pg_types

Erlang library for encoding and decoding postgres data types. Used by pgo.

How it works

A type module implements the pg_types behaviour. init/1 must return a 2-tuple with the first element being a list of typsend binary strings. typsend is what Postgres calls the conversion function and has defined for each OID. The second element is any sort of configuration the type needs included when encoding or decoding. For example the json implementation might want to support setting the json module to use when encoding and decoding.

encode/2 and decode/2 are passed the valuel to encode or decode and a type_info record. The record contains the configuration returned from init/1 under config, so #type_info{config=Config}. The other fields are columns from Postgres' pg_type table.

Using

pg_types:update/3 will lookup all type modules, based on the modules in pg_types and the pg_types application environment variable modules, which is how to add implementations not found in this application. The update function takes an atom namespace that usually corresponds to whatever pool the client is setting up types for, a list of type_info records populated from the Postgres table's rows and any option map to pass to each type module's init/1. It then sets a persistent_term with the type_info record updated to include the module that implement's the types typsend and the configuration returned from init/1.

pg_types:update/4 with a map as the last argument to update a map with oid() => type_info()} instead of having them inserted into a persistent_term.

To find the module for encoding/decoding an oid use lookup_type_info/2, passing the namespace (pool) and oid, it will return the type_info record with module and config set.

Configuring

To configure if json is to be encoded and decoded by a library, include the json library, like jsone, in theprojects deps list and set json_config in pg_types application environment to a 3-tuple {module(), encode_options(), decode_options()}. For example to use jsone and decode keys as atoms:

{pg_types, [{json_config, {jsone, [], [{keys, atom}]}}]}

To configure if enums should be converted to atoms, set enum_config in pg_types application environment to the atom atoms, or the atom existing_atoms. Using existing_atoms will result in binary_to_existing_atom(Value, utf8) being used to decode values.

{pg_types, [{enum_config, atoms}]}

Incomplete Types

Timezones: I don't think the timestamptz is done
Not a type but need to decide on the atoms for and define macros for nan, infinity, -infinity and make sure all types use them
Timestamps: currently only supports integer timestamps. Not sure if there is really a reason to support the old float version?

Acknowledgements