Add Hash/Indexable#dig/dig?

[Sija]

Followup to #3536.

[j8r]

It would be also great to have #dig(keys : Enumerable), where keys represents a variable sequence of keys.

[RX14]

Thought we decided against this. It was .dig or JSON::Any and JSON::Any won because my proposal forgot that JSON::Type included more than Hash and Array. I proposed it could be solved by making JSON.parse return Hash | Array because the 95% usecase for json is arrays or hashes at the top level, but that idea wasn't well recieved.

@j8r that method would just produce insane types.

[Sija]

@RX14 I understood differently, also according to your own #3536 (comment).
@j8r That would forgo all compile-time safety, ditto what @RX14 said.

I still see it as a very handy addition to stdlib and see no reason not to implement it.

[asterite]

I was against dig because I thought the implementation would make the return type be the union of everything. But this implementation is brilliant.

a = {1 => [[1, 2], [3, 4]]}

p typeof(a.dig(1, 0)) # => Array(Int32)
p a.dig(1, 0) # => [1, 2]

p typeof(a.dig(1, 0, 1)) # => Int32
p a.dig(1, 0, 1) # => 2

So I see it as a very nice, convenient way, to traverse deep structures. Now we would just need to add it to JSON::Any and YAML::Any... or remove those ugly Any wrappers altogether (I'd start by adding dig/dig? to those types, and eventually decide how to remove them).

[Sija]

So I see it as a very nice, convenient way, to traverse deep structures.

Exactly.

... or remove those ugly Any wrappers altogether (I'd start by adding dig/dig? to those types, and eventually decide how to remove them).

Yes, please! 🎉

[asterite]

@Sija Do you want to add dig/dig? to Any then? At least that way this PR will be "complete".

[Sija]

@asterite Sure, will do.

[codenoid]

hi @Sija , long time no c

[Sija]

@asterite Done.
@codenoid Indeed, good to c u back! :)

[Sija]

Is it GTG now? CI fails seems not related, btw.

[Sija]

Hello? Is this got stalled for a reason?

[bcardiff]

GTG on my side. Maybe NamedTuple should also have #dig to traverse them directly.

[Sija]

@bcardiff I like it 👍, it's here in the last commit.

[bcardiff]

@Sija CircleCI had a bad week in their servers.

In darwin is still failing because it trying to compile the compiler with llvm 7, which is still not supported. The dependency of the formula is llvm plain and that leads to llvm 7 right now.