feat(halfvec): support halfvec for postgres target (#1171)

* feat(postgres): allow to put SQL type in target states

* feat(halfvec): support `halfvec` for postgres target

Author: georgeh0

Cargo.lock

@@ -61,7 +61,7 @@ tower = "0.5.2"
 tower-http = { version = "0.6.6", features = ["cors", "trace"] }
 indexmap = { version = "2.10.0", features = ["serde"] }
 blake2 = "0.10.6"
-pgvector = { version = "0.4.1", features = ["sqlx"] }
+pgvector = { version = "0.4.1", features = ["sqlx", "halfvec"] }
 phf = { version = "0.12.1", features = ["macros"] }
 indenter = "0.3.4"
 itertools = "0.14.0"

Cargo.toml

@@ -49,6 +49,11 @@ The spec takes the following fields:
 
 * `schema` (`str`, optional): The PostgreSQL schema to create the table in. If unspecified, the table will be created in the default schema (usually `public`). When specified, `table_name` must also be explicitly specified. CocoIndex will automatically create the schema if it doesn't exist.
 
+* `column_options` (`dict[str, PostgresColumnOptions]`, optional): Options for the columns in the table.
+  Key is the column name, and value is the specific option, with the following fields:
+
+  * `type` (`str`, optional): The specific type of the column in Postgres. Currently only supports `"vector"` and `"halfvec"`. By default, we use `"vector"` for vector columns, and you can use this field to override it to `"halfvec"` for some columns.
+
 ## Attachments
 
 ### PostgresSqlCommand

docs/docs/targets/postgres.md

@@ -1,20 +1,29 @@
 """All builtin targets."""
 
 from dataclasses import dataclass
-from typing import Sequence
+from typing import Sequence, Literal
 
 from .. import op
 from .. import index
 from ..auth_registry import AuthEntryReference
 from ..setting import DatabaseConnectionSpec
 
 
+@dataclass
+class PostgresColumnOptions:
+    """Options for a Postgres column."""
+
+    # Specify the specific type of the column in Postgres. Can use it to override the default type derived from CocoIndex schema.
+    type: Literal["vector", "halfvec"] | None = None
+
+
 class Postgres(op.TargetSpec):
     """Target powered by Postgres and pgvector."""
 
     database: AuthEntryReference[DatabaseConnectionSpec] | None = None
     table_name: str | None = None
     schema: str | None = None
+    column_options: dict[str, PostgresColumnOptions] | None = None
 
 
 class PostgresSqlCommand(op.TargetAttachmentSpec):