(Apologies in advance for the large PR, but this a pretty foundational change that required quite a bit of boilerplate!)
This PR implements call chains as a new AbstractTokenTarget: BreakableCallChainEntry. This means that multilining for call chains -- just like for existing breakables -- will be decided on-the-fly during rendering, where we have more contextual information and can better enforce line length.
This is done with a combination of the normal breakable machinery and some additional "magic tokens." Specifically, this adds tokens at the beginning and end of the indent-able portion of the call chain, and then during rendering, we maintain a count of how many nested call chains we're in and adjust indentation accordingly. This isn't exactly the cleanest implementation, but given the complexity in use-cases for call chains, it was the best I could muster in a way that functioned how users would generally expect.
This is a fairly large PR, but there's primarily three main portions which each are about a third of the total diff each: fixtures, the boilerplate in ripper_tree_types to support getting expression starting lines, and finally the main formatting changes. I've done my best to comment these areas appropriately, but it's helpful to add further explanation there, I'm more than happy to add more.
Resolves #397
(Apologies in advance for the large PR, but this a pretty foundational change that required quite a bit of boilerplate!)
This PR implements call chains as a new
AbstractTokenTarget
:BreakableCallChainEntry
. This means that multilining for call chains -- just like for existing breakables -- will be decided on-the-fly during rendering, where we have more contextual information and can better enforce line length.This is done with a combination of the normal breakable machinery and some additional "magic tokens." Specifically, this adds tokens at the beginning and end of the indent-able portion of the call chain, and then during rendering, we maintain a count of how many nested call chains we're in and adjust indentation accordingly. This isn't exactly the cleanest implementation, but given the complexity in use-cases for call chains, it was the best I could muster in a way that functioned how users would generally expect.
This is a fairly large PR, but there's primarily three main portions which each are about a third of the total diff each: fixtures, the boilerplate in
ripper_tree_types
to support getting expression starting lines, and finally the main formatting changes. I've done my best to comment these areas appropriately, but it's helpful to add further explanation there, I'm more than happy to add more.