DortDB - SQL

This package is a language plugin for DortDB. It adds support for SQL SELECT queries.

Data adapter

The default data adapter implementation allows queries against iterables of JS objects.

``ts
const t1 = [
{ a: 1, b: 2, c: 3 },
{ a: 4, b: 5, c: 6 },
];
db.registerSource(['t1'], t1);
db.query('SELECT a, b FROM t1');
`

Dialect

The package is based on the PostgreSQL dialect. For convenience, identifiers _are_ case-sensitive. Only SELECT queries are supported.

Notable features

PostgreSQL includes several features that are not usual in other SQL flavors. Among them belong the following:

$3

Lateral joins allow joining of correlated subqueries. In other words, it is possible for the joined subquery to refer to previous tables, and thus to be re-evaluated for
different outer contexts.

`sql
SELECT t1.attr1, s.attr2 FROM t1
JOIN LATERAL (
SELECT t2.attr2 FROM t2
-- the WHERE clause refers to the outer context
-- the subquery is reevaluated for each t1 row
WHERE t2.attr3 + t1.attr4 > t1.attr5
) AS s
`

$3

The DISTINCT modifier filters out duplicate values. PostgreSQL allows customizing this behavior.

`sql
SELECT DISTINCT ON (attr1, attr2 % 10) attr1, attr2, attr3
FROM t1
`

$3

The arguments of aggregate calls such as count or sum may be filtered or sorted.

`sql
SELECT
count(DISTINCT attr1) AS distinct_count,
count(attr1) FILTER (WHERE attr1 % 2 = 0) AS even_count,
collect(attr1 ORDER BY attr1) AS ordered_array
FROM t1
`

Limitations

Currently, some SQL features are not supported, including:

- window functions
- common table expressions (CTEs)
- special GROUP BY behavior (ROLLUP, CUBE, GROUPING SETS)
- any schema-dependent features (read further)

$3

DortDB is by design schema-less. This clashes with some SQL expressions, that only make sense in a schema-based context. Consider the following:

`sql
-- which tables are a, b, or c from?
SELECT a, b, c
-- what columns are used in the natural join?
FROM t1 NATURAL JOIN t2
`

The allowed queries therefore face some restrictions.

- It must be clear from which tables specific columns originate.
- Non-qualified column names are only allowed when there is only a single, non-joined data source.
- Natural joins are disabled.
- All FROM clause subqueries must have an alias.
- Selecting all attributes using asterisk \* is currently also disabled, although in theory it should be possible.

When an attribute should not be interpreted as belonging to the current table, it should be prefixed by the nonlocal schema. This is relevant when inserting SQL subqueries
into languages that do not prefix identifiers, such as Cypher.

`sql
-- t1.attr1
SELECT attr1 FROM t1
-- t2.attr2, t2.attr3, attr1 from elsewhere
WHERE (SELECT attr2 FROM t1 WHERE attr3 < nonlocal.attr1)
``