I am using James Harton’s Heap / Priority Queue and I would like to know if it’s possible to implement List-like pattern matching for a custom datastructure. I.e. the heap is implemented as a struct, but I would like to:
# my heap actually contains {DateTime, object} tuples.
heap = 1..10 |> Enum.shuffle() |> Enum.into(Heap.min)
root = hd heap
rest = tl heap
# and
[root | rest] = heap
I first looked at Protocols and Behaviours, but found that the Heap already implements Enumerable and Collectable, which I thought might enable this. Then I looked at the source of hd and saw that it’s just calling into Erlang.
Is what I’m trying to do possible at the moment? (It seems strange to me that hd would only work on lists and not tuples for example.)
[root | rest] = Enum.to_list heap is not ideal, since it turns the rest into a List. I guess the Heap module will have to implement a function that returns a tuple with {root, rest} (or a 2 element list), if I want to access both with a one liner.
Yeah, thanks The Heap does implement Heap.root/1 to get the first element and Heap.pop/1 to get the rest. I was just looking for an Elixiridiomatic way to implement something that returns both.
Yeah, Enum.split/2 is the closest to what I’m aiming for. (Might also be a good name, since pop is already taken.) Would also be cool if there was a protocol’ish way to implement “match like a list” and “match like a map”.
Follow-up question: What’s cheaper / more idiomatic, returning a tuple or a list containing head and tail?
Depends, is it a fixed-length set? Then tuples. Is it an unknown length at runtime, then lists. ^.^
They are both very cheap, the BEAM is highly optimized for both. Tuples are slightly faster on longer sizes than lists technically, since their length can be matched on immediately if that is an important attribute for your matchers.
Hehe, cool. I’ll definitely check it out when I’ve got a working prototype. Probably shouldn’t dive too deep into metaprogramming just for syntactic sugar before my code works.
To be honest, after what I’ve learned from my datastructure lessons, I’d have expected there were an operation which is equivalent to {Heap.root(h), Heap.pop(h)} (perhaps a PR against the repository helps, biggest problem is to come up with an appropriate name… This is exactly what I had pop/1 expected to do…).
I’d not use anything from Enum if you want to retain the Heaps attributes! Anything in Enum will convert your heap into an ordinary list!
I’ve been thinking of using my ProtocolEx system to build a replacement Enum (EnumEx?) module that implements the proper styles and names of the algorithms. I could even use the new property testing library in it automatically to ensure that things follow the interfaces properly too. ^.^
But it would be so nice to have an enum library that actually returns the same types, a proper Monad Enum library. Enum just breaks SO many of the Monad Laws (the most basic of which is to always return the same-bleeping-type of monad when a monad is returned unless explicitly casting). >.>
HI. I believe what you think of is not possible. Here are some suggestions based on speculation (ie I havent tried this code in practice)
To get hd/tl perhaps create your own module Heap.hd and Heap.tl and then include them into your module when using Heap. Your implementations of hd and tl must fall back to Kernel module if the argument is not a Heap struct.
For [h | t] you could use sigil syntax, for example %H[h | t] and solve that with macros quite easily. You should probably have a named fallback in case of naming collisions for that one. For examples look at the Kernel module.
My suggestion overall would be to skip the fancy stuff and just create functions called split, hd and tl in the Heap module. The split function should return a two item tuple. That extra syntactic sugar is probably not worth it, considering the complexity added…
The Heap does implement 
