PostgreSQL uses the format $1, $2 in the protocol, so I think it's just that nobody has bothered to implement named parameters in clients.

In another style, postgres.js uses calls such as sql`select * from t where id = ${variable}` (which is safe because it's a tagged template, not string interpolation).

This is basically how we do it in .NET. With Dapper it’s particularly neat sometimes because you can just pass an object that you were using anyway, and it will match parameter names to its properties case-insensitively.

I.e.

Query("select * from MyTable where Id = @Id", myEntity)

The idiom is to use anonymous objects which you can new up inline like “new { id, age = 5 }”, where id is an existing variable that will automatically lend its name. So it’s pretty concise.

The syntax is Sql Server native (which supports named params at the protocol level), but the Npgsql dat provider converts it to PG’s positional system automatically.

Kysely is a query builder: you build queries by calling javascript functions, while with pg-typesafe you write SQL directly.

I've used kysely before creating pg-typesafe, and came to the conclusion that writing SQL directly is more convenient.

A query builder works well for simple cases (db.selectFrom("t").where("id","=","1") looks a lot like the equivalent SQL), however, for more complicated queries it all falls apart. I often had to look at the docs to find how to translate some predicate from SQL to the required idiom. Also, I don't think kysely can automatically infer the return type of PostgreSQL functions, while pg-typed does (it asks PostgreSQL for it).

It only changes the types; not the code.

Kanel is great! That's what I used before pg-typesafe, and I still use it for dynamic queries.

Regarding the nominal/branded types, the typegen is configurable, so instead of e.g. mapping oid 20 to a bigint, you could map the field id of table foo to Foo["id"] from kanel.

see this example https://github.com/n-e/pg-typesafe?tab=readme-ov-file#type-j...

Very interesting, thanks!

I see they use the same global approach as pg-typed (asking for a ParameterDescription / RowDescription, which aren't usually exposed by the PG drivers), but there are interesting differences in the details. Also this made me realise that I could also type enums automatically.

I think plenty of ML languages have similar things. In F# SQL Type Providers have long been a mainstay (which makes using an ORM with F# almost meaningless IMO)

https://learn.microsoft.com/en-us/dotnet/fsharp/tutorials/ty...

SQLProvider is probably the most well known one: https://fsprojects.github.io/SQLProvider/

It’s really beautiful. You get type safety with SQL. If your code compiles, you guaranteed to have valid executable both F# code and SQL. Also you get to create composable queries.

And there are other, thinner/leaner type providers as well. My favourite Postgres one is: https://github.com/Zaid-Ajaj/Npgsql.FSharp

This simple sample executes query and read results as table then map the results

open Npgsql.FSharp

type User = { Id: int FirstName: string LastName: string }

let getAllUsers (connectionString: string) : User list = connectionString |> Sql.connect |> Sql.query "SELECT * FROM users" |> Sql.execute (fun read -> { Id = read.int "user_id" FirstName = read.text "first_name" LastName = read.text "last_name" })

Unfortunately I haven't found a way to make it work.

It would be quite easy to extract the queries to compute the types, but TypeScript doesn't handle tagged template literals well enough to link the query passed to the sql`` template to the return type.

Per "how to handle dynamic queries", it's admittedly pretty different b/c we're an ORM (https://joist-orm.io/) that "fetches entities" instead of adhoc SQL queries, but our pattern for "variable number of filters/joins" looks like:

const { date, name, status } = args.filter;

await em.find(Employee, { date, name, employer: { status } });

Where the "shape" of the query is static, but `em.find` will drop/prune any filters/joins that are set to `undefined`.

So you get this nice "declarative / static structure" that gets "dynamically pruned to only what's applicable for the current query", instead of trying to jump through "how do I string together knex .orWhere clauses for this?" hoops.

I haven’t found a good way to handle dynamic queries in pg-typesafe yet.

For now, I type these manually, which is acceptable for my usage as they are pretty rare compared to static queries.

I didn't look too much into sqlc-gen-typescript because the project looks abandoned (no commits in 2 years, many open PRs).

Regarding sqlc in general, it is focused on having the SQL queries in .sql files, while pg-typed is focused on having the queries inline (though I plan to add .sql file support). I like the latter approach better, as for small queries used in only one place, it is a little cumbersome to add them to a different file and find a name for them.

That’s what I thought.

The queries look a but more clumsy then but you won’t cause problems when spelling the column names.

  const [domain] = await db
    .select()
    .from(cmsDomains)
    .where(eq(cmsDomains.id, domainId))
    .limit(1);

Please tell us why we should.

That is the way node-postgres works. pg-typesafe adds type safety but doesn’t change the node-postgres methods