Add the new record_class parameter to the create_pool() and
connect() functions, as well as to the cursor(), prepare(),
fetch() and fetchrow() connection methods.

This not only allows adding custom functionality to the returned
objects, but also assists with typing (see #577 for discussion).

Fixes: #40.

[1st1]

Instead of adding record_class everywhere I'd allow passing it to the constructor. Or add a set_record_class_factory method.

[1st1]

Is there any evidence that users want multiple different record types on one connection?

[elprans]

The entire impetus of this push was from #577, where the record_class is used as a TypeDict-like typehint for the result, which gets typechecked also.

I'd allow passing it to the constructor.

It is also allowed to be passed in the constructor if you want a connection-wide subclass.

@elprans is this PR ok to be merged? I'm really looking forward for this and the typings PR 😄

@elprans Can you provide an example how to implement custom record_class with dot-notation?
Because it's unclear from the docs. I think it would be very useful for many people. Thanks!

@elprans Can you provide an example how to implement custom record_class with dot-notation?
Because it's unclear from the docs. I think it would be very useful for many people. Thanks!

Not sure if this is correct but should work

    class MyRecord(Record):
        def __getattr__(self, item):
            return self.get(item)

#960 adds an example of this to the FAQ item.

I tried implementing that example, but I kept getting errors from the GraphQL execution engine because it passed __await__ (via hasattr()) and received None — based on my reading of the docs, you're supposed to raise AttributeError if the lookup fails:

https://docs.python.org/3/reference/datamodel.html#object.__getattr__

So I updated the code to this, and it started working:

class AttrRecord( asyncpg.Record ):
    def __getattr__( self, name ):
        if name not in self:
            raise AttributeError
        return self[ name ]