I thought I’d raise a discussion on just what exactly iterators are, a simple view might be that they’re just a sequence of values which can be gotten using .next(). However this is simplified because iterators .next can actually do three things:


   They can yield a value


   They can return a value


   They can throw an error

With these three things in mind how should we actually consume these things called iterators.

The current methods are (in addition to the await versions in async-iteration):

1. The for-of loop but it can only use non-completion values
2. Manually calling .next e.g. in a while loop

One place I’m aware of all three things being used is in coroutines (which have two way communication) but are there other places where errors/returns are actually used for iterators? As an example where I might consider using both non-completion and completion values in an iterator is for hinting progress values as an example I’ve actually written (but using Workers and progress events instead) a library to generate a Range Minimum Query for a very large array of data, but suppose we had implemented this using some fictional IterableWorker:

|// Inside iterable worker function* createMinQueryTree(data) { const result = new Uint8Array(data.length*2) const chunkSize = Math.floor(data.length/1000) for (let i=data.length; i >= 0; i--) { const leftChild = 2*i + 1 const rightChild = 2*i + 2 result[i] = Math.min(data[leftChild], data[rightChild]) if (i % chunkSize === 0) { yield i/data.length } } return result } // Send progress messages somehow ... // Inside main program const worker = new IterableWorker('./createMinQueryTree.js') const iter = worker.postMessage(data)[Symbol.iterator]() bar = new ProgressBar() let result while (true) { const {done, value} = iter.next() if (done) { // If done then our value is the result result = value break } bar.displayProgress(value) } bar.dispose() ... // Do stuff with result |

Of course this might be an antipattern (if so can someone suggest how one would /actually/ go about doing something like this), but assuming this is reasonable it would be far nicer if we had some fictional construct such as:

|const bar = new ProgressBar() for (let progress of worker.postMessage(data)) { bar.displayProgress(progress) } then (result) { ... // Do stuff with result } bar.dispose() ... // Do stuff with result |

We could even extend this construct with catch e.g.

|const bar = new ProgressBar() for (let progress of worker.postMessage(data)) { bar.displayProgress(progress) } catch (err) { doSomethingToRecover(err) } then (result) { ... // Do stuff with result } finally { bar.dispose() } |


Questions I have:

 * What do people use iterators for other than just data sequences?
 * What ways are there to represent sequences such as progress that
   have a completion value?
 * Even if the above example is an anti-pattern would the expository
   syntax still be useful for other purposes?
 * If the above example is an anti-pattern what might code for the
   above situation look like using Iterators/AsyncIterators and
   for-of/for-await-of loops or is this problem fundamentally not
   expressible in a clean way using these constructs?
 * I believe this syntax would cause ambiguity with different styles
   e.g. this would be ambiguous

   |for (let i of it) { ... } then (...) // This would be parsed a
   function call { // Followed by a block ... } |

   so what alternatives might there be (|else| maybe?)?

All other feedback is welcome too.

es-discuss mailing list

Reply via email to