Dedicated AST nodes for patterns? about patma

[Tobias-Kohn]

I concur: it is certainly a trade-off that could go either way and I have no strong feelings for or against either variant. Of course, Nick is using some dramatic rhetoric to make his point, quite overblowing the importance of this.

Without doubt, it would be a cleaner design with dedicated nodes, but at the expense of a significant population growth in the realm of AST nodes. While I like a clean design as much as the next man, I am not too sure about having to deal with a bunch of additional nodes—although it would probably be 'honest'. As you say, reusing the 'context' field does not seem like a good idea and would IMHO lead to a more brittle and complex design.

At the end of the day, it might be a good idea to have dedicated nodes because of the resulting cleaner design that provides a better abstraction from the concrete syntax used. If you squint a little, you might even put that into the category of "explicit is better than implicit" in that the nodes along with their handlers receive more descriptive names 😉.

[gvanrossum]

(What about the next woman? :-)

[brandtbucher]

Can we agree to defer this, then? It can easily be changed in a later alpha (I'm even comfortable with changing it in an early beta if Pablo is). If somebody makes noise about an actual use case where these new nodes are a quality-of-life improvement (mypy?), then that's probably enough to convince me. The only client I've converted so far is ast.unparse, which is much simpler with the current design.

I'm really not enthusiastic about creating, documenting, supporting, and testing a dozen new stdlib classes when our collective opinion is just "meh".

[Tobias-Kohn]

Yes, I am absolutely on board with the idea of deferring it. It seems to be mostly an implementation detail that is important rather for "meta-tools". Frankly, I share your feeling of not being too excited about creating all these new classes if there is not a real need for it.

[gvanrossum]

Sounds good to me. I think mypy will survive; it has its own AST structure that is derived from ast.AST by a transformation, and if it is necessary to use different nodes to represent e.g. a class pattern than a function call (which is likely) that can be done quite cleanly in the transformation.