reddb-io-rql 1.28.21

//! Parser error types

use std::fmt::{self, Write};

use crate::lexer::{LexerError, LexerLimitHit, Position, Token};

/// Parse error
#[derive(Debug, Clone)]
pub struct ParseError {
    /// Error message
    pub message: String,
    /// Position where error occurred
    pub position: Position,
    /// Expected tokens (for better error messages)
    pub expected: Vec<String>,
    /// Optional structured kind for hardening / DoS errors
    pub kind: ParseErrorKind,
}

/// Categorical kind for a parse error.
///
/// Most parse errors are plain `Syntax` failures; the variants
/// below carry structured information for the parser-hardening
/// layer (issue #87) so callers can distinguish DoS-style refusals
/// from grammar errors without string matching.
#[derive(Debug, Clone, PartialEq, Eq)]
pub enum ParseErrorKind {
    /// Generic syntax / semantic error.
    Syntax,
    /// Recursion-depth limit exceeded during parsing.
    DepthLimit {
        limit_name: &'static str,
        value: usize,
    },
    /// Input larger than the configured byte cap.
    InputTooLarge {
        limit_name: &'static str,
        value: usize,
    },
    /// Identifier longer than the configured character cap.
    IdentifierTooLong {
        limit_name: &'static str,
        value: usize,
    },
    /// Parser consumed more tokens than the configured cap.
    TokenLimit {
        limit_name: &'static str,
        value: usize,
    },
    /// A literal value (integer / float) parsed cleanly but lies
    /// outside the semantic range expected for its slot — e.g.
    /// `MAX_SIZE 0`, `lat = 91.0`, `K = 0`, or a negative integer
    /// where a positive one is required. The structured payload lets
    /// the snapshot/property harness distinguish these from generic
    /// syntax errors without string matching.
    ValueOutOfRange {
        /// Stable slot name, e.g. `"MAX_SIZE"`, `"lat"`, `"radius"`.
        field: &'static str,
        /// Free-text constraint, e.g. `"must be > 0"`,
        /// `"must be in -90.0..=90.0"`.
        constraint: &'static str,
    },
    /// The lexer recognized this token, but the parser does not support
    /// it in the current grammar position.
    UnsupportedToken { token: String },
}

impl ParseError {
    /// Create a new parse error
    pub fn new(message: impl Into<String>, position: Position) -> Self {
        Self {
            message: message.into(),
            position,
            expected: Vec::new(),
            kind: ParseErrorKind::Syntax,
        }
    }

    /// Create error with expected tokens
    ///
    /// `found` is rendered through [`SafeTokenDisplay`] so caller-controlled
    /// bytes inside `Token::Ident` / `Token::String` / `Token::JsonLiteral` /
    /// `Token::Float` / `Token::Integer` payloads are escaped via Rust's
    /// `escape_debug` rules (CR / LF / NUL / quote bytes become `\n`,
    /// `\r`, `\0`, `\"`, …). Static keyword and punctuation arms keep their
    /// existing UPPER-CASE rendering so error messages and snapshot tests
    /// stay readable. This prevents F-05 smuggling through the downstream
    /// JSON / audit / log / gRPC sinks that embed `ParseError::message`.
    pub fn expected(expected: Vec<&str>, found: &Token, position: Position) -> Self {
        Self {
            message: format!("Unexpected token: {}", SafeTokenDisplay(found)),
            position,
            expected: expected.into_iter().map(|s| s.to_string()).collect(),
            kind: ParseErrorKind::Syntax,
        }
    }

    /// Create an error when a lexer-known keyword appears in a parser
    /// position where that keyword has no supported production.
    pub fn unsupported_recognized_token(found: &Token, position: Position) -> Option<Self> {
        let token = recognized_keyword_name(found)?;
        Some(Self {
            message: format!("token {token} is recognized but not supported in this position"),
            position,
            expected: Vec::new(),
            kind: ParseErrorKind::UnsupportedToken { token },
        })
    }

    /// Recursion depth limit hit. The structured `kind` carries the
    /// name + numeric value so the snapshot/property harness can
    /// pattern-match without string slicing.
    pub fn depth_limit(limit_name: &'static str, value: usize, position: Position) -> Self {
        Self {
            message: format!(
                "recursion depth limit exceeded ({} = {})",
                limit_name, value
            ),
            position,
            expected: Vec::new(),
            kind: ParseErrorKind::DepthLimit { limit_name, value },
        }
    }

    /// Input bytes exceeded the configured cap.
    pub fn input_too_large(limit_name: &'static str, value: usize, position: Position) -> Self {
        Self {
            message: format!(
                "input exceeds maximum size ({} = {} bytes)",
                limit_name, value
            ),
            position,
            expected: Vec::new(),
            kind: ParseErrorKind::InputTooLarge { limit_name, value },
        }
    }

    /// Identifier exceeded the configured character cap.
    pub fn identifier_too_long(limit_name: &'static str, value: usize, position: Position) -> Self {
        Self {
            message: format!(
                "identifier exceeds maximum length ({} = {} chars)",
                limit_name, value
            ),
            position,
            expected: Vec::new(),
            kind: ParseErrorKind::IdentifierTooLong { limit_name, value },
        }
    }

    /// Token budget exceeded during parsing.
    pub fn token_limit(limit_name: &'static str, value: usize, position: Position) -> Self {
        Self {
            message: format!("parser token limit exceeded ({} = {})", limit_name, value),
            position,
            expected: Vec::new(),
            kind: ParseErrorKind::TokenLimit { limit_name, value },
        }
    }

    /// A literal value lies outside the allowed range for its slot.
    /// The free-text `constraint` is included verbatim in the message
    /// so callers can render a single line without re-formatting.
    pub fn value_out_of_range(
        field: &'static str,
        constraint: &'static str,
        position: Position,
    ) -> Self {
        Self {
            message: format!("{} {}", field, constraint),
            position,
            expected: Vec::new(),
            kind: ParseErrorKind::ValueOutOfRange { field, constraint },
        }
    }

    /// Didactic (#1704): a document-model newcomer wrote the Postgres
    /// `CREATE TABLE <name> DOCUMENT (…)` form. Documents are schemaless
    /// and have no column list — the native form is `CREATE DOCUMENT`.
    pub fn document_create_table(position: Position) -> Self {
        Self {
            message: "documents are schemaless: write `CREATE DOCUMENT <name>` \
                      (no column list) instead of `CREATE TABLE <name> DOCUMENT (…)` — \
                      top-level body fields auto-flatten into queryable columns"
                .to_string(),
            position,
            expected: Vec::new(),
            kind: ParseErrorKind::Syntax,
        }
    }

    /// Didactic (#1704): a document-model newcomer used the Postgres JSON
    /// operators `->` / `->>` in a query. The native idiom is the dotted
    /// path. (Graph-mode `->` / `<-` traversal is parsed elsewhere and is
    /// unaffected.)
    pub fn json_arrow_operator(position: Position) -> Self {
        Self {
            message: "the `->` / `->>` JSON operators are not supported: use the \
                      dotted-path idiom instead (e.g. `body.details.ip` rather than \
                      `body->'details'->>'ip'`)"
                .to_string(),
            position,
            expected: Vec::new(),
            kind: ParseErrorKind::Syntax,
        }
    }

    /// Didactic (#1709 / ADR 0067): the document INSERT column list is
    /// removed. Documents are written with a bare inline JSON literal, so
    /// the old ceremonial `(body)` column list no longer exists.
    pub fn document_insert_column_list(position: Position) -> Self {
        Self {
            message: "the document INSERT column list is removed: write \
                      `INSERT INTO <collection> DOCUMENT VALUES ({\"level\": \"info\"})` — \
                      a bare inline JSON literal with no `(body)` column list"
                .to_string(),
            position,
            expected: Vec::new(),
            kind: ParseErrorKind::Syntax,
        }
    }

    /// Didactic (#1709 / ADR 0067): legacy `_ttl` / `_ttl_ms` /
    /// `_expires_at` document INSERT columns are removed; expiry is set
    /// with the `WITH TTL` clause instead.
    pub fn document_insert_ttl_column(position: Position) -> Self {
        Self {
            message: "legacy `_ttl` metadata columns are removed from document INSERT: \
                      set expiry with the `WITH TTL <duration>` clause (e.g. \
                      `INSERT INTO events DOCUMENT VALUES ({\"level\": \"info\"}) WITH TTL 30 m`)"
                .to_string(),
            position,
            expected: Vec::new(),
            kind: ParseErrorKind::Syntax,
        }
    }

    /// Didactic (#1709 / ADR 0067): the quoted-string body coercion
    /// (`VALUES ('{…}')`) is removed. JSON is written as an inline literal;
    /// `JSON_PARSE(<expr>)` converts a runtime string.
    pub fn document_insert_quoted_body(position: Position) -> Self {
        Self {
            message: "document INSERT no longer coerces a quoted string to JSON: write the \
                      body as an inline JSON literal — \
                      `INSERT INTO events DOCUMENT VALUES ({\"level\": \"info\"})` — or wrap a \
                      runtime string with `JSON_PARSE(<expr>)`"
                .to_string(),
            position,
            expected: Vec::new(),
            kind: ParseErrorKind::Syntax,
        }
    }
}

fn recognized_keyword_name(token: &Token) -> Option<String> {
    match token {
        Token::String(_)
        | Token::Integer(_)
        | Token::Float(_)
        | Token::JsonLiteral(_)
        | Token::Ident(_)
        | Token::QuotedIdent(_)
        | Token::Eq
        | Token::Ne
        | Token::Lt
        | Token::Le
        | Token::Gt
        | Token::Ge
        | Token::Plus
        | Token::Minus
        | Token::Star
        | Token::Slash
        | Token::Percent
        | Token::LParen
        | Token::RParen
        | Token::LBracket
        | Token::RBracket
        | Token::LBrace
        | Token::RBrace
        | Token::Comma
        | Token::Dot
        | Token::Colon
        | Token::Semi
        | Token::Dollar
        | Token::Question
        | Token::Arrow
        | Token::ArrowLeft
        | Token::Dash
        | Token::DotDot
        | Token::Pipe
        | Token::DoublePipe
        | Token::Eof => None,
        other => Some(SafeTokenDisplay(other).to_string()),
    }
}

impl fmt::Display for ParseError {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        write!(f, "Parse error at {}: {}", self.position, self.message)?;
        if !self.expected.is_empty() {
            write!(f, " (expected: {})", self.expected.join(", "))?;
        }
        Ok(())
    }
}

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

/// `Display` adapter that emits a `Token` while escaping the
/// caller-controlled byte payload of `Ident` / `String` / `JsonLiteral` /
/// `Integer` / `Float` arms.
///
/// F-05 (serialization-boundary audit, 2026-05-06): SQL parser error
/// messages flow into JSON HTTP bodies, JSONL audit rows, gRPC
/// `Status::message`, PG3 `ErrorResponse`, and `tracing::warn!` log
/// lines. The default `Token` Display arms emit raw user bytes for
/// `Token::Ident("foo\nbar")` etc., which lets a tenant smuggle CR /
/// LF / NUL / quote bytes through every downstream sink at once.
///
/// This adapter renders user-controlled arms via `escape_debug` (the
/// same rules `{:?}` applies to a `&str`) and leaves keyword /
/// punctuation arms untouched so existing snapshot tests and operator
/// log readability are preserved.
pub struct SafeTokenDisplay<'a>(pub &'a Token);

impl fmt::Display for SafeTokenDisplay<'_> {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        match self.0 {
            // User-controlled byte payloads. Render via `escape_debug`
            // so embedded CR / LF / NUL / quote bytes do not reach
            // downstream serialization sinks unescaped.
            Token::Ident(s) | Token::QuotedIdent(s) => write_escaped(f, s),
            Token::String(s) => {
                f.write_str("'")?;
                write_escaped(f, s)?;
                f.write_str("'")
            }
            Token::JsonLiteral(s) => write_escaped(f, s),
            // Numeric tokens come straight from the lexer; their
            // canonical Display form is bounded ASCII, but the lexer
            // builds them via `to_string` so they cannot carry control
            // bytes. Pass through Display.
            Token::Integer(_) | Token::Float(_) => fmt::Display::fmt(self.0, f),
            // Static keyword / punctuation arms — fall back to the
            // existing Display output verbatim.
            other => fmt::Display::fmt(other, f),
        }
    }
}

fn write_escaped(f: &mut fmt::Formatter<'_>, s: &str) -> fmt::Result {
    for ch in s.chars() {
        // `escape_debug` matches Rust's Debug rules: ASCII control
        // bytes become `\n`, `\r`, `\0`, `\t`, …; non-ASCII printable
        // characters pass through; backslash and double-quote are
        // escaped.
        for esc in ch.escape_debug() {
            f.write_char(esc)?;
        }
    }
    Ok(())
}

impl From<LexerError> for ParseError {
    fn from(e: LexerError) -> Self {
        let kind = match &e.limit_hit {
            Some(LexerLimitHit::IdentifierTooLong { limit_name, value }) => {
                ParseErrorKind::IdentifierTooLong {
                    limit_name,
                    value: *value,
                }
            }
            None => ParseErrorKind::Syntax,
        };
        ParseError {
            message: e.message,
            position: e.position,
            expected: Vec::new(),
            kind,
        }
    }
}