I'm also willing to assist with implementing this but I have no clue as to where to start as I've never contributed to this project before.

this is possibly not trivial, not on the reflection side, but on the mechanics of INSERT. all of the PG internals assume SERIAL for autoincrement and getting this format to work would be the hard part.

Further Reading:

• PostgreSQL 10 identity columns explained Overview on Identity Columns by a PostgreSQL developer who (co?)authored the feature
• PostgreSQL Docs: Generated Columns

we do support computed columns in general however i think PG's "identity" is a separate syntax.

Aside from reflection:

• new column type and DDL.
• potential config options: generated=OneOf(["always", "default"]), **sequence_options
• inserts will need the potential for OVERRIDING SYSTEM VALUE
• update/remove inline docs for /dialects/postgresql/base.py (https://github.com/sqlalchemy/sqlalchemy/blob/master/lib/sqlalchemy/dialects/postgresql/base.py#L14-L78)

@zzzeek check out this link that goes into some of what is done regarding the auto-creation of serials. https://www.depesz.com/2017/04/10/waiting-for-postgresql-10-identity-columns/

We should check if the generated always as identity is already piked up by the reflection logic in 1.4, since the syntax is the same as a normal compute column.

We should check if the generated always as identity is already piked up by the reflection logic in 1.4, since the syntax is the same as a normal compute column.

the syntax is not what we generate, I tried this:

class A(Base):
    __tablename__ = 'a'

    id = Column(Integer, Computed("AS IDENTITY"), primary_key=True)
    data = Column(String)
e = create_engine("postgresql://scott:tiger@localhost/test", echo='debug')

Base.metadata.drop_all(e)
Base.metadata.create_all(e)

and got

sqlalchemy.exc.ProgrammingError: (psycopg2.errors.SyntaxError) syntax error at or near "("
LINE 3:  id INTEGER GENERATED ALWAYS AS (AS IDENTITY) STORED NOT NUL...
                                        ^

[SQL: 
CREATE TABLE a (
	id INTEGER GENERATED ALWAYS AS (AS IDENTITY) STORED NOT NULL, 
	data VARCHAR, 
	PRIMARY KEY (id)
)

I think this looks like a totally different (but very similar) syntax. One proposal is to add this as a column option:

class A(Base):
    __tablename__ = 'a'

    id = Column(Integer, postgresql_identity="BY DEFAULT", primary_key=True)
    data = Column(String)


class A(Base):
    __tablename__ = 'a'

    id = Column(Integer, postgresql_identity="ALWAYS", primary_key=True)
    data = Column(String)