Iterator and Table interface

[Moelf]

This allows smaller memory footprint by avoiding materialize the entire array into memory. And we can use something like OnlineStats

I tried to implement one for a few days now and find fBasketSeek is very annoying and want some suggestion. I'm not sure what would be the canonical way to do this but:

struct BRANCH_ITR
    io
    seeks
    thetype
    count::Int
end

function Base.iterate(S::BRANCH_ITR, idx=1)
    basket_seek = S.seeks[idx]
    if idx > S.count
        return nothing
    elseif basket_seek==0
        return (nothing, idx+1)
    else
        s = datastream(io, basketkey)
        return (readtype(s, S.thetype), idx)
    end
end

this doesn't work because readtype advance the cursor and we lost track of it in the next iteration, where would you suggest to keep track of:

• basket cursor
• cursor within the basket?

https://quinnj.home.blog/2020/11/13/partition-all-the-datas/ Leverage this as ROOT provides natural partition too.

[tamasgal]

Sorry @Moelf I was very busy with work stuff and put UnROOT on hold (a bit selfish, but currently UnROOT works for my own analysis, so I put it on low prio in favour of finalising my PhD... 🙈 )

Anyways, the cursor interface needs to be fixed. I was playing with two different ways to implement it and the newer one is thread-safe and I think it will also easily solve the issue with your iterator interface.

So in fact, readtype() should use a copy of the Cursor instead of the actual io. What do you think?

I'd also like to add caching at some point.