if you reply to an answer or comment, those in the thread should get an email (unless they turned that off)

> Is there a reason for this? no, probably not

Good. Now I can sleep more easily.

While on the topic of reduce, in ArrayChunk we have

public Object reduce(IFn f, Object start) {
		Object ret = f.invoke(start, array[off]);
		if(RT.isReduced(ret))
			return ret;
		for(int x = off + 1; x < end; x++)
			{
			ret = f.invoke(ret, array[x]);
			if(RT.isReduced(ret))
				return ret;
			}
		return ret;
}

In short, yes - this is a different reduce method (part of IChunk), than every other reduce (IReduce, CollReduce)

That's what I was missing. Even with that hint, it took me a bit to piece it together. (The CollReduce bone connects to the InternalReduce bone connects to the IChunkedSeq bone. I think. ) Thx.

yes

I have made the mistake of implementing both interfaces in the same object which caused some very subtle errors that took a long time to figure out

(wrt reduced)

So, before I commit myself to an error, is the following correct for reduce in IReduceInit and IReduce (but not IChunkedSeq.reduce): One essential rule when writing a reduce method: after each invocation of the reduction function, check the result to see if it is an instance of Reduced; if so, stop immediately and return the deref value.

yes

Thanks. Planning on that note for my next blog post. Not much information out there about clojure.lang.Reduced The Transducers article seems to be about it. I may have missed others.

there is all the trickiness in expanding transducers that may do a sub-reduce that need to re-wrap reduced, but that doesn't change the rule above

Fortunately, all that complication lies out in the Clojure source code, which I can mostly ignore at this stage -- focusing just on the Java/C# parts. Except of course for solving my confusion about ArrayChunk .