faucet_source_postgres_cdc/pgoutput/

values.rs

//! Postgres type OID -> JSON Value mapping for text-mode tuple cells.
//!
//! Reference: <https://github.com/postgres/postgres/blob/master/src/include/catalog/pg_type.dat>
//! Only the OIDs that ship with every Postgres install are special-cased;
//! anything else falls back to a JSON string.

use base64::Engine;
use faucet_core::FaucetError;
use serde_json::Value;

// Selected pg_type built-in OIDs.
const OID_BOOL: u32 = 16;
const OID_BYTEA: u32 = 17;
const OID_INT2: u32 = 21;
const OID_INT4: u32 = 23;
const OID_INT8: u32 = 20;
const OID_FLOAT4: u32 = 700;
const OID_FLOAT8: u32 = 701;
const OID_NUMERIC: u32 = 1700;
const OID_JSON: u32 = 114;
const OID_JSONB: u32 = 3802;
// Types passed through as their Postgres ISO text form (stable under the
// default `DateStyle ISO` / `IntervalStyle`): date/time/timestamp/
// timestamptz/uuid/interval. Downstream consumers don't agree on a single
// binary encoding, and the text form is already ISO-8601-ish, so we keep
// the string verbatim rather than risk a lossy reformat.

/// Map a built-in Postgres *array* type OID to its element type OID, so a
/// `{...}` array literal can be decoded element-by-element instead of being
/// emitted as one opaque string (e.g. `int4[]` → `"{1,2,3}"`) (#78/#45).
/// Returns `None` for non-array OIDs.
fn array_element_oid(array_oid: u32) -> Option<u32> {
    Some(match array_oid {
        1000 => OID_BOOL,    // _bool
        1001 => OID_BYTEA,   // _bytea
        1005 => OID_INT2,    // _int2
        1007 => OID_INT4,    // _int4
        1016 => OID_INT8,    // _int8
        1021 => OID_FLOAT4,  // _float4
        1022 => OID_FLOAT8,  // _float8
        1231 => OID_NUMERIC, // _numeric
        199 => OID_JSON,     // _json
        3807 => OID_JSONB,   // _jsonb
        1009 => 25,          // _text → text
        1015 => 1043,        // _varchar → varchar
        1014 => 1042,        // _bpchar → bpchar
        2951 => 2950,        // _uuid → uuid
        1115 => 1114,        // _timestamp
        1185 => 1184,        // _timestamptz
        1182 => 1082,        // _date
        1183 => 1083,        // _time
        _ => return None,
    })
}

/// Decode a text-encoded value with the given column type OID into JSON.
///
/// Unknown OIDs and decode failures both fall back to wrapping the raw text
/// in a JSON string — this is the safest default for a generic CDC
/// connector. We never panic on bad data; the only way this returns `Err` is
/// if a structurally promised invariant is violated (e.g. `OID_BYTEA` text
/// that doesn't start with `\x`).
pub fn text_to_json(type_oid: u32, text: &str) -> Result<Value, FaucetError> {
    Ok(match type_oid {
        OID_BOOL => match text {
            "t" => Value::Bool(true),
            "f" => Value::Bool(false),
            other => {
                return Err(FaucetError::Source(format!(
                    "pgoutput: bool column has non-t/f text {other:?}"
                )));
            }
        },
        OID_INT2 | OID_INT4 | OID_INT8 => {
            let n: i64 = text.parse().map_err(|e| {
                FaucetError::Source(format!(
                    "pgoutput: int (oid={type_oid}) parse {text:?}: {e}"
                ))
            })?;
            Value::from(n)
        }
        OID_FLOAT4 | OID_FLOAT8 => match text {
            // JSON has no NaN/Inf. Preserve them as strings so the
            // non-finite value is distinguishable from a SQL NULL, which
            // also maps to `Value::Null` (#78/#45).
            "NaN" => Value::String("NaN".into()),
            "Infinity" => Value::String("Infinity".into()),
            "-Infinity" => Value::String("-Infinity".into()),
            other => {
                let n: f64 = other.parse().map_err(|e| {
                    FaucetError::Source(format!("pgoutput: float parse {text:?}: {e}"))
                })?;
                serde_json::Number::from_f64(n)
                    .map(Value::Number)
                    .unwrap_or(Value::Null)
            }
        },
        OID_NUMERIC => Value::String(text.into()),
        OID_BYTEA => {
            let stripped = text.strip_prefix("\\x").ok_or_else(|| {
                FaucetError::Source(format!(
                    "pgoutput: bytea text {text:?} missing '\\x' prefix"
                ))
            })?;
            let bytes = hex_decode(stripped)?;
            Value::String(base64::engine::general_purpose::STANDARD.encode(bytes))
        }
        OID_JSON | OID_JSONB => serde_json::from_str(text).map_err(|e| {
            FaucetError::Source(format!("pgoutput: json/jsonb parse {text:?}: {e}"))
        })?,
        other => {
            // One-dimensional arrays of a known scalar element type decode
            // into a JSON array; everything else (incl. multi-dimensional
            // arrays, ranges, composites, enums, and unmapped scalars) falls
            // back to the raw Postgres text string (#78/#45).
            if let Some(elem_oid) = array_element_oid(other)
                && let Some(elements) = parse_pg_array(text)
            {
                let mut out = Vec::with_capacity(elements.len());
                for elem in elements {
                    match elem {
                        Some(s) => out.push(text_to_json(elem_oid, &s)?),
                        None => out.push(Value::Null),
                    }
                }
                Value::Array(out)
            } else {
                Value::String(text.into())
            }
        }
    })
}

/// Parse a one-dimensional Postgres array literal (`{a,b,"c,d",NULL}`) into a
/// vector of element texts, where `None` marks an unquoted `NULL`. Returns
/// `None` (caller falls back to the raw string) for anything this simple
/// parser doesn't handle: a non-`{...}`-delimited input or a nested
/// (multi-dimensional) array.
fn parse_pg_array(text: &str) -> Option<Vec<Option<String>>> {
    let bytes = text.as_bytes();
    if bytes.first() != Some(&b'{') || bytes.last() != Some(&b'}') {
        return None;
    }
    let inner = &text[1..text.len() - 1];
    if inner.is_empty() {
        return Some(Vec::new());
    }

    let mut out: Vec<Option<String>> = Vec::new();
    let mut cur = String::new();
    let mut in_quotes = false;
    let mut quoted = false; // current element was quoted (so "NULL" != NULL)
    let mut chars = inner.chars().peekable();

    while let Some(c) = chars.next() {
        if in_quotes {
            match c {
                '\\' => {
                    // Escaped next char (\" or \\) is taken literally.
                    if let Some(next) = chars.next() {
                        cur.push(next);
                    }
                }
                '"' => in_quotes = false,
                _ => cur.push(c),
            }
            continue;
        }
        match c {
            '"' => {
                in_quotes = true;
                quoted = true;
            }
            // A nested array — bail out, let the caller keep the raw string.
            '{' => return None,
            ',' => {
                out.push(finish_array_element(&cur, quoted));
                cur.clear();
                quoted = false;
            }
            _ => cur.push(c),
        }
    }
    if in_quotes {
        return None; // unterminated quote — malformed, fall back to string
    }
    out.push(finish_array_element(&cur, quoted));
    Some(out)
}

/// An unquoted `NULL` token is a SQL NULL; anything else (incl. a quoted
/// `"NULL"`) is the literal text.
fn finish_array_element(raw: &str, quoted: bool) -> Option<String> {
    if !quoted && raw == "NULL" {
        None
    } else {
        Some(raw.to_owned())
    }
}

fn hex_decode(s: &str) -> Result<Vec<u8>, FaucetError> {
    if !s.len().is_multiple_of(2) {
        return Err(FaucetError::Source(format!(
            "pgoutput: bytea hex has odd length: {s:?}"
        )));
    }
    let mut out = Vec::with_capacity(s.len() / 2);
    for i in (0..s.len()).step_by(2) {
        out.push(
            u8::from_str_radix(&s[i..i + 2], 16)
                .map_err(|e| FaucetError::Source(format!("pgoutput: bytea hex {s:?}: {e}")))?,
        );
    }
    Ok(out)
}

#[cfg(test)]
mod tests {
    use super::*;
    use serde_json::json;

    #[test]
    fn bool_t_and_f() {
        assert_eq!(text_to_json(OID_BOOL, "t").unwrap(), json!(true));
        assert_eq!(text_to_json(OID_BOOL, "f").unwrap(), json!(false));
        assert!(text_to_json(OID_BOOL, "yes").is_err());
    }

    #[test]
    fn integer_types() {
        assert_eq!(text_to_json(OID_INT2, "32000").unwrap(), json!(32000));
        assert_eq!(text_to_json(OID_INT4, "-1").unwrap(), json!(-1));
        assert_eq!(
            text_to_json(OID_INT8, "9223372036854775807").unwrap(),
            json!(9223372036854775807_i64)
        );
        assert!(text_to_json(OID_INT4, "abc").is_err());
    }

    #[test]
    fn floats() {
        assert_eq!(text_to_json(OID_FLOAT8, "3.5").unwrap(), json!(3.5));
        // NaN/Inf are preserved as strings (distinct from a NULL → Value::Null).
        assert_eq!(text_to_json(OID_FLOAT8, "NaN").unwrap(), json!("NaN"));
        assert_eq!(
            text_to_json(OID_FLOAT4, "Infinity").unwrap(),
            json!("Infinity")
        );
        assert_eq!(
            text_to_json(OID_FLOAT8, "-Infinity").unwrap(),
            json!("-Infinity")
        );
    }

    #[test]
    fn int_array_decodes_to_json_array() {
        assert_eq!(text_to_json(1007, "{1,2,3}").unwrap(), json!([1, 2, 3]));
        assert_eq!(text_to_json(1007, "{}").unwrap(), json!([]));
        assert_eq!(text_to_json(1016, "{-9,0,9}").unwrap(), json!([-9, 0, 9]));
    }

    #[test]
    fn text_array_handles_quotes_nulls_and_commas() {
        assert_eq!(
            text_to_json(1009, r#"{a,"b,c",NULL,"NULL"}"#).unwrap(),
            json!(["a", "b,c", null, "NULL"])
        );
        // Escaped quote inside a quoted element.
        assert_eq!(
            text_to_json(1009, r#"{"he said \"hi\""}"#).unwrap(),
            json!(["he said \"hi\""])
        );
    }

    #[test]
    fn bool_array_decodes() {
        assert_eq!(
            text_to_json(1000, "{t,f,t}").unwrap(),
            json!([true, false, true])
        );
    }

    #[test]
    fn multidimensional_array_falls_back_to_string() {
        assert_eq!(
            text_to_json(1007, "{{1,2},{3,4}}").unwrap(),
            json!("{{1,2},{3,4}}")
        );
    }

    #[test]
    fn numeric_kept_as_string() {
        assert_eq!(
            text_to_json(OID_NUMERIC, "12345.67890").unwrap(),
            json!("12345.67890")
        );
    }

    #[test]
    fn bytea_base64() {
        // \xDEADBEEF -> base64 "3q2+7w=="
        assert_eq!(
            text_to_json(OID_BYTEA, "\\xDEADBEEF").unwrap(),
            json!("3q2+7w==")
        );
        assert!(text_to_json(OID_BYTEA, "deadbeef").is_err()); // missing \x
        assert!(text_to_json(OID_BYTEA, "\\xZZ").is_err()); // not hex
    }

    #[test]
    fn json_columns_parsed() {
        assert_eq!(
            text_to_json(OID_JSON, r#"{"a":1}"#).unwrap(),
            json!({"a": 1})
        );
        assert_eq!(
            text_to_json(OID_JSONB, r#"[1,2,3]"#).unwrap(),
            json!([1, 2, 3])
        );
    }

    #[test]
    fn unknown_oid_falls_back_to_string() {
        assert_eq!(
            text_to_json(99999, "2026-05-17 12:34:56+00").unwrap(),
            json!("2026-05-17 12:34:56+00")
        );
    }
}