Creating a Tree is expensive

[epage]

I am using this crate to show what parts of an assertion failed and why (see assert-rs/predicates-rs#7).

For simple nodes, it seems like the current API is great. Unfortunately, I have heterogeneous nodes for my expression tree. so with treelines current API, I have to copy this into a treeline specific tree and convert each of my nodes into a Box<Display> to conform to the API, making a lot of unnecessary work.

[epage]

I wonder if we could create a Tree trait, maybe something like:

pub trait Tree {
    fn root(&self) -> &Display;
    fn leaves(&self) -> Box<Iterator<Item=&Tree>>;

    fn display(&self) -> DisplayTree;
}

pub struct DisplayTree(...);

impl Display for DisplayTree {
...
}

This would allow heterogeneous nodes, reducing unnecessary Box::new(item.to_string()), instead allowing items existing Display to be directly leverage.

The existing Tree implementation could still exist for rapid prototyping and those that don't have an complex trees.

[softprops]

In like your idea. If you have some time I'd be open to a PR. Otherwise I can take a look

[epage]

Unfortunately, this is one of the lower items on my priority list (using various tricks to avoid treeline being in my public API, allowing me to have it be improved later).

If curious, you can see my use of treeline in assert-rs/predicates-rs#60. predicates is forming the basis of a flexible assertion system I'm working on, with assert_fs and assert_cmd being my priority clients.

[epage]

I have mixed feelings about the display method vs impl<T> Display for T where T: Tree

For my use case, I'm probably going to do this

pub trait CasseTreeDisplayExt {
    fn tree(&self) -> CaseTree;
}

pub CaseTree(Case);

impl Tree for CaseTree

and ideally, CaseTree is Display. The reason for the newtype is my tree code will live in a separate crate from Case for minimizing semver breaks for my core logic.