Rethink composition

[mofeing]

As of right now, * or composition of Bijections doesn't respect the dictionary types. We should think about them before releasing 1.0.

If we are not sure, we should remove Bijection composition and add it in the future when we think the feature is ready. I don't think it's a critical feature and I don't want to deal with backward compatibility issues in the future.

[mofeing]

Also, given that it's "composition" and mathematically a bijection is a function, we should be using the circ symbol.

[scheinerman]

Also, given that it's "composition" and mathematically a bijection is a function, we should be using the circ symbol.

I suggest both. Have $\circ$ as a synonym.

[goerz]

Could we resolve this and push out 1.0.0 soon?

I was holding back the release of DocumenterCitations 1.4.0, since I wanted to avoid having to push a compat-bump for Bijections 1.0.0. People are getting antsy, though: JuliaDocs/DocumenterCitations.jl#98.

I suppose I could proactively declare compatibility with Bijections 1.0.0, which would of course mean we absolutely must make sure that the upcoming v1.0.0 release is compatible (which I think is the plan anyway)

[mofeing]

Yeah, sorry. Busy week. From my side I would just remove it for now. Since adding it in the future is non-breaking, we could think about readding it in v1.1.0.

@scheinerman opinions?

[goerz]

Removing it as done in #39 seems fine to me. However, we should include instructions in the CHANGELOG on how to adapt to the removal (and maybe a test). I think that's

a = Bijection{Int,Int}(); a[1] = 10; a[2] = 20;
b = Bijection(1=>2, 3=>4, 5=>6)
Bijection(merge(a, b))

Right?

[goerz]

Oh, hang on… that's not what the old composition was doing at all.

So I don't think there is any obvious replacement. I'm not totally sure how/why this composition is useful exactly. So we could indeed remove it, even though that leaves anyone who was using it hanging. Is there anyone using this? What was the motivation for adding this in the first place?

The Base.:(∘) does seem like the most obviously correct symbol to use for this.

When I ask ChatGPT "Julia has the ∘ symbols for function decomposition. How do I compose functions without using a unicode symbol?" it hallucinates a non-existent Base.compose function. I gotta agree with ChatGPT! That function really should exist, not just Base.ComposedFunction.

So unless we have an argument that composition of two Bijections shouldn't exist because it it confusing / unhelpful / not part of the AbstractDict interface, I would add it again just with Base.:(∘) replacing Base.:(*), and maybe implemented by a Bijections.compose function for those who do not like unicode.

P.S.: I'm not sure I understand the original concern

doesn't respect the dictionary types

julia> c
Bijection{String, Int64, Dict{String, Int64}, Dict{Int64, String}} with 2 entries:
  "bye" => 20
  "hi"  => 10

Is there something wrong with that?

[goerz]

Maybe have a look at #41