Squee: A Typed, Composable Database Query Language
Squee is an experimental language that breaks relational queries down in to composable functions that can be fully inferred with a few extensions to Hindley Milner. It compiles to SQL and so can be used with relational databases.
WARNING: Squee is EXPERIMENTAL and INCOMPLETE. Don't use this for anything important!
Docs
Rationale
As much as I love languages with a good type system, integrating types with relational querying is often a weak point. It can be complicated, clunky, and result in poor error messages.
Squee's type system has been designed to handle relational concepts and give good, understandable error messages (WIP - error messages are currently terrible).
Unlike SQL, Squee breaks queries down in to composable parts so you can define and re-use joins, maps and filters across your code.
Some Quick Examples
Given a PostgreSQL database with the following definition:
CREATE TABLE example (a int not null, b text not null);
INSERT INTO example VALUES (1, 'example1'), (2, 'example2');
CREATE TABLE join_example (a int not null, c text not null);
INSERT INTO join_example VALUES (1, 'join_example1'), (2, 'join_example2');Squee will introspect the database and make tables available:
SQUEE> example
: [{a: ~int4, b: ~text}]
| a | b |
+---+----------+
| 1 | example1 |
| 2 | example2 |
SQUEE> join_example
: [{a: ~int4, c: ~text}]
| a | c |
+---+---------------+
| 1 | join_example1 |
| 2 | join_example2 |
Queries can be built out of map, filter, order, natjoin, join and aggregate:
map : ({α} → {β}) → [{α}] → [{β}]filter : ({α} → ~bool) → [{α}] → [{α}]order : (Comparable β) ⇒ ({α} → β) → [{α}] → [{α}]natjoin : ({γ} = {α} ⋈ {β}) ⇒ [{α}] → [{β}] → [{γ}]join : ({α} → {β} → ~bool) → ({α} → {β} → {γ}) → [{α}] → [{β}] → [{γ}]aggregate : ({β} = Agg {γ}) ⇒ ({α} → {β}) → [{α}] → [{γ}]
The types are explained in the language docs.
The filter function can be abstracted and used in multiple queries:
SQUEE> example | filter (\t -> t.a = 1)
: [{a: ~int4, b: ~text}]
| a | b |
+---+----------+
| 1 | example1 |
SQUEE> def filterA1 := filter (\t -> t.a = 1)
filterA1 : [{a: ~int4, ..α}] → [{a: ~int4, ..α}]
SQUEE> filterA1 example
: [{a: ~int4, b: ~text}]
| a | b |
+---+----------+
| 1 | example1 |
SQUEE> filterA1 join_example
: [{a: ~int4, c: ~text}]
| a | c |
+---+---------------+
| 1 | join_example1 |
as well as the natjoin function:
SQUEE> example | natjoin join_example
: [{a: ~int4, b: ~text, c: ~text}]
| a | b | c |
+---+----------+---------------+
| 1 | example1 | join_example1 |
| 2 | example2 | join_example2 |
SQUEE> def joinExample := natjoin join_example
joinExample : ({β} = {a: ~int4, c: ~text} ⋈ {α}) ⇒ [{α}] → [{β}]
SQUEE> example | filterA1 | joinExample
: [{a: ~int4, b: ~text, c: ~text}]
| a | b | c |
+---+----------+---------------+
| 1 | example1 | join_example1 |
The queries can be exported to other languages.
Given the file example.squee:
export exportedExample := example
export filteredExportedExample a := example | filter (\t -> t.a = a)
The command squee generate sql-prepare example.squee will generate:
PREPARE exportedExample AS
SELECT "a","b" FROM "example" AS _t;
PREPARE filteredExportedExample AS
SELECT "a","b" FROM "example" AS _t WHERE ("a") = ($1);Since Squee is fully type inferred, it can also generate templates for languages that require type annotations:
squee generate hs-postgresql-simple example.squee:
exportedExample :: Connection -> IO [(Int, String)]
exportedExample connection = do
query_ connection "SELECT \"a\",\"b\" FROM \"example\" AS _t"
filteredExportedExample :: Connection -> Int -> IO [(Int, String)]
filteredExportedExample connection a = do
query connection "SELECT \"a\",\"b\" FROM \"example\" AS _t WHERE (\"a\") = (?)" (Only a)Installation
Requirements:
- Stack
- PostgreSQL development libraries (e.g. on Debian/Ubuntu:
apt install libpq-dev) - ncurses development libraries (e.g. on Debian/Ubuntu:
apt install libtinfo-dev)
Clone the repo, and then stack install.
Notes and Limitations
- Squee doesn't currently handle nulls or nullable fields.
- Rows have a flat namespace, as opposed to SQL's qualified field names i.e.
table.column. A row cannot have duplicate field names because there's no way to disambiguate them. - Very few SQL operators/functions are currently available.
- Only natural and inner joins are currently available. Left/right joins require null handling.
- The error reporting is currently poor, but quality error reporting is a long-term goal of the project.
Interesting Links
If you know of any similar projects not listed here, please let me know!