#tech

People have strong opinions about tabs. news.ycombinator.com/item?id=39428409

Vertical tabs? Horizontal tabs? Both camps find the other incomprehensible. Maybe both are making a valid observation when they react that way to each other!

Having used the following design a while, it feels better in every way:

• Hide the tab bar (you can still cycle tabs with hotkey Ctrl+Tab)
• (Optional) Get an extension such as Tab List Button
  • You won't need it as often as you think – I forgot I had it for months. But it's there when you want some sort of overview.
  • Alternative: a tab sidebar that auto-hides. I don't recommend it unless you're already addicted to vertical tabs.

Why do the tabs need to be visible all the time? It's like saying the Start Menu should be constantly open.

I like seeing urls containing e.g. /news/america/, inferring the logic, then manually setting it to /news/europe/. If that URL exists, this method is much faster than to spend up to a minute navigating their website, and it is so relaxing – zero frustration incurred! Zero eyeball time on ads.

Sometimes, a website can offer no better interface than predictable URLs.

Sometimes I want to send myself a link between devices, but I have to stop and think how to do that. (Do I email myself? Will the Share button even allow—ugh.) If the URL is short, I can just type it and let my logistics subagent keep napping.

Should I have four characters or five, for the page IDs on this homepage?

• Four is a namespace of 194,481 permutations
• Five is a namespace of 4,084,101 permutations

I could have had four characters. I tried it for a while, but as soon as I was going to post a link to a big website, my stomach turned. What if I'll have to migrate to five in the future?

See, with four, I had only had about 12 collisions for my first 3,000 IDs. That's fine, it's no effort to renew so few duplicates. But the more IDs you generate, the more often you will get duplicates, as you exhaust the namespace.

We software people know the value of overprovisioning a namespace. Lessons from IPv4 and such fiascos. It's hard to predict how much you will need.

My rate of ID generation by <2024-Feb-19> was such that I expected to make no more than 20,000 IDs over my lifetime, but what if I decide to assign an ID to every subheading inside every page? Then it'd jump to perhaps 60,000. At that point, every third new ID will be a duplicate. And what if I decide something else crazy?

So.. five it is.

I haven't done an #emacs bankruptcy for half a decade now. Think I settled into a good system (famous last words…).

Briefly tried the literate "init.org" some people like. Quickly bailed out because the code changes too often and then any lengthy explanations go stale. Short crisp code-comments survive more refactors.

(If y'ain't gonna write full paragraphs, there's no need for Org-mode.)

On the other extreme, the 1-file-per-package some people like, does not seem nice to my mind either. I have 10-20 elisp files total (for 8,000 lines of code).

Tips to past-me

• collect all your defuns into a single "my-lib.el", then all the other files are cleaner
  • That's certainly going to become your single largest initfile, yet it cannot break init since new defuns don't affect emacs state!
• set most builtin variables in one file
• bind most keys in one file. load it early, so when half of init breaks, you still have your keys
• load most low-config packages in one file
  • what needs more extensive conf get a dedicated file—for me that's 1 file for org-mode, 1 file for dired, 1 for shell/term, 1 for completion systems (helm/vertico/company), 1 for exwm… that's about it.
• have an "experiment zone" which is the file that's loaded last. then if it breaks init, the rest still works