &Handle) -> Vec<Attribute> {
    if let NodeData::Element { attrs, .. } = &node.data {
        attrs.borrow().clone()
    } else {
        Vec::new()
    }
}

pub(super) fn attr_value(node: &Handle, name: &str) -> Option<String> {
    if let NodeData::Element { attrs, .. } = &node.data {
        for a in attrs.borrow().iter() {
            if a.name.local.as_ref().eq_ignore_ascii_case(name) {
                return Some(a.value.as_ref().to_string());
            }
        }
    }
    None
}

pub(super) fn attrs_contains(attrs: &[Attribute], name: &str, value: &str) -> bool {
    attrs.iter().any(|a| {
        a.name.local.as_ref().eq_ignore_ascii_case(name)
            && a.value.as_ref().eq_ignore_ascii_case(value)
    })
}

pub(super) fn has_class(attrs: &[Attribute], class: &str) -> bool {
    attrs.iter().any(|a| {
        if !a.name.local.as_ref().eq_ignore_ascii_case("class") {
            return false;
        }
        a.value.split_ascii_whitespace().any(|c| c == class)
    })
}

pub(super) fn find_body(node: &Handle) -> Option<Handle> {
    if element_tag(node).as_deref() == Some("body") {
        return Some(node.clone());
    }
    for child in node.children.borrow().iter() {
        if let Some(found) = find_body(child) {
            return Some(found);
        }
    }
    None
}

pub(super) fn direct_children_of_tag_any(node: &Handle, tags: &[&str]) -> Vec<Handle> {
    node.children
        .borrow()
        .iter()
        .filter(|c| {
            element_tag(c)
                .as_deref()
                .map(|t| tags.contains(&t))
                .unwrap_or(false)
        })
        .cloned()
        .collect()
}

pub(super) fn collect_table_rows(table_el: &Handle) -> Vec<Handle> {
    let mut rows: Vec<Handle> = Vec::new();
    for child in table_el.children.borrow().iter() {
        match element_tag(child).as_deref() {
            Some("tbody" | "thead" | "tfoot") => {
                for tr in child.children.borrow().iter() {
                    if element_tag(tr).as_deref() == Some("tr") {
                        rows.push(tr.clone());
                    }
                }
            }
            Some("tr") => rows.push(child.clone()),
            _ => {}
        }
    }
    rows
}

// ── Helpers: inline extraction ────────────────────────────────────

#[derive(Default, Debug, Clone)]
pub(super) struct MarkSet {
    bold: bool,
    italic: bool,
    underline: bool,
    strike: bool,
    code: bool,
    link: Option<String>,
    color: Option<String>,
    highlight: Option<String>,
}

impl MarkSet {
    pub(super) fn empty() -> Self {
        Self::default()
    }

    pub(super) fn to_marks(&self) -> Vec<Mark> {
        let mut out = Vec::new();
        if self.bold {
            out.push(Mark::bold());
        }
        if self.italic {
            out.push(Mark::italic());
        }
        if self.underline {
            out.push(Mark::underline());
        }
        if self.strike {
            out.push(Mark::strike());
        }
        if self.code {
            out.push(Mark::code());
        }
        if let Some(href) = &self.link {
            let mut a = Map::new();
            a.insert("href".into(), Value::String(href.clone()));
            out.push(Mark::with_attrs("link", a));
        }
        if let Some(color) = &self.color {
            let mut style = Map::new();
            style.insert("color".into(), Value::String(color.clone()));
            let mut a = Map::new();
            a.insert("style".into(), Value::Object(style));
            out.push(Mark::with_attrs("color", a));
        }
        if let Some(bg) = &self.highlight {
            let mut style = Map::new();
            style.insert("backgroundColor".into(), Value::String(bg.clone()));
            let mut a = Map::new();
            a.insert("style".into(), Value::Object(style));
            out.push(Mark::with_attrs("highlight", a));
        }
        out
    }
}

pub(super) fn extend_marks(base: &MarkSet, tag: &str, attrs: &[Attribute]) -> MarkSet {
    let mut next = base.clone();
    match tag {
        "strong" | "b" => next.bold = true,
        "em" | "i" => next.italic = true,
        "u" | "ins" => next.underline = true,
        "s" | "strike" | "del" => next.strike = true,
        "code" | "kbd" | "samp" => next.code = true,
        "a" => {
            if let Some(href) = attrs
                .iter()
                .find(|a| a.name.local.as_ref().eq_ignore_ascii_case("href"))
            {
                let v = href.value.as_ref().to_string();
                if !v.is_empty() {
                    next.link = Some(v);
                }
            }
        }
        _ => {}
    }
    // Table structure tags: DO NOT promote inline style to text marks.
    if TABLE_STRUCTURE_TAGS.contains(&tag) {
        return next;
    }
    // Inline style fallback — Word/HWP/Google Docs rely on these.
    if let Some(style_attr) = attrs
        .iter()
        .find(|a| a.name.local.as_ref().eq_ignore_ascii_case("style"))
        .map(|a| a.value.as_ref().to_string())
    {
        let decl = parse_inline_style(&style_attr);
        if let Some(fw) = decl.font_weight
            && is_bold_weight(&fw)
        {
            next.bold = true;
        }
        if let Some(fs) = decl.font_style
            && (fs == "italic" || fs == "oblique")
        {
            next.italic = true;
        }
        if let Some(td) = decl.text_decoration {
            if td.contains("underline") {
                next.underline = true;
            }
            if td.contains("line-through") {
                next.strike = true;
            }
        }
        if let Some(color) = decl.color {
            next.color = Some(color);
        }
        if let Some(bg) = decl.background_color {
            next.highlight = Some(bg);
        }
    }
    next
}

pub(super) fn is_bold_weight(v: &str) -> bool {
    if v.eq_ignore_ascii_case("bold") || v.eq_ignore_ascii_case("bolder") {
        return true;
    }
    v.parse::<u32>().map(|n| n >= 600).unwrap_or(false)
}

pub(super) fn extract_spans(node: &Handle) -> Vec<TextSpan> {
    // Inherit marks from the root element (e.g. <p style="color:red">).
    let initial = match &node.data {
        NodeData::Element { name, attrs, .. } => extend_marks(
            &MarkSet::empty(),
            &name.local.as_ref().to_ascii_lowercase(),
            &attrs.borrow(),
        ),
        _ => MarkSet::empty(),
    };
    let mut spans: Vec<TextSpan> = Vec::new();
    for child in node.children.borrow().iter() {
        collect_inline_into(child, &mut spans, &initial);
    }
    // merge adjacent equal-mark spans; drop empties
    normalize_spans(&mut spans);
    spans
}

pub(super) fn collect_inline_into(node: &Handle, out: &mut Vec<TextSpan>, marks: &MarkSet) {
    match &node.data {
        NodeData::Text { contents } => {
            let text = contents.borrow().to_string();
            if !text.is_empty() {
                out.push(TextSpan::with_marks(text, marks.to_marks()));
            }
        }
        NodeData::Element { name, attrs, .. } => {
            let tag = name.local.as_ref().to_ascii_lowercase();
            if tag == "br" {
                out.push(TextSpan::with_marks("\n", marks.to_marks()));
                return;
            }
            if tag == "input" {
                return; // checkboxes contribute no text
            }
            let next = extend_marks(marks, &tag, &attrs.borrow());
            for c in node.children.borrow().iter() {
                collect_inline_into(c, out, &next);
            }
        }
        _ => {}
    }
}

/// Extract spans from a `<li>`, excluding nested `<ul>`/`<ol>`/
/// `<details>` so their descendants don't leak into the parent item.
pub(super) fn extract_spans_from_li(li: &Handle) -> Vec<TextSpan> {
    let filtered = clone_node_filter_direct(li, &["ul", "ol", "details"]);
    extract_spans(&filtered)
}

/// Collect every text descendant into a string preserving whitespace —
/// used for `<pre>` contents.
pub(super) fn collect_raw_text(node: &Handle) -> String {
    let mut s = String::new();
    walk_text(node, &mut s);
    s
}
pub(super) fn walk_text(node: &Handle, out: &mut String) {
    match &node.data {
        NodeData::Text { contents } => out.push_str(&contents.borrow()),
        NodeData::Element { .. } => {
            for c in node.children.borrow().iter() {
                walk_text(c, out);
            }
        }
        _ => {}
    }
}

// ── Helpers: inline-style subset parser ──────────────────────────

#[derive(Default, Debug, Clone)]
pub(super) struct InlineStyle {
    pub(super) background_color: Option<String>,
    pub(super) border_color: Option<String>,
    pub(super) border_width: Option<String>,
    pub(super) border_style: Option<String>,
    pub(super) vertical_align: Option<String>,
    pub(super) padding: Option<String>,
    pub(super) height: Option<String>,
    pub(super) width: Option<String>,
    pub(super) color: Option<String>,
    pub(super) font_weight: Option<String>,
    pub(super) font_style: Option<String>,
    pub(super) text_decoration: Option<String>,
}

pub(super) fn parse_inline_style(style_attr: &str) -> InlineStyle {
    let mut out = InlineStyle::default();
    for decl in style_attr.split(';') {
        let Some((k, v)) = decl.split_once(':') else {
            continue;
        };
        let key = k.trim().to_ascii_lowercase();
        let value = v.trim();
        if value.is_empty() {
            continue;
        }
        match key.as_str() {
            "background-color" => out.background_color = Some(value.to_string()),
            "border-color" => out.border_color = Some(value.to_string()),
            "border-width" => out.border_width = Some(value.to_string()),
            "border-style" => out.border_style = Some(value.to_string()),
            "vertical-align" => out.vertical_align = Some(value.to_string()),
            "padding" => out.padding = Some(value.to_string()),
            "height" => out.height = Some(value.to_string()),
            "width" => out.width = Some(value.to_string()),
            "color" => out.color = Some(value.to_string()),
            "font-weight" => out.font_weight = Some(value.to_string()),
            "font-style" => out.font_style = Some(value.to_string()),
            "text-decoration" | "text-decoration-line" => {
                let cur = out.text_decoration.unwrap_or_default();
                let combined = if cur.is_empty() {
                    value.to_string()
                } else {
                    format!("{cur} {value}")
                };
                out.text_decoration = Some(combined);
            }
            _ => {}
        }
    }
    out
}

// ── Helpers: table prop extraction ───────────────────────────────

pub(super) fn extract_cell_props(cell_el: &Handle) -> Option<Map<String, Value>> {
    let mut out = decode_props_from_element(cell_el).unwrap_or_default();

    if let Some(cs) = attr_value(cell_el, "colspan").and_then(|s| s.parse::<u64>().ok())
        && cs > 1
    {
        out.insert("colspan".into(), Value::from(cs));
    }
    if let Some(rs) = attr_value(cell_el, "rowspan").and_then(|s| s.parse::<u64>().ok())
        && rs > 1
    {
        out.insert("rowspan".into(), Value::from(rs));
    }

    if let Some(style_attr) = attr_value(cell_el, "style") {
        let decl = parse_inline_style(&style_attr);
        if !out.contains_key("backgroundColor")
            && let Some(v) = decl.background_color
        {
            out.insert("backgroundColor".into(), Value::String(v));
        }
        if !out.contains_key("borderColor")
            && let Some(v) = decl.border_color
        {
            out.insert("borderColor".into(), Value::String(v));
        }
        if !out.contains_key("borderWidth")
            && let Some(v) = decl.border_width
        {
            out.insert("borderWidth".into(), Value::String(v));
        }
        if !out.contains_key("borderStyle")
            && let Some(v) = decl.border_style
        {
            out.insert("borderStyle".into(), Value::String(v));
        }
        if !out.contains_key("verticalAlign")
            && let Some(v) = decl.vertical_align
        {
            out.insert("verticalAlign".into(), Value::String(v));
        }
        if !out.contains_key("padding")
            && let Some(v) = decl.padding
        {
            out.insert("padding".into(), Value::String(v));
        }
    }

    if out.is_empty() { None } else { Some(out) }
}

pub(super) fn extract_row_props(row_el: &Handle) -> Option<Map<String, Value>> {
    let mut out = decode_props_from_element(row_el).unwrap_or_default();
    if !out.contains_key("height")
        && let Some(style_attr) = attr_value(row_el, "style")
        && let Some(v) = parse_inline_style(&style_attr).height
    {
        out.insert("height".into(), Value::String(v));
    }

    // Normalize height: string "48px" → number 48.
    //
    // Clipboard HTML is untrusted input. We guard against every f64
    // edge case (NaN, ±∞, subnormals, overflow) so a crafted `style`
    // attribute cannot panic the Rust engine — which would take the
    // whole editor down via WASM trap. Each branch either produces a
    // valid JSON number or removes the `height` key entirely.
    if let Some(Value::String(s)) = out.get("height") {
        let parsed = s
            .trim_end_matches("px")
            .trim_end_matches("PX")
            .parse::<f64>()
            .ok();
        // Reject anything we can't safely serialise as a JSON number:
        // NaN has no ordering, ±∞ overflow JSON number, non-positive
        // heights don't make semantic sense for a table row.
        let normalised = parsed.filter(|v| v.is_finite() && *v > 0.0);
        match normalised {
            // float_cmp: We want **exact** integral comparison here,
            // not a fuzzy epsilon check. `v == v.trunc()` is the
            // standard idiom for "v has no fractional part"; a margin
            // comparison would misclassify 47.9999999 as integral and
            // round-trip to `48` via the i64 branch below, which is
            // wrong.
            #[allow(clippy::float_cmp)]
            Some(v) if v == v.trunc() => {
                // cast_possible_truncation / cast_sign_loss: v is
                // guaranteed finite and positive by the filter above.
                // Rust's `f64 as i64` saturates on overflow (values
                // above i64::MAX become i64::MAX, NaN becomes 0) so
                // this cast cannot panic regardless of input. The
                // only lossy case is heights > i64::MAX px, which no
                // real browser will ever produce.
                #[allow(clippy::cast_possible_truncation, clippy::cast_sign_loss)]
                let as_int = v as i64;
                out.insert("height".into(), Value::from(as_int));
            }
            Some(v) => {
                // Finite, positive, non-integer → safe for JSON.
                // `from_f64` only returns None for NaN/±∞ which we
                // already excluded, but we defensively fall through
                // to `remove` if that invariant ever changes.
                match serde_json::Number::from_f64(v) {
                    Some(n) => {
                        out.insert("height".into(), Value::from(n));
                    }
                    None => {
                        out.remove("height");
                    }
                }
            }
            None => {
                out.remove("height");
            }
        }
    }

    if out.is_empty() { None } else { Some(out) }
}

pub(super) fn decode_props_from_element(el: &Handle) -> Option<Map<String, Value>> {
    let raw = attr_value(el, "data-devup-props")?;
    decode_props(&raw)
}

pub(super) fn extract_colgroup_widths(table_el: &Handle, cols: usize) -> Option<Vec<f64>> {
    let colgroup = table_el
        .children
        .borrow()
        .iter()
        .find(|c| element_tag(c).as_deref() == Some("colgroup"))
        .cloned()?;
    let col_els: Vec<Handle> = colgroup
        .children
        .borrow()
        .iter()
        .filter(|c| element_tag(c).as_deref() == Some("col"))
        .cloned()
        .collect();
    if col_els.is_empty() {
        return None;
    }
    let mut widths = Vec::with_capacity(cols);
    let mut saw_any = false;
    for i in 0..cols {
        let col = col_els.get(i);
        let w = col
            .and_then(|c| attr_value(c, "style"))
            .and_then(|s| parse_inline_style(&s).width)
            .and_then(|s| {
                let trimmed = s.trim().trim_end_matches("px");
                trimmed.parse::<f64>().ok().filter(|n| *n > 0.0)
            })
            .or_else(|| {
                col.and_then(|c| attr_value(c, "width"))
                    .and_then(|s| s.parse::<f64>().ok())
                    .filter(|n| *n > 0.0)
            });
        match w {
            Some(n) => {
                widths.push(n);
                saw_any = true;
            }
            None => widths.push(120.0),
        }
    }
    if saw_any { Some(widths) } else { None }
}

// ── Helpers: node cloning / filtering ─────────────────────────────

use std::cell::RefCell;
use std::rc::Rc;

/// Build a synthetic `<div>` host whose children are the given
/// handles. The node is NOT inserted into the original tree; it just
/// lets us reuse `process_children_with_indent` on a subrange.
pub(super) fn build_synthetic_parent(handles: &[Handle]) -> Handle {
    let synthetic = markup5ever_rcdom::Node::new(NodeData::Element {
        name: html5ever::QualName::new(
            None,
            markup5ever::ns!(html),
            markup5ever::local_name!("div"),
        ),
        attrs: RefCell::new(Vec::new()),
        template_contents: RefCell::new(None),
        mathml_annotation_xml_integration_point: false,
    });
    for h in handles {
        synthetic.children.borrow_mut().push(h.clone());
    }
    synthetic
}

/// Clone a node but drop every direct child whose tag is in `strip_tags`.
/// Used to lift toggles / nested lists out of an `<li>` before inline
/// extraction.
pub(super) fn clone_node_filter_direct(node: &Handle, strip_tags: &[&str]) -> Handle {
    let cloned = deep_clone_element(node);
    cloned.children.borrow_mut().retain(|c| {
        element_tag(c)
            .map(|t| !strip_tags.contains(&t.as_str()))
            .unwrap_or(true)
    });
    cloned
}

pub(super) fn clone_node_without_checkboxes(node: &Handle) -> Handle {
    let cloned = deep_clone_element(node);
    strip_checkboxes_in_place(&cloned);
    cloned
}

pub(super) fn strip_checkboxes_in_place(node: &Handle) {
    node.children.borrow_mut().retain(|c| {
        if let Some(t) = element_tag(c)
            && t == "input"
        {
            let is_cb = element_attrs(c).iter().any(|a| {
                a.name.local.as_ref().eq_ignore_ascii_case("type")
                    && a.value.as_ref().eq_ignore_ascii_case("checkbox")
            });
            if is_cb {
                return false;
            }
        }
        true
    });
    for c in node.children.borrow().iter() {
        strip_checkboxes_in_place(c);
    }
}

pub(super) fn strip_nested_blocks(node: &Handle) -> Handle {
    let cloned = deep_clone_element(node);
    cloned.children.borrow_mut().retain(|c| {
        if let Some(t) = element_tag(c) {
            !matches!(
                t.as_str(),
                "ul" | "ol" | "details" | "pre" | "blockquote" | "table"
            )
        } else {
            true
        }
    });
    cloned
}

/// Deep clone an element (element data + attrs + recursively its
/// children). Text nodes are cloned by copying contents.
pub(super) fn deep_clone_element(node: &Handle) -> Handle {
    let data = match &node.data {
        NodeData::Element { name, attrs, .. } => NodeData::Element {
            name: name.clone(),
            attrs: RefCell::new(attrs.borrow().clone()),
            template_contents: RefCell::new(None),
            mathml_annotation_xml_integration_point: false,
        },
        NodeData::Text { contents } => NodeData::Text {
            contents: RefCell::new(contents.borrow().clone()),
        },
        NodeData::Comment { contents } => NodeData::Comment {
            contents: contents.clone(),
        },
        NodeData::Doctype {
            name,
            public_id,
            system_id,
        } => NodeData::Doctype {
            name: name.clone(),
            public_id: public_id.clone(),
            system_id: system_id.clone(),
        },
        NodeData::ProcessingInstruction { target, contents } => NodeData::ProcessingInstruction {
            target: target.clone(),
            contents: contents.clone(),
        },
        NodeData::Document => NodeData::Document,
    };
    let new_node = Rc::new(markup5ever_rcdom::Node {
        parent: std::cell::Cell::new(None),
        children: RefCell::new(Vec::new()),
        data,
    });
    for c in node.children.borrow().iter() {
        new_node.children.borrow_mut().push(deep_clone_element(c));
    }
    new_node
}

// ── Helpers: list / checkbox detection ────────────────────────────

pub(super) fn is_notion_v3_toggle(li: &Handle) -> bool {
    let block_children_count = 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
            }
        })
        .count();
    if block_children_count < 2 {
        return false;
    }
    // First block child must be a `<p>` (title).
    li.children
        .borrow()
        .iter()
        .find_map(element_tag)
        .map(|t| t == "p")
        .unwrap_or(false)
}

pub(super) fn detect_direct_checkbox(li: &Handle) -> Option<bool> {
    for c in li.children.borrow().iter() {
        if let Some(t) = element_tag(c) {
            if t == "input" && is_checkbox(c) {
                return Some(checkbox_is_checked(c));
            }
            if t == "div" {
                for gc in c.children.borrow().iter() {
                    if element_tag(gc).as_deref() == Some("input") && is_checkbox(gc) {
                        return Some(checkbox_is_checked(gc));
                    }
                }
            }
        }
    }
    None
}

pub(super) fn detect_any_checkbox(li: &Handle) -> Option<bool> {
    fn walk(node: &Handle) -> Option<bool> {
        if let Some(t) = element_tag(node)
            && t == "input"
            && is_checkbox(node)
        {
            return Some(checkbox_is_checked(node));
        }
        for c in node.children.borrow().iter() {
            if let Some(r) = walk(c) {
                return Some(r);
            }
        }
        None
    }
    walk(li)
}

pub(super) fn is_checkbox(input: &Handle) -> bool {
    attr_value(input, "type")
        .as_deref()
        .unwrap_or("")
        .eq_ignore_ascii_case("checkbox")
}

pub(super) fn checkbox_is_checked(input: &Handle) -> bool {
    if let NodeData::Element { attrs, .. } = &input.data {
        attrs
            .borrow()
            .iter()
            .any(|a| a.name.local.as_ref().eq_ignore_ascii_case("checked"))
    } else {
        false
    }
}

pub(super) fn has_descendant_with_class(node: &Handle, class: &str) -> bool {
    if let NodeData::Element { attrs, .. } = &node.data
        && has_class(&attrs.borrow(), class)
    {
        return true;
    }
    for c in node.children.borrow().iter() {
        if has_descendant_with_class(c, class) {
            return true;
        }
    }
    false
}

pub(super) fn find_descendant_with_any_class(node: &Handle, classes: &[&str]) -> Option<Handle> {
    if let NodeData::Element { attrs, .. } = &node.data
        && classes.iter().any(|c| has_class(&attrs.borrow(), c))
    {
        return Some(node.clone());
    }
    for c in node.children.borrow().iter() {
        if let Some(found) = find_descendant_with_any_class(c, classes) {
            return Some(found);
        }
    }
    None
}

pub(super) fn is_all_whitespace(spans: &[TextSpan]) -> bool {
    spans.iter().all(|s| s.text.trim().is_empty())
}