sequelts VS assert-combinators

You can use these template literal types + infer to build an entire SQL parser. I did a POC that infers SQL query types by parsing the SQL query on a type level:

https://github.com/nikeee/sequelts

However, building this parser is pretty cumbersome and supporting multiple SQL dialects would be lots of pain. While I'm not a fan of query builders per se, Kysely pretty much covers everything that my POC tried to cover (except that 0 runtime overhead). However, you get the option to use different DBMs in tests than in production (pg in prod, sqlite in tests), which is a huge benefit for a lot of people. sequelts was designed to work with sqlite only. And it's not a hack.

The only thing I can imagine where this would be useful is when you don't have control about what DB is being used, for example, when building a product that should be compatible with Postgres and MariaDB (and each is getting used). However, in the age of containerization, this isn't a big problem any more.

In some ORMs, I need to create types that the result of a query containing JOINs is mapped to. Others don't support them _at all_. In TypeORM, there is a query builder which forces you to put in _some_ SQL for things like "WHERE a in (b, c)".

I created a proof of concept of a different approach: Just embrace SQL and provide static typing based on the query. The return type of a query is whatever that thing is that the query returns in the context of the database schema. It's possible to do in TypeScript, by parsing the SQL query at development time:

https://github.com/nikeee/sequelts

One benefit is that it does not need any runtime code, as it's just a type layer over SQL. You don't have to rely on some type-metadata that TypeScript emits. That's why it also works with JavaScript only.

I don't like ORMs that use runtime types either. Most of the time, I want to write raw SQL.

So as an experiment, I created a library that statically types raw SQL:

https://github.com/nikeee/sequelts

The idea is to parse the SQL queries using TS's type system. The parsed query is combined with the database schema and therefore, we know what type the query will return.

This is especially useful due to TS's structural type system.

We use in prod variant of no 1. [0]. Why? Because:

* it's extremely lightweight (built on pure, functional combinators)

* it allows us to use more complex patterns ie. convention where every json field ends with Json which is automatically parsed; which, unlike datatype alone, allows us to create composable query to fetch arbitrarily nested graphs and promoting single [$] key ie. to return list of emails as `string[]` not `{ email: string }[]` with `select email as [$] from Users` etc.

* has convenience combinators for things like constructing where clauses from monodb like queries

* all usual queries like CRUD, exists etc. and some more complex ie. insertIgnore, merge1n etc has convenient api

We resort to runtime type assertions [1] which works well for this and all other i/o; runtime type assertions are necessary for cases when your running service is incorrectly attached to old or future remote schema (there are other protections against it but still happens).

[0] https://github.com/appliedblockchain/tsql

[1] https://github.com/appliedblockchain/assert-combinators

We use not so much frameworks but combination of lightweight libraries:

- runtime assertions [0] - to map unknown values at i/o boundary into statically typed code (rpc input parameters, sql results etc)

- template based sql combinators to sanitize sql/generate sql [1]

- jsonrpc over websockets - for bidirectional comms between f/e and b/e

[0] https://github.com/appliedblockchain/assert-combinators

[1] https://github.com/appliedblockchain/tsql

Types can be asserted at runtime (parsed) at IO boundaries (reading http request or response, websocket message, parsing json file etc). Once they enter statically type system they don't need to be asserted again.

The difference it makes is illusion of type-safety vs type-safety this article touches on.

You can try to bind service with client somehow but in many cases this will fail in production as you can't guarantee paired versioning, due to normal situations by design of your architecture or temporary mid-deployment state or other team doing something they were not suppose to do etc. It's hard to avoid runtime parsing in general.

Functional combinators [0] or faster [1] with predicate/assert semantics work very well with typescript, which is very pleasant language to work with.

[0] https://github.com/appliedblockchain/assert-combinators

[1] https://github.com/preludejs/refute

Once i/o boundaries are parsing unknown types into static types, your type safety is guaranteed.

[0] https://github.com/appliedblockchain/assert-combinators