Idea: Allow Mammoth to serialize/deserialize values, for more precise customization

[cakoose]

I problem I have:

• I want different int8 columns to deserialize differently: number vs BigInt vs Buffer.
• I have schema-aware JSON deserializers. I'd like to use those to deserialize different jsonb columns differently.

With "pg", you can only have a single serializer/deserializer for a type.

Mammoth has richer type information, e.g. jsonb<Article>. If Mammoth allowed transforming values, I could get exactly the serialization/deserialization I want.

One downside is performance. But maybe there won't be a performance hit if Mammoth hooks in to "pg" at a lower level, parsing the text/binary protocol directly. (Hopefully "pg-protocol" can do most of the work.)

[cakoose]

Just ran into another reason: "pg" doesn't correctly deserialize arrays of enums: brianc/node-pg-types#56

Mammoth has the type information to do it correctly.

[martijndeh]

And would you suggest to let mammoth handle the pool and everything pg is doing right now as well or do you see another way?

[cakoose]

The statement about "parsing the text/binary protocol directly" was a bit extreme. I think we might get acceptable performance just passing in a custom value parser, e.g:

import pg from "pg";

const client = new pg.Client({
    connectionString: "...",
    types: {
        getTypeParser(typeId: ..., format: 'text' | 'binary'): any {
            return v => v;
        }
    }
} as any); 

This just relays the raw string/binary value, so Mammoth can parse the values depending on the Mammoth schema. (It can use 'pg-types' to do the low-level work.)