usability issue: `connect()`-like functionality is only available on slices, not on Iterator(Item=Str)

[Byron]

In other languages, this kind of functionality is easily achieved:

# my_fields can be an iterator
", ".join(my_fields)

In Rust, this currently only works on slices and vectors. However, in generic code, one usually has an iterator only, and I believe it should be simple to connect such an iterator if the item type is compatible.

Otherwise I fear people will start to throw in an intermediate collect::<Vec<String>>() call, which has performance implications.

All details, code, and the current workaround using itertools.intersperse() can be found in the respective question on stackoverflow.

Please note that this seems to be related to a more general issue, namely that some useful functionality is not implemented on Iterator, even though it could be: http://internals.rust-lang.org/t/add-dedup-for-iterator/1567 .

Meta

rustc 1.0.0-nightly (522d09d 2015-02-19) (built 2015-02-19)
binary: rustc
commit-hash: 522d09d
commit-date: 2015-02-19
build-date: 2015-02-19
host: x86_64-apple-darwin
release: 1.0.0-nightly

[tbu-]

In Python this is basically just implemented on the equivalent of Vecs, if you pass in an iterator, the function first collects these elements into a vector and then joins them.

[Byron]

I think the point is that one can pass an iterable, whereas in Rust this is currently not possible. Also there seems to be no reason to not have such functionality readily available. Of course I might be totally mistaken there given my greenling status with Rust. Nonetheless I try to point out a usability issue, and addressing it will make using Rust easier in the given context.

[tbu-]

I just addressed your

Otherwise I fear people will start to throw in an intermediate collect::<Vec<String>>() call, which has performance implications.

by pointing out that Python does that exact thing in its "connect" (join) function.

[Byron]

I think it would be comparable if one would have to write something like this: ", ".join(list(my_iterator)). As far as I am concerned, if I don't know what it does, I won't care as long as it is comfortable to use.

Also I hope what I wrote makes at least some sense, I don't know if I this issue is complying with requirements, or if it is at the right spot even.

[bluss]

1. .dedup() for an iterator is super simple to add by third parties (see itertools), while Vec's .dedup() is much more involved and requires unsafe code and concern for exception safety.

2. .connect() and .concat() are available on Vec I think because: We can do it "perfectly". We allocate the exact space needed beforehand, and do the concatenation. An iterator implementation has this imperfection that we can't know how much memory to allocate beforehand -- this has a certain performance cost. It's almost Rust tradition to punt the issue until it can be resolved in the best way.

And note, the string-joining functionality on iterators was implemented by a trait in collect-rs. I've imported it into itertools, so at least it's available there.

[tbu-]

@bluss Shouldn't connect and concat be implemented on slices though?

[Byron]

Thank you - the newly added join() implementation will work for me.

@bluss About 2) Isn't there a rewindable iterator Trait somewhere, which would allow to traverse sequences multiple times. In that case, one could have one run to compute the required string size, and a second one to fill in the result string.

[tbu-]

@Byron That would probably be a Clone bound. If you can clone the iterator before traversing it for collection, you can call .count() on it.

[Byron]

True ! It could be that easy ! With that in mind, maybe connect() could actually make it into the IteratorExt Trait, as an optimal implementation would be possible (?).

[steveklabnik]

Triage: no change here. The itertools crate has this.

[tbu-]

Now that a crate implements this, it could be considered fixed. Adding it to the standard library would likely require an RFC.