It's right along what I just played with regarding sending bytea as binary, the issue is deciding return results as binary or text or mixed.

I'll play a bit more with it, but until then you could get a long way simply casting json(b) to text in your query ::text and passing it directly.

You can also override the json types and set the serializer to null which will disable JSON.parse

I'll play a bit more with it, but until then you could get a long way simply casting json(b) to text in your query ::text and passing it directly.

I know about this ugly hack )

You can also override the json types and set the serializer to null which will disable JSON.parse

I can do this, only for some queries, or only settings for the connection?

types are set per instance.

Why do you find casting to text ugly? Now I'm curious if there's any overhead compared to skipping JSON.parse 😋

1. because I have to edit the original queries
2. this will worsen the understanding of the SQL code for people who will support it and there may be a situation when I need to execute the same prepared request in different places and somewhere I want to get the result in the form of an object, I need to call JSON.parse there )

That sounds reasonable.

How about adding the following .raw() method?

sql`select 1 as x, 2 as x`.raw()

That would resolve to this:

[
  [ <Buffer 31>, <Buffer 32> ],
  count: 1,
  columns: [
    { name: 'x', parser: [Function: parse], type: 23 },
    { name: 'x', parser: [Function: parse], type: 23 }
  ],
  command: 'SELECT',
  state: { status: 'I', pid: 62257, secret: 1364597943 }
]

This would help some other requests as well who wanted rows as arrays instead of objects, and allow same named columns to be retrieved.

For json like in your example

sql`select '{"a":"b"}'::jsonb`.raw()

you'd get this:

[
  [ <Buffer 7b 22 61 22 3a 20 22 62 22 7d> ],
  count: 1,
  columns: [ { name: 'jsonb', parser: [Function: parse], type: 3802 } ],
  command: 'SELECT',
  state: { status: 'I', pid: 62208, secret: -1778630945 }
]

Great, I like your interfaces

Would you mind trying out the #raw branch to see if it works well for you?

How about unsafe method, how to get buffer?

Same way :) the methods on a tagged query and an unsafe query are the same:

const [[buffer]] = await sql.unsafe('select '{"a":"b"}'::jsonb').raw()

Would you mind trying out the #raw branch to see if it works well for you?

I checked in my test version:

const [[buffer]] = await sql`SELECT '{"key": "value"}'::jsonb`.raw()
console.log(JSON.parse(buffer))

Everything works fine, thanks!

P.S.
Unfortunately, I cannot check the performance in real conditions, we do not have enough of this functionality for production #170

I still need some performance improvements around connection handling, but copy from / to works now https://github.com/porsager/postgres/tree/copy