function next(eit::EdgeIter, state) modifying state

[IssamT]

I think state should not be modified by the function next.

http://stackoverflow.com/questions/39181313/julia-iteration-start-next-done-side-effects

Fixing it requires a minor modification. I can suggest a PR if you don't mind.

[sbromberger]

Please do. Thanks!

[jpfairbanks]

Is the only argument against mutating the state variable is to allow caching the iteration state? Is there an example of code doing this? I can only recall seeing the start, next, done interface used in for loops, which do not cache the state variable.

[sbromberger]

@Keno - your input appreciated here, especially given http://stackoverflow.com/questions/39181313/julia-iteration-start-next-done-side-effects - see also the PR (#423).

[Keno]

Is the only argument against mutating the state variable is to allow caching the iteration state? Is there an example of code doing this? I can only recall seeing the start, next, done interface used in for loops, which do not cache the state variable.

This capability is used extensively in some of the iterator wrappers e.g. those from https://github.com/JuliaLang/Iterators.jl. In general the state variable is not very large and immutable. If that's not the case, the trade-off needs to be carefully considered.

[Keno]

In this case, I haven't looked at the code too much, but from that PR, it looks like EdgeIterState could be made immutable in which case the non-mutating version is probably better.

[sbromberger]

the PR shows O(n) allocations with immutable state. This looks like it would be a big problem. Your thoughts?

[Keno]

Did you change EdgeIterState to an immutable?

[jpfairbanks]

PR #423 does not. In fact it creates a new instance and then mutates it.

[IssamT]

I am sorry I just see the conversation here and only read the messages in #423. I will try to use the immutable as @Keno suggests

[IssamT]

I m surprised by the results when using immutable. To be honest I don't understand the reason behind that... I added the results in #423

[sbromberger]

Closing via #423 - thanks for the really great discussion.