pg barely was able to do it, and he cared a lot about whether the world wanted to use arc.

> it would really benefit from having the real latest version (or as close to that as is possible).

I independently implemented almost every feature in HN. But it exists in a repository that gets autokilled whenever I link to it here, so, that's about par for arc's progress in modern times.

Took so much work to get all the features done. Even implemented Dan's thread-local state ideas. Never gonna see the light of day.

The fact is, Arc's progress stopped with pg. It's best to accept that and move on.

It's tempting to think you need the latest in order to understand arc's power. But take it from a decade-long arc veteran: that's not true. All you need is to tug at the question, "why did pg do it this way? out of all possible ways, why this?"

And then realize that HN today could not have happened if pg had done it any other way.

If you care enough about this question, you'll come to respect arc. Especially what was left out, as much as what was implemented. Less is more.

Over the years I've implemented at least 5 different flavors of arc, including a fusion of lumen and arc, both on lumen (javascript-based) and in arc (racket-based).

I always come back to vanilla arc and vanilla lumen as starting points. And there's a reason for that. The more you add, the more you lose. The flexibility is the power, and the flexibility is a result of arc's smallness.

By the way, pg did exactly what you said, in the original release of arc. If you grep the codebase for "defhook", those are all the spots where the secret code was refactored out.