ocpi_tariffs/

json.rs

//! JSON parsing and typed decoding for the OCPI CDR pricing/generating and CDR/Tariff linting pipeline.
//!
//! # Parsing vs decoding
//!
//! Parsing and decoding are intentionally separated so the linter can emit
//! actionable warnings rather than hard parse errors.
//!
//! **Parsing** ([`parse_object`]) converts a raw JSON `&str` into an
//! [`Element`] tree. The parser is deliberately lenient about string content:
//! it only verifies structural correctness (balanced delimiters, valid
//! top-level values) and leaves escape sequences and control characters
//! untouched inside [`RawStr`].
//!
//! **Decoding** ([`decode`]) interprets the raw JSON String as a `&str`.
//! Calling [`RawStr::decode_escapes`] validates escape sequences and
//! rejects control characters, returning [`decode::Warning`]
//! values instead of hard errors. This lets the linter pinpoint the exact
//! field, report what is wrong, and suggest a corrected encoding.
//! The `price` and `generate` mods can choose to hard fail on specific `Warning`s.
//!
pub mod decode;
mod parser;
pub mod write;

#[cfg(test)]
pub(crate) mod test;

#[cfg(test)]
mod test_line_col;

#[cfg(test)]
mod test_path;

#[cfg(test)]
mod test_path_matches_glob;

#[cfg(test)]
mod test_removal;

#[cfg(test)]
mod test_source_json;

use std::{
    borrow::{Borrow, Cow},
    collections::{btree_set, BTreeMap, BTreeSet},
    fmt::{self, Write as _},
    rc::Rc,
};

use crate::{
    string,
    warning::{Caveat, CaveatDeferred},
};

pub(crate) use parser::parse;
pub use parser::{Error, ErrorKind as ParseErrorKind};

/// Parse a raw JSON `&str` into a [`Document`] and require the root value to be a JSON object.
///
/// The input size is gated: 5 MB (5,000,000 bytes) or more returns
/// [`ParseError::SizeExceedsMax`].
pub fn parse_object(json: &str) -> Result<Document<'_>, ParseError> {
    let json = string::ReasonableLen::new(json).map_err(|_e| ParseError::SizeExceedsMax)?;
    let doc = parse(json).map_err(ParseError::Json)?;

    if !doc.root().is_object() {
        return Err(ParseError::ShouldBeAnObject);
    }

    Ok(doc)
}

#[derive(Debug)]
/// Why a raw JSON `&str` could not be turned into a [`Document`].
pub enum ParseError {
    /// The JSON parser was unable to parse the JSON str.
    Json(Error),

    /// The OCPI object should be a JSON object.
    ShouldBeAnObject,

    /// The size of the input `str` exceeds the maximum deemed reasonable.
    SizeExceedsMax,
}

impl fmt::Display for ParseError {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        match self {
            Self::Json(error) => write!(f, "{error}"),
            Self::ShouldBeAnObject => f.write_str("The CDR should be an object."),
            Self::SizeExceedsMax => write!(
                f,
                "The input `&str` exceeds the reasonable maximum `{} MB`.",
                string::ReasonableLen::FACTOR
            ),
        }
    }
}

impl std::error::Error for ParseError {
    fn source(&self) -> Option<&(dyn std::error::Error + 'static)> {
        match &self {
            ParseError::Json(err) => Some(err),
            ParseError::ShouldBeAnObject | ParseError::SizeExceedsMax => None,
        }
    }
}

/// The output of [`parse_object`]: the element tree with path resolution embedded in each
/// element.
#[derive(Clone, Debug)]
pub struct Document<'buf> {
    /// Shared inner state; also held by every element in the tree.
    inner: Rc<DocumentInner<'buf>>,
    /// Root element of the parsed tree.
    root: Element<'buf>,
}

impl<'buf> Document<'buf> {
    /// Returns the source JSON string this document was parsed from.
    pub fn source(&self) -> &'buf str {
        self.inner.source
    }

    /// Returns the root element of this document.
    pub fn root(&self) -> &Element<'buf> {
        &self.root
    }

    /// Returns the element with `id`, or `None` if this document has no such element.
    ///
    /// Walks the parent chain up from `id` and then descends back down the tree, so the
    /// cost is O(depth) plus a scan of each ancestor's children. Meant for looking up
    /// the occasional edit site, not for bulk traversal.
    ///
    /// NOTE: an [`ElemId`] produced by a different parse of the same source `&str`
    /// cannot be told apart from one of this document's own ids.
    pub fn element(&self, id: ElemId) -> Option<&Element<'buf>> {
        let ancestry = self.inner.paths.ancestry(id)?;
        let (root_id, descent) = ancestry.split_first()?;

        if *root_id != self.root.id {
            return None;
        }

        let mut element = &self.root;

        for child_id in descent {
            element = child_of(element, *child_id)?;
        }

        Some(element)
    }

    /// Resolves the source spans to erase in order to remove every element in `ids`.
    ///
    /// The returned spans are sorted by start offset and never overlap, so erasing them
    /// from [`Document::source`] leaves JSON that still parses.
    ///
    /// Siblings are resolved as a group, because which span removes a child depends on
    /// which of its siblings survive:
    ///
    /// | Case | Span erased |
    /// |---|---|
    /// | Run of children followed by a surviving sibling | the run's first key or value, up to the start of that sibling |
    /// | Run of children reaching the closing delimiter | the end of the last surviving sibling's value, up to the end of the run |
    /// | Every child of an object or array | everything between the `{}` or `[]` delimiters |
    ///
    /// A run that reaches the closing delimiter carries no trailing comma of its own, so
    /// it has to swallow the comma of the sibling before it. Resolving each id on its own
    /// and merging the spans afterwards gets that case wrong: it leaves the earlier
    /// sibling's comma dangling.
    ///
    /// An id nested inside another id in the same set needs nothing beyond being dropped,
    /// since the ancestor's span already covers it.
    pub fn removal_spans(&self, ids: &BTreeSet<ElemId>) -> Result<Vec<Span>, RemovalError> {
        let mut by_parent: BTreeMap<ElemId, BTreeSet<ElemId>> = BTreeMap::new();

        for id in ids {
            let parent = self.parent_of(*id)?;
            by_parent.entry(parent).or_default().insert(*id);
        }

        let mut spans: Vec<Span> = Vec::new();

        for (parent_id, removed) in by_parent {
            let parent = self
                .element(parent_id)
                .ok_or(RemovalError::UnknownElement(parent_id))?;

            spans.extend(sibling_removal_spans(parent, &removed)?);
        }

        spans.sort_unstable();

        Ok(without_nested(spans))
    }

    /// Return the id of the parent of the [`Element`] with the given `id`.
    fn parent_of(&self, id: ElemId) -> Result<ElemId, RemovalError> {
        match self.inner.paths.entries.get(id.0) {
            None => Err(RemovalError::UnknownElement(id)),
            Some(PathEntry::Root) => Err(RemovalError::Root),
            Some(PathEntry::Field { parent, key: _ } | PathEntry::Item { parent, index: _ }) => {
                Ok(*parent)
            }
        }
    }
}

/// Why a set of elements could not be resolved to removal spans.
#[derive(Copy, Clone, Debug, Eq, PartialEq)]
pub enum RemovalError {
    /// The root element has no parent, so removing it would leave no document behind.
    Root,

    /// No element in this document has this id.
    UnknownElement(ElemId),
}

impl fmt::Display for RemovalError {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        match self {
            Self::Root => f.write_str("The root element can not be removed."),
            Self::UnknownElement(id) => write!(f, "The document has no element with id `{id}`."),
        }
    }
}

impl std::error::Error for RemovalError {}

/// Returns the direct child of `parent` that has `id`.
fn child_of<'a, 'buf>(parent: &'a Element<'buf>, id: ElemId) -> Option<&'a Element<'buf>> {
    match parent.value() {
        Value::Object(fields) => fields
            .iter()
            .map(Field::element)
            .find(|element| element.id == id),
        Value::Array(items) => items.iter().find(|element| element.id == id),
        Value::Null | Value::True | Value::False | Value::String(_) | Value::Number(_) => None,
    }
}

/// Resolves the spans to erase to remove `removed` from `parent`, in source order.
fn sibling_removal_spans(
    parent: &Element<'_>,
    removed: &BTreeSet<ElemId>,
) -> Result<Vec<Span>, RemovalError> {
    let children = child_spans(parent);

    for id in removed {
        if !children.iter().any(|child| child.id == *id) {
            return Err(RemovalError::UnknownElement(*id));
        }
    }

    let mut spans: Vec<Span> = Vec::new();
    let mut kept_end: Option<u32> = None;
    let mut run: Option<Run> = None;

    for child in &children {
        if removed.contains(&child.id) {
            let start = run.map(|open| open.start).unwrap_or(child.start);
            run = Some(Run {
                start,
                end: child.end,
                next_sibling_start: child.next_sibling_start,
            });
            continue;
        }

        // A surviving sibling closes the open run: erase up to where that sibling starts,
        // which takes the run's own trailing comma with it.
        if let Some(open) = run.take() {
            spans.push(Span::new(open.start, open.next_sibling_start));
        }

        kept_end = Some(child.end);
    }

    let Some(open) = run else {
        return Ok(spans);
    };

    // A run still open here runs up to the closing delimiter, so it has no trailing comma
    // of its own and has to swallow the comma of the sibling before it. With no surviving
    // sibling before it, every child was removed and the whole interior goes.
    let span = match kept_end {
        Some(end) => Span::new(end, open.end),
        None => interior_span(parent),
    };

    spans.push(span);

    Ok(spans)
}

/// A maximal run of adjacent children being removed from the same parent.
#[derive(Copy, Clone, Debug)]
struct Run {
    /// First byte to erase; the `start` of the run's first child.
    start: u32,
    /// One past the last byte of the run's last value.
    end: u32,
    /// The `next_sibling_start` of the run's last child.
    next_sibling_start: u32,
}

/// Where a child of an object or array sits in the source, as far as removal is concerned.
#[derive(Copy, Clone, Debug)]
struct ChildSpan {
    id: ElemId,
    /// First byte to erase: the key of an object field, the value of an array item.
    start: u32,
    /// One past the last byte of the child's value.
    end: u32,
    /// Where the next sibling starts: one past this child's trailing comma and the
    /// whitespace after it. Equal to `end` for the last child, which has no comma.
    next_sibling_start: u32,
}

/// The removal spans of every child of `parent`, in source order.
fn child_spans(parent: &Element<'_>) -> Vec<ChildSpan> {
    match parent.value() {
        Value::Object(fields) => fields
            .iter()
            .map(|field| ChildSpan {
                id: field.element.id,
                start: field.key_span.start,
                end: field.element.span.end,
                next_sibling_start: field.element.full_span_end,
            })
            .collect(),
        Value::Array(items) => items
            .iter()
            .map(|item| ChildSpan {
                id: item.id,
                start: item.span.start,
                end: item.span.end,
                next_sibling_start: item.full_span_end,
            })
            .collect(),
        Value::Null | Value::True | Value::False | Value::String(_) | Value::Number(_) => {
            Vec::new()
        }
    }
}

/// The span between an object's or array's delimiters, each of which is one byte wide.
fn interior_span(element: &Element<'_>) -> Span {
    let end = element.span.end.saturating_sub(1);
    let start = element.span.start.saturating_add(1).min(end);

    Span::new(start, end)
}

/// Drops the spans that sit inside an earlier span.
///
/// A removed element nested inside another removed element is already covered by its
/// ancestor's span. JSON is a tree, so two removal spans either nest or are disjoint;
/// they never partially overlap.
fn without_nested(spans: Vec<Span>) -> Vec<Span> {
    let mut kept: Vec<Span> = Vec::new();

    for span in spans {
        let Some(last) = kept.last_mut() else {
            kept.push(span);
            continue;
        };

        if span.start >= last.end {
            kept.push(span);
            continue;
        }

        // Nesting means `last` already reaches past `span`, so this leaves it as it is.
        // Extending rather than ignoring keeps a partial overlap from under-erasing.
        last.end = last.end.max(span.end);
    }

    kept
}

/// A JSON [`Element`] with identity, source span, and value.
///
/// Each element carries a shared reference to the document it was parsed from,
/// so [`Element::path()`] can resolve its path.
#[derive(Clone, Debug)]
pub struct Element<'buf> {
    /// Shared document state.
    doc: Rc<DocumentInner<'buf>>,
    /// Unique identifier within the document; sequentially assigned depth-first.
    id: ElemId,
    /// Byte range of the value only; use for replacement edits.
    span: Span,
    /// End of the value plus any trailing comma and whitespace; use for removal edits.
    /// Equal to `span.end` when there is no trailing comma (root element, or last sibling).
    full_span_end: u32,
    /// Parsed value, borrowing from the source `&str`.
    value: Value<'buf>,
}

impl PartialEq for Element<'_> {
    fn eq(&self, other: &Self) -> bool {
        self.id == other.id
            && self.span == other.span
            && self.full_span_end == other.full_span_end
            && self.value == other.value
    }
}

impl Eq for Element<'_> {}

impl<'buf> Element<'buf> {
    /// Returns this element's identity within its [`Document`].
    pub fn id(&self) -> ElemId {
        self.id
    }

    /// Returns the span covering this element's value alone, without any trailing comma.
    ///
    /// See [`Element::full_span`] for the span that reaches the next sibling.
    pub fn span(&self) -> Span {
        self.span
    }

    /// Returns the span covering the value plus any trailing comma and the whitespace
    /// after it, so the span ends where the next sibling begins.
    ///
    /// When there is no trailing comma (the root element, or the last child of its
    /// parent), `full_span() == span()`.
    ///
    /// Use [`Element::span`] to replace a value. Removal is not simply a matter of
    /// erasing `full_span`: which span removes a child depends on which of its siblings
    /// survive, and erasing the `full_span` of a last child leaves a dangling comma on
    /// the sibling before it. Use [`Document::removal_spans`], which resolves a whole set
    /// of removals against the surviving siblings.
    pub fn full_span(&self) -> Span {
        Span {
            start: self.span.start,
            end: self.full_span_end,
        }
    }

    /// Returns this element's JSON value.
    pub fn value(&self) -> &Value<'buf> {
        &self.value
    }

    /// Returns the RFC 9535 path to this element.
    ///
    /// NOTE: The `Path` is constructed anew every time this functions is called.
    pub fn path(&self) -> Path {
        self.doc.paths.path_of(self)
    }

    /// Returns the slice of the source JSON that this element spans.
    #[expect(
        clippy::string_slice,
        reason = "spans are produced by the parser from the same source, so slices are always valid"
    )]
    #[expect(
        clippy::as_conversions,
        reason = "The index is guaranteed within bounds by the parser"
    )]
    pub fn source_json_value(&self) -> &'buf str {
        &self.doc.source[self.span.start as usize..self.span.end as usize]
    }

    /// The full source string; all element spans are relative to this.
    pub fn source(&self) -> &'buf str {
        self.doc.source
    }

    /// Return the location that this element begins at.
    #[expect(
        clippy::string_slice,
        reason = "spans are produced by the parser from the same source, so slices are always valid"
    )]
    #[expect(
        clippy::as_conversions,
        reason = "The index is guaranteed within bounds by the parser"
    )]
    pub fn location(&self) -> Location {
        let source = self.doc.source;

        // Slice up to the start of the span to calculate line and col numbers.
        let lead_in = &source[..self.span.start as usize];
        line_col(lead_in)
    }

    /// Return the inner `Value` by ref.
    pub fn as_value(&self) -> &Value<'buf> {
        &self.value
    }

    /// Return `Some(&str)` if the `Value` is a `String`.
    pub fn to_raw_str(&self) -> Option<RawStr<'buf>> {
        self.value.to_raw_str()
    }

    /// Return `Some(&[Field])` if the `Value` is a `Object`.
    pub fn as_object_fields(&self) -> Option<&[Field<'buf>]> {
        self.value.as_object_fields()
    }

    /// Returns the items if this element is a JSON array, `None` otherwise.
    pub fn as_array(&self) -> Option<&[Element<'buf>]> {
        self.value.as_array()
    }

    /// Returns the number as it was written in the source if this element is a JSON number,
    /// `None` otherwise. The text is not parsed, so it keeps the precision the document had.
    pub fn as_number_str(&self) -> Option<&str> {
        self.value.as_number()
    }

    /// Return true if the `Element`s `Value` is null.
    pub fn is_null(&self) -> bool {
        self.value.is_null()
    }

    /// Return true if the `Element`s `Value` is an object.
    pub fn is_object(&self) -> bool {
        self.value.is_object()
    }

    /// Return true if the `Element`s `Value` is an array.
    pub fn is_array(&self) -> bool {
        self.value.is_array()
    }
}

/// A JSON value that borrows its content from the source JSON `&str`.
#[derive(Clone, Debug, Eq, PartialEq)]
pub enum Value<'buf> {
    /// JSON `null` literal.
    Null,
    /// JSON `true` literal.
    True,
    /// JSON `false` literal.
    False,
    /// String content with quotes removed; escape sequences are not decoded.
    String(RawStr<'buf>),
    /// Raw number text; not guaranteed to fit any specific numeric type.
    Number(&'buf str),
    /// Ordered list of child elements.
    Array(Vec<Element<'buf>>),
    /// Ordered list of key-value fields.
    Object(Vec<Field<'buf>>),
}

impl fmt::Display for Value<'_> {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        match self {
            Self::Null => write!(f, "null"),
            Self::True => write!(f, "true"),
            Self::False => write!(f, "false"),
            Self::String(s) => write!(f, "{}", s.as_unescaped_str()),
            Self::Number(s) => write!(f, "{s}"),
            Self::Array(..) => f.write_str("[...]"),
            Self::Object(..) => f.write_str("{...}"),
        }
    }
}

/// Byte range of a JSON token within the source string.
#[derive(Copy, Clone, Debug, Default, Eq, PartialEq, Ord, PartialOrd)]
pub struct Span {
    /// Byte offset of the first byte of the token.
    pub start: u32,
    /// Byte offset one past the last byte of the token.
    pub end: u32,
}

impl Span {
    fn new(start: u32, end: u32) -> Self {
        Self { start, end }
    }
}

/// A file location expressed as line and column.
#[derive(Clone, Copy, Debug, PartialEq, Eq)]
pub struct Location {
    /// The line index is 0 based.
    pub line: u32,

    /// The col index is 0 based.
    pub col: u32,
}

impl From<(u32, u32)> for Location {
    fn from(value: (u32, u32)) -> Self {
        Self {
            line: value.0,
            col: value.1,
        }
    }
}

impl From<Location> for (u32, u32) {
    fn from(value: Location) -> Self {
        (value.line, value.col)
    }
}

impl PartialEq<(u32, u32)> for Location {
    fn eq(&self, other: &(u32, u32)) -> bool {
        self.line == other.0 && self.col == other.1
    }
}

impl fmt::Display for Location {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        write!(f, "{}:{}", self.line, self.col)
    }
}

/// Return the line and column indices of the end of the slice.
///
/// The line and column indices are zero based.
pub fn line_col(s: &str) -> Location {
    let mut chars = s.chars().rev();
    let mut line = 0_u32;
    let mut col = 0_u32;

    // The col only needs to be calculated on the final line so we iterate from the last char
    // back to the start of the line and then only continue to count the lines after that.
    //
    // This is less work than continuously counting chars from the front of the slice.
    for c in chars.by_ref() {
        // If the `&str` is multiline, we count the line and stop accumulating the col count too.
        if c == '\n' {
            let Some(n) = line.checked_add(1) else {
                break;
            };
            line = n;
            break;
        }
        let Some(n) = col.checked_add(1) else {
            break;
        };
        col = n;
    }

    // The col is now known, continue to the start of the str counting newlines as we go.
    for c in chars {
        if c == '\n' {
            let Some(n) = line.checked_add(1) else {
                break;
            };
            line = n;
        }
    }

    Location { line, col }
}

/// Unique sequential index of a JSON [`Element`] within a document.
///
/// Assigned depth-first by the parser. `Parser::alloc_id` uses `checked_add`
/// so the counter never wraps silently — overflow becomes `ParseError::TooLarge`.
#[derive(Copy, Clone, Debug, Eq, Hash, PartialEq, Ord, PartialOrd)]
pub struct ElemId(usize);

impl fmt::Display for ElemId {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        fmt::Display::fmt(&self.0, f)
    }
}

/// Records how one element was reached from its parent.
#[derive(Debug)]
enum PathEntry<'buf> {
    /// The root element; has no parent.
    Root,
    /// An object field; the element's path ends with a key segment.
    Field {
        /// Id of the parent object element.
        parent: ElemId,
        /// Key text borrowed from the source JSON, without surrounding quotes.
        key: RawStr<'buf>,
    },
    /// An array item; the element's path ends with an index segment.
    Item {
        /// Id of the parent array element.
        parent: ElemId,
        /// Zero-based position within the parent array.
        index: u32,
    },
}

/// Shared state carried by every [`Element`] produced from the same parse.
///
/// Wrapped in [`Rc`] so that each element can resolve its own path without
/// holding a live reference to the original [`Document`].
#[derive(Debug)]
struct DocumentInner<'buf> {
    /// The full source string; all element spans are relative to this.
    source: &'buf str,
    /// Parent-pointer table used to reconstruct element paths.
    paths: PathTable<'buf>,
}

/// A table recording the parentage of every [`Element`] produced by a parse.
#[derive(Debug, Default)]
struct PathTable<'buf> {
    /// The `entries` `Vec` is indexed using `ElemId`.
    entries: Vec<PathEntry<'buf>>,
}

impl<'buf> PathTable<'buf> {
    fn push(&mut self, entry: PathEntry<'buf>) {
        self.entries.push(entry);
    }

    /// Returns the chain of ids from the root down to `id`, both included.
    ///
    /// Cost is O(depth). Returns `None` if `id` is not in this table. The walk always
    /// terminates because a parent is allocated before its children, so every parent id
    /// is smaller than the id it was reached from.
    fn ancestry(&self, id: ElemId) -> Option<Vec<ElemId>> {
        let mut chain = vec![id];
        let mut current = id;

        loop {
            match self.entries.get(current.0)? {
                PathEntry::Root => break,
                PathEntry::Field { parent, key: _ } | PathEntry::Item { parent, index: _ } => {
                    chain.push(*parent);
                    current = *parent;
                }
            }
        }

        chain.reverse();

        Some(chain)
    }

    /// Build the full RFC 9535 path to `element` by walking the parent-pointer chain.
    ///
    /// Cost is O(depth) per path construction.
    ///
    /// NOTE: The `'buf` lifetime shared by `element` and `self` prevents cross-document
    /// misuse at compile time for documents with distinct source lifetimes.
    ///
    /// # Panics
    ///
    /// Panics if `element` was not produced by the same parse that created this table.
    fn path_of(&self, element: &Element<'buf>) -> Path {
        let mut entries: Vec<&PathEntry<'buf>> = Vec::new();
        let mut elem_id = element.id;

        // Walk back up the path chain.
        loop {
            let entry = self
                .entries
                .get(elem_id.0)
                .expect("ElemId always refers to a valid PathEntry");

            match entry {
                PathEntry::Root => {
                    entries.push(entry);
                    break;
                }
                PathEntry::Field { parent, key: _ } | PathEntry::Item { parent, index: _ } => {
                    entries.push(entry);
                    elem_id = *parent;
                }
            }
        }

        // Reverse the elements so we can walk forward along the chain.
        entries.reverse();

        let mut out = String::with_capacity(30);

        for entry in entries {
            let res = match entry {
                PathEntry::Root => write!(out, "$"),
                PathEntry::Field { parent: _, key } => {
                    write!(out, ".{}", key.as_unescaped_str())
                }
                PathEntry::Item { parent: _, index } => {
                    // Array indices use bracket notation per RFC 9535, e.g.
                    // `$.elements[0]`, rather than a dotted `.0` segment.
                    write!(out, "[{index}]")
                }
            };

            res.expect("Writing to a String can only fail if the system runs out of heap memory");
        }

        Path(out)
    }
}

#[derive(Clone, PartialOrd, Ord, PartialEq, Eq)]
/// The RFC 9535 path locating an [`Element`] in its [`Document`], such as
/// `$.charging_periods[0].dimensions`.
///
/// A warning is reported against a path, so this is what tells a caller which field a
/// warning is about.
pub struct Path(String);

impl Path {
    /// Consume the `Path` and return the underlying `String`.
    pub fn into_string(self) -> String {
        self.0
    }

    /// Return the path as a `str`.
    pub fn as_str(&self) -> &str {
        &self.0
    }

    /// Iterate the [`Component`]s of this path in order, skipping the `$` root.
    ///
    /// For example, `$.elements[0].id` yields `Member("elements")`,
    /// `Index("0")`, `Member("id")`. The root path `$` yields nothing.
    pub fn components(&self) -> Components<'_> {
        Components::over(&self.0)
    }
}

/// A single [`Path`] component: an object member or an array index.
#[derive(Clone, Copy, Debug, PartialEq, Eq)]
pub enum Component<'a> {
    /// An object member, the `name` in a `.name` segment.
    Member(&'a str),
    /// An array index, the decimal digits in a `[n]` segment.
    Index(&'a str),
}

/// Iterator over the [`Component`]s of a [`Path`]; see [`Path::components`].
#[derive(Clone, Debug)]
pub struct Components<'a> {
    rest: &'a str,
}

impl<'a> Components<'a> {
    /// Iterate the components of a raw `JSONPath` string, skipping a leading `$`.
    pub(crate) fn over(path: &'a str) -> Self {
        Self {
            rest: path.strip_prefix('$').unwrap_or(path),
        }
    }
}

impl<'a> Iterator for Components<'a> {
    type Item = Component<'a>;

    fn next(&mut self) -> Option<Self::Item> {
        if let Some(after) = self.rest.strip_prefix('.') {
            // `.name`: read up to the next segment delimiter.
            let end = after.find(['.', '[']).unwrap_or(after.len());
            let (name, tail) = after.split_at(end);
            self.rest = tail;
            Some(Component::Member(name))
        } else if let Some(after) = self.rest.strip_prefix('[') {
            // `[index]`: read up to the closing bracket.
            let end = after.find(']').unwrap_or(after.len());
            let (index, tail) = after.split_at(end);
            self.rest = tail.strip_prefix(']').unwrap_or(tail);
            Some(Component::Index(index))
        } else {
            // Empty (root) or malformed input: stop iterating.
            None
        }
    }
}

impl PartialEq<str> for Path {
    fn eq(&self, other: &str) -> bool {
        self.0 == other
    }
}

impl PartialEq<&str> for Path {
    fn eq(&self, other: &&str) -> bool {
        self.0 == *other
    }
}

impl fmt::Debug for Path {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        f.write_str(&self.0)
    }
}

impl fmt::Display for Path {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        fmt::Display::fmt(&self.0, f)
    }
}

/// Set of path with a common issue.
#[derive(Debug)]
pub struct PathSet<'set>(BTreeSet<&'set Path>);

impl<'set> PathSet<'set> {
    pub(crate) fn new(paths: BTreeSet<&'set Path>) -> Self {
        Self(paths)
    }

    /// Return the field paths as a `Vec` of `String`s.
    pub fn to_strings(&self) -> Vec<String> {
        self.0.iter().map(ToString::to_string).collect()
    }

    /// Return the field paths as a `Vec` of `String`s.
    pub fn into_strings(self) -> Vec<String> {
        self.0.into_iter().map(ToString::to_string).collect()
    }

    /// Return true if the list of unexpected fields is empty.
    pub fn is_empty(&self) -> bool {
        self.0.is_empty()
    }

    /// Return the number of unexpected fields.
    pub fn len(&self) -> usize {
        self.0.len()
    }

    /// Return an Iterator over the unexpected fields.
    pub fn iter(&self) -> btree_set::Iter<'_, &Path> {
        self.0.iter()
    }
}

impl<'set> IntoIterator for PathSet<'set> {
    type Item = &'set Path;

    type IntoIter = btree_set::IntoIter<&'set Path>;

    fn into_iter(self) -> Self::IntoIter {
        self.0.into_iter()
    }
}

impl<'a, 'set> IntoIterator for &'a PathSet<'set> {
    type Item = &'a &'set Path;

    type IntoIter = btree_set::Iter<'a, &'set Path>;

    fn into_iter(self) -> Self::IntoIter {
        self.0.iter()
    }
}

#[derive(Copy, Clone, Debug, Eq, PartialEq, Ord, PartialOrd)]
/// Which of the six JSON kinds a [`Value`] is, without its content.
///
/// A schema mismatch is reported in terms of these, saying which kind was expected and
/// which was found.
pub enum ValueKind {
    /// The JSON `null` literal.
    Null,
    /// The JSON `true` or `false` literal.
    Bool,
    /// A JSON number.
    Number,
    /// A JSON string.
    String,
    /// A JSON array.
    Array,
    /// A JSON object.
    Object,
}

impl fmt::Display for ValueKind {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        match self {
            ValueKind::Null => write!(f, "null"),
            ValueKind::Bool => write!(f, "bool"),
            ValueKind::Number => write!(f, "number"),
            ValueKind::String => write!(f, "string"),
            ValueKind::Array => write!(f, "array"),
            ValueKind::Object => write!(f, "object"),
        }
    }
}

impl<'buf> Value<'buf> {
    /// Returns which JSON kind this value is, discarding its content.
    pub fn kind(&self) -> ValueKind {
        match self {
            Value::Null => ValueKind::Null,
            Value::True | Value::False => ValueKind::Bool,
            Value::String(_) => ValueKind::String,
            Value::Number(_) => ValueKind::Number,
            Value::Array(_) => ValueKind::Array,
            Value::Object(_) => ValueKind::Object,
        }
    }

    /// Return true if the `Value` is null.
    pub fn is_null(&self) -> bool {
        matches!(self, Value::Null)
    }

    /// Return true if the `Value` is an array.
    pub fn is_array(&self) -> bool {
        matches!(self, Value::Array(..))
    }

    /// Return true if the `Value` is an object.
    pub fn is_object(&self) -> bool {
        matches!(self, Value::Object(..))
    }

    /// Return true if the `Value` can't contain child elements.
    pub fn is_scalar(&self) -> bool {
        matches!(
            self,
            Value::Null | Value::True | Value::False | Value::String(_) | Value::Number(_)
        )
    }

    /// Returns the items if this is a JSON array, `None` otherwise.
    pub fn as_array(&self) -> Option<&[Element<'buf>]> {
        if let Value::Array(elems) = self {
            Some(elems)
        } else {
            None
        }
    }

    /// Returns the number as it was written in the source if this is a JSON number,
    /// `None` otherwise.
    pub fn as_number(&self) -> Option<&str> {
        if let Value::Number(s) = self {
            Some(s)
        } else {
            None
        }
    }

    /// Return `Some(&str)` if the `Value` is a `String`.
    pub fn to_raw_str(&self) -> Option<RawStr<'buf>> {
        if let Value::String(s) = self {
            Some(*s)
        } else {
            None
        }
    }

    /// Return `Some(&[Field])` if the `Value` is a `Object`.
    pub fn as_object_fields(&self) -> Option<&[Field<'buf>]> {
        if let Value::Object(fields) = self {
            Some(fields)
        } else {
            None
        }
    }
}

/// An object field; upholds the invariant that the inner [`Element`]'s path ends with a key.
#[derive(Clone, Debug, Eq, PartialEq)]
pub struct Field<'buf> {
    /// Span of the key token, including surrounding `"` delimiters.
    key_span: Span,
    /// The value element; its path ends with the key from `key_span`.
    element: Element<'buf>,
}

impl<'buf> Field<'buf> {
    /// Consume the `Field` and return the inner `Element`.
    pub fn into_element(self) -> Element<'buf> {
        self.element
    }

    /// Return the inner `Element`.
    pub fn element(&self) -> &Element<'buf> {
        &self.element
    }

    /// Returns the span covering the field's quoted key alone.
    pub fn key_span(&self) -> Span {
        self.key_span
    }

    /// Returns the span covering `"key": value` plus any trailing comma and the whitespace
    /// after it, so the span ends where the next field begins.
    ///
    /// When there is no trailing comma (the last field), `full_span()` covers
    /// `"key": value` only.
    ///
    /// As with [`Element::full_span`], removing a field is not simply a matter of erasing
    /// this span; use [`Document::removal_spans`]. Erasing the `full_span` of a last field
    /// leaves a dangling comma on the field before it, and a field's value cannot be
    /// erased on its own without leaving a key with nothing after the colon.
    pub fn full_span(&self) -> Span {
        Span {
            start: self.key_span.start,
            end: self.element.full_span_end,
        }
    }

    /// Returns the key text without surrounding `"` delimiters.
    #[expect(
        clippy::arithmetic_side_effects,
        reason = "key_span always spans a quoted string, so +1/-1 to strip the surrounding quote bytes is safe"
    )]
    #[expect(
        clippy::string_slice,
        reason = "key_span is produced by the parser from the same source; +1/-1 strips the ASCII quote bytes"
    )]
    #[expect(
        clippy::as_conversions,
        reason = "The index is guaranteed within bounds by the parser"
    )]
    pub fn key(&self) -> RawStr<'buf> {
        let src = self.element.source();
        let s = &src[self.key_span.start as usize + 1..self.key_span.end as usize - 1];
        RawStr::from_str(s)
    }

    /// Returns the slice of the source JSON spanning `"key": value`.
    #[expect(
        clippy::string_slice,
        reason = "spans are produced by the parser from the same source, so slices are always valid"
    )]
    #[expect(
        clippy::as_conversions,
        reason = "The index is guaranteed within bounds by the parser"
    )]
    pub fn source_json(&self) -> &'buf str {
        let src = self.element.source();
        &src[self.key_span.start as usize..self.element.span.end as usize]
    }
}

/// An object's fields keyed by their undecoded name, for looking a field up by key.
pub type RawMap<'buf> = BTreeMap<RawStr<'buf>, Element<'buf>>;
/// A [`RawMap`] that borrows its elements instead of owning them.
pub type RawRefMap<'a, 'buf> = BTreeMap<RawStr<'buf>, &'a Element<'buf>>;

#[expect(dead_code, reason = "pending use in `tariff::lint`")]
pub(crate) trait FieldsIntoExt<'buf> {
    fn into_map(self) -> RawMap<'buf>;
}

impl<'buf> FieldsIntoExt<'buf> for Vec<Field<'buf>> {
    fn into_map(self) -> RawMap<'buf> {
        self.into_iter()
            .map(|field| (field.key(), field.into_element()))
            .collect()
    }
}

/// A `&str` with surrounding quotes removed; escape sequences are not decoded.
#[derive(Copy, Clone, Debug, Eq, PartialEq, Ord, PartialOrd)]
pub struct RawStr<'buf>(&'buf str);

/// Impl `Borrow` so `RawStr` plays well with hashed collections.
impl Borrow<str> for RawStr<'_> {
    fn borrow(&self) -> &str {
        self.0
    }
}

/// Impl `Borrow` so `RawStr` plays well with hashed collections.
impl Borrow<str> for &RawStr<'_> {
    fn borrow(&self) -> &str {
        self.0
    }
}

impl<'buf> RawStr<'buf> {
    fn from_str(source: &'buf str) -> Self {
        Self(source)
    }

    /// Compare `other` against this raw `&str`, decoding any JSON escape
    /// sequences in the `&str` on the fly without allocating.
    ///
    /// Returns `Ok(true)`/`Ok(false)` for the comparison, or `Err` if the key
    /// contains a decoding problem (an invalid escape or a control character) at
    /// or before the first differing character.
    pub fn eq_escape_aware(&self, other: &str) -> Result<bool, decode::Warning> {
        decode::eq(self.0, other)
    }

    /// Compare this raw `&str` against a list of `&str`s, decoding any JSON escape
    /// sequences in the `&str` on the fly without allocating.
    ///
    /// Returns true if any of the `other` `&str`s match self.
    pub fn eq_any_escape_aware(&self, other: &[&str]) -> bool {
        other
            .iter()
            .any(|s| decode::eq(self.0, s).ok().unwrap_or(false))
    }

    /// Like [`RawStr::eq_any_escape_aware`], but compares ASCII letters case-insensitively.
    pub fn eq_any_escape_aware_ignore_ascii_case(&self, other: &[&str]) -> bool {
        other.iter().any(|s| {
            decode::eq_ignore_ascii_case(self.0, s)
                .ok()
                .unwrap_or(false)
        })
    }

    /// Return the raw unescaped `&str`.
    pub fn as_unescaped_str(&self) -> &'buf str {
        self.0
    }

    /// Return the `&str` with all escapes decoded.
    pub fn decode_escapes(&self) -> CaveatDeferred<Cow<'_, str>, decode::Warning> {
        decode::from_raw(self.0)
    }

    /// Return a `&str` marked as either having escapes or not.
    pub fn has_escapes(&self, elem: &Element<'buf>) -> Caveat<PendingStr<'buf>, decode::Warning> {
        decode::analyze(self.0, elem)
    }

    /// Report whether the string contains escape sequences and whether its decoded form
    /// contains non-printable ASCII, in a single pass without allocating.
    ///
    /// This combines the escape test of [`RawStr::has_escapes`] with the printability
    /// check a caller would otherwise run on a string returned from [`RawStr::decode_escapes`].
    pub fn lexical_issues(&self) -> LexicalIssues {
        decode::lexical_issues(self.0)
    }
}

/// The lexical issues a [`RawStr`] may contain, discovered in a single pass by
/// [`RawStr::lexical_issues`].
#[derive(Clone, Copy, Debug, Eq, PartialEq)]
pub struct LexicalIssues {
    /// The raw string contains one or more JSON escape sequences.
    pub escapes: bool,

    /// The decoded string contains non-printable ASCII: an ASCII control character or
    /// ASCII whitespace.
    pub non_printable_ascii: bool,
}

/// Marks a `&str` as having escapes or not.
pub enum PendingStr<'buf> {
    /// The `&str` has no escapes and can be used as is.
    NoEscapes(&'buf str),

    /// The `&str` has escape chars and needs to be unescaped before trying to parse into another form.
    HasEscapes(EscapeStr<'buf>),
}

/// A `&str` with escape chars.
pub struct EscapeStr<'buf>(&'buf str);

impl<'buf> EscapeStr<'buf> {
    /// Decode the escape sequences, reporting each one that is invalid as a
    /// [`decode::Warning`] rather than failing.
    pub fn decode_escapes(&self) -> CaveatDeferred<Cow<'buf, str>, decode::Warning> {
        decode::from_raw(self.0)
    }

    /// Consume the `EscapeStr` and return the raw bytes as a str.
    pub fn into_raw(self) -> &'buf str {
        self.0
    }
}