Supported Cypher

• Clauses: MATCH, OPTIONAL MATCH, WHERE, CREATE, MERGE (+ ON CREATE SET / ON MATCH SET), SET (n.p = v, n = {…}, n += {…}, n:Label), REMOVE, DELETE, DETACH DELETE, UNWIND, WITH, RETURN, DISTINCT, ORDER BY, SKIP, LIMIT, UNION [ALL], FOREACH (x IN list | updates), LOAD CSV [WITH HEADERS] FROM 'file:///data.csv' AS row [FIELDTERMINATOR ';'] (files inside the import directory only: Database::set_import_dir / DatabaseManager::set_import_dir, --import-dir for the server, the current directory by default in the CLI; each record is a list of strings, or a map by header; an unquoted empty field is null, "" an empty string), CALL { … } subqueries (run once per row; import outer variables with CALL (a, b) { … }, CALL (*) { … } or a leading WITH a, b; they may change the graph, and a subquery without RETURN keeps each row once)

• Procedures: CALL db.labels() YIELD label AS l WHERE l STARTS WITH 'P', anywhere in a query (one call per row; YIELD is required there). A query that is only a CALL returns every output, and may leave out YIELD and the parentheses (CALL db.labels; arguments then come from the parameters of the same names). Built in: db.labels(), db.relationshipTypes(), db.propertyKeys(), db.indexes(), db.constraints(), db.awaitIndexes(), dbms.procedures(), dbms.components(); SHOW PROCEDURES lists them. Applications add read-only procedures in Rust with Database::register_procedure (or DatabaseManager::register_procedure for every database):

  use graphbuff::{ProcedureSignature, Value};
  db.register_procedure(
      ProcedureSignature::new("my.range", ["n"], ["i"]).description("0 .. n-1"),
      |_ctx, args| {
          let n = match args[0] { Value::Int(n) => n, _ => 0 };
          Ok((0..n).map(|i| vec![Value::Int(i)]).collect())
      },
  )?;
  

• Schema: CREATE [RANGE] INDEX [name] [IF NOT EXISTS] FOR (n:L) ON (n.p [, n.q …]), … FOR ()-[r:T]-() ON (r.p [, r.q …]) (relationship indexes), CREATE INDEX ON :L(p [, q …]), DROP INDEX name [IF EXISTS], DROP INDEX ON :L(p [, q …]), SHOW INDEXES, CREATE CONSTRAINT [name] [IF NOT EXISTS] FOR (n:L) REQUIRE n.p IS UNIQUE (also IS NOT NULL, and (n.a, n.b) IS UNIQUE | NODE KEY; the older ON … ASSERT … form works too), DROP CONSTRAINT name [IF EXISTS], SHOW CONSTRAINTS

• Databases (through a Session): CREATE [OR REPLACE] DATABASE name [IF NOT EXISTS] [OPTIONS {…}], DROP DATABASE name [IF EXISTS], SHOW DATABASES, SHOW DATABASE name, SHOW DEFAULT DATABASE, USE name, USE name <query>

• Patterns: labels, inline property maps, $param maps, directions --> <-- --, type alternatives [:A|B], variable length *, *n, *min..max, named paths p = (…), pattern predicates in WHERE, shortestPath((a)-[:T*]->(b)) and allShortestPaths(…) (a bidirectional breadth-first search; WHERE conditions on the path or its relationships are applied during the search, so the result is the shortest path that passes them)

• Expressions: arithmetic, string concatenation, comparison chains, three-valued AND/OR/XOR/NOT, IN, STARTS WITH, ENDS WITH, CONTAINS, =~ (Java-style regular expressions, whole-string match, flags such as (?i); no lookaround or backreferences), IS [NOT] NULL, IS [NOT] [NFC|NFD|NFKC|NFKD] NORMALIZED, indexing and slicing, maps, lists, CASE, list comprehensions, reduce(acc = init, x IN list | expr), all/any/none/single, parameters, EXISTS { … }, COUNT { … } and COLLECT { … } subqueries (a pattern with an optional WHERE, or a full read-only query; they see every outer variable)

• Aggregates: count, count(*), sum, avg, min, max, collect, stdev, stdevp, percentileCont, percentileDisc (with DISTINCT), implicit grouping

• Functions: id, elementId, labels, type, keys, properties, startNode, endNode, nodes, relationships, length, size, degree, head, last, tail, range, reverse, coalesce, isEmpty, toUpper / upper, toLower / lower, trim / btrim, ltrim, rtrim (each with an optional string of characters to trim), replace, substring, left, right, split, char_length, character_length, normalize(s [, NFC|NFD|NFKC|NFKD]) (Unicode 17.0), toString, toStringOrNull, toInteger, toFloat, toBoolean (and their …OrNull forms), toIntegerList, toFloatList, toBooleanList, toStringList, nullIf, valueType, randomUUID, abs, sign, ceil, floor, round (also round(x, precision[, mode]) with UP, DOWN, CEILING, FLOOR, HALF_UP (default), HALF_DOWN, HALF_EVEN, rounding the number as written: round(2.675, 2) is 2.68), sqrt, exp, log, log10, sin, cos, tan, cot, asin, acos, atan, atan2, haversin, degrees, radians, isNaN, pi, e, rand, timestamp, exists

• apoc-style basics, built in (no plugin): apoc.coll.sum, apoc.coll.avg, apoc.coll.min, apoc.coll.max, apoc.coll.toSet, apoc.coll.sort, apoc.coll.flatten (true for deep), apoc.coll.contains, apoc.coll.indexOf, apoc.text.join, apoc.map.merge, apoc.map.fromPairs. Other apoc.* names return Unsupported

• Temporal values: date, localtime, time, localdatetime, datetime and duration, built from ISO 8601 strings (date('2015-07-21'), datetime('2015-07-21T21:40:32.142+01:00'), duration('P1Y2M10DT2H30M')), from maps (date({year: 2015, week: 30, dayOfWeek: 2}), datetime({date: d, hour: 10, timezone: '+02:00'}), duration({hours: 1.5})), from each other (date(dt)), or with no argument for the current time (one clock per statement; also date.statement() and friends). Fields such as d.year, d.week, t.hour, dt.offset, dt.epochMillis, dur.months, dur.minutesOfHour. Arithmetic: temporal ± duration (months first, clamped to the month's end; then days; then seconds), duration ± duration, duration * and / a number; duration.between(a, b), duration.inMonths/inDays/inSeconds, datetime.fromepoch(s, ns), datetime.fromepochmillis(ms). Values of one kind compare (zoned values by instant: datetime('…T01:00+01:00') = datetime('…T00:00Z')); durations only test equality. They can be stored as properties (and in lists) and indexed for equality. Time zones are UTC offsets only: named zones such as [Europe/Paris] need a time zone database and return Unsupported

• Spatial values: 2D and 3D points, Cartesian (point({x: 1, y: 2}), point({x: 1, y: 2, z: 3})) or WGS-84 (point({longitude: 12.5, latitude: 56.1}), with height for 3D), or chosen with crs: 'cartesian' | 'cartesian-3d' | 'wgs-84' | 'wgs-84-3d' or srid: 7203 | 9157 | 4326 | 4979. A null coordinate gives null. Fields x, y, z, longitude, latitude, height, crs, srid. point.distance(a, b): straight line for Cartesian points, haversine metres for geographic ones (height difference added in 3D); null across coordinate systems. point.withinBBox(p, lowerLeft, upperRight), where a geographic box may cross the 180th meridian. Points test equality but do not compare with < (null); ORDER BY sorts them after paths and before temporal values. They can be stored as properties (and in lists) and indexed for equality; HTTP returns them as {"$type": "point", "crs": …, "srid": …, "x": …, "y": …}

Not yet supported (these return an Unsupported error rather than a wrong result): procedures that change the graph, CALL { … } IN TRANSACTIONS, LOAD CSV from URLs other than local files, relationship constraints, named time zones, date.truncate and the other temporal truncation functions, spatial index seeks for point.distance / point.withinBBox (they filter instead), the trim(BOTH | LEADING | TRAILING 'x' FROM s) form (use btrim / ltrim / rtrim with a second argument).

$ docker run --rm -p 127.0.0.1:7475:7475 -p 127.0.0.1:7687:7687 ghcr.io/janvencl/graphbuff --demo movies