add a __getattr__ to Records so it looks cleaner

[nhumrich]

The syntax row.col looks a lot cleaner than row['col'].
What are you thoughts on adding a __getattr__ method to a Record so that both are possible?

[1st1]

We considered having __getattr__ instead of __getitem__ but decided against that, because:

1. There are a few methods in the Record object, such as .keys(), .values() etc. We don't want field names to clash with Record's API.
2. Having both __getattr__ and __getitem__ doing the same thing feels non-pythonic.

[runfalk]

Couldn't this and #17 be solved using a way of specifying an alternative Record class, or factory function, to use? This way the user could even use namedtuples if so desired for memory efficiency.

[1st1]

Record is as memory efficient as Python's tuple (it's implemented in pure C). It's also faster than namedtuple to instantiate, since we have a freelist for it.

[1st1]

I don't think we'll implement this. Adding __getattr__ will preclude us from adding methods/APIs to the Record class, so let's keep the status quo.