When you're done making the requested changes, leave the comment: I have made the requested changes; please review again.

The pickling and copying should be a separate tracker item for Kristjan Jonnson to update his own code (I think he is the only user of that feature and would be willing to shoulder some of the recurring maintenance burden arising from that code).

I don't think that ignoring initial=None is right. And you did not think this too, according to your initial implementation.

It depends on whether you think None would be a useful input value to an accumulation.

The start=_sentinel approach gives more flexibility but is problematic for complicating the pure python version in the docs and for wrestling with argument clinic (I suspect we would have to remove the clinic code entirely and forgo the signature it generates). If so, I'm inclined to leave the PR as is.

This will make a subtle difference between accumulate() and reduce(), and also max(), iter(), next(), getattr(), etc which accept an optional argument, but distinguish None from no argument. I think this will be a bug magnet.

As for the pure python version in the docs, the pure python version of reduce() uses None as a sentinel.

Added pickle support (when initial is set, use chain([initial], it)). Please take a look.

The pickling part LGTM. But I think that initial=None should not be ignored.

Serhiy, thank you for the suggestion about initial=None, but I'm going to decline. I want to keep the argument clinic here and I agree with Tim's post on the subject. As module maintainer, I get to make the final call on the API.

Let's clear the way for Lisa to apply her first patch own her own. Please clear the review request so this can move forward. It has already consumed far too much time in relative to the value of the feature.

This feature is just too complex for its value.