The list at https://news.ycombinator.com/topcolors is definitely not "the complete set of colors users have set" as claimed. I know this because it doesn't include my top color (namely #64BFBD).
I fired off an email to Scott, frantically telling him "I think I took down HN. If you’re scrambling to figure out what’s wrong, it’s because so-and-so story is on the front page, and you can fix it by booting it off." A minute or two later the site came back to life, so I emailed saying "oh, never mind. I guess it was something else." And to my surprise he emailed back saying you were exactly right, thanks.
Presumably they either added caching, prevented it from going to the front page, froze the list for posterity, or it’s still a ticking time bomb that can take down HN again if you happen to submit it. Either way, maybe the explanation of your missing topcolor is there somewhere.
I’m just guilty without knowing the crime.
And yes my choice is there (#ffaa33)
EDIT: You can see for yourself how it's ordered here: https://github.com/shawwn/arc/blob/arc3.1/news.arc#L2611-L26...
This is the original code, which survived untouched till at least 2019 or so. I've wondered if they changed it since then.
The key is the (users) function:
https://github.com/shawwn/arc/blob/arc3.1/news.arc#L144-L145
This returns the keys of the profs* hash table (short for "profiles"). At the time, profs was populated each time the server restarts:
https://github.com/shawwn/arc/blob/arc3.1/news.arc#L88-L91
(dir profdir*) essentially returns an `ls` of the directory in which user profiles are stored. (Yes, they are (or were) stored to disk, not a database.) This was one reason it was hard to add a feature to rename users, which Dan somehow figured out.
So it takes the `ls` of that directory, giving a list in mostly random order, and loads them one by one, putting them into a hash table, whose iteration order is also random.
In modern times, profs* is populated in a lazy fashion whenever users post after a server reboot.
Arc uses "maptable" to iterate over a hash table, which under the hood used hash-table-for-each: https://github.com/shawwn/arc/blob/arc3.1/ac.scm#L1082
Nowadays in Racket, it probably uses hash-for-each: https://docs.racket-lang.org/reference/hashtables.html#%28de...
It does have a "try-order?" option, which gives a deterministic ordering if set. But whenever I set it, I sometimes get strange errors, at least on MacOS. Since it's disabled by default, I doubt they set it. Even if they did, it would be ordered alphabetically, which means your topcolor wouldn't be anywhere close to the top unless someone else whose name starts with "a" also set it.
I only know all this because I've spent five years updating Arc for the modern era over at https://github.com/shawwn/sparc
Past discussion: https://news.ycombinator.com/item?id=38586832