@simontatham when I first started using Linux as my main operating system, I had an extensive custom configuration for the window manager Sawmill. Then it got completely broken by an upgrade. I've never done more than the simplest configuration for any program since.
@pozorvlak oddly enough, that's _my_ window manager, and I too have a lot of complicated Lisp configuration for it.
But I must have come in after the breaking change you mention. Certainly I started using it after it renamed to Sawfish, so if you're calling it Sawmill then I guess the breakage happened first out of those two things.
@simontatham I think they happened at the same time!
@simontatham @pozorvlak Odd. I used it as Sawmill and I don't remember a vast change, but then as you know my rcfile doesn't do all that much. Maybe I got lucky.
@pozorvlak @simontatham This is something that I feel like has gotten a lot better in the LLM world. "Take this existing setup and move it over to the new API in a mostly mechanical way" is a task that LLMs are pretty good at.
It's still a pain in the ass, but I do think it's taken a bunch of the sting out of breaking changes.
@pozorvlak @simontatham (This doesn't of course fix the bit where you may also have to update your mental model of the software significantly. It's only helpful when there's a mostly equivalent set of functionality that changed its shape)
@DRMacIver @pozorvlak flipping that round, this points out a new risk this decade of large-scale breaking changes. One person using an LLM for a job like this may be able to doublethink their way around the environmental cost, but someone introducing a change that breaks 100 million users' scripts might well be causing _most_ of them to throw an environmental disaster at the problem, which is a far larger cost!
@simontatham @pozorvlak I'm pretty happy to singlethink my way around the environmental costs (which are comparable to just running the computer the software you're upgrading is on...) but probably not interested in arguing the case.
@simontatham @pozorvlak There are genuine environmental concerns around training (mostly insufficient grid capacity/storage and the resulting short-term significant increase in non-renewable energy), but inference isn't any more of an environmental disaster than normal computer use.
Jun 30, 2026, 3:50 PM@pozorvlak @simontatham all my config is in #git, and the script that applies the contents of the repository to my home for is unreasonably portable. It even works on #HPUX! If my config doesn’t apply cleanly anywhere I make it more portable immediately.
@DrHyde I also keep my (minimal) dotfiles in git, but that doesn't solve the API-breakage problem - if anything, it makes it worse, if the same dotfiles repo must work across many machines!
@simontatham
Jun 30, 2026, 10:26 PM@pozorvlak @simontatham I’ve had to go with platform-specific config or wrapper scripts a few times but nothing major. Perhaps I just got lucky!
@DrHyde or the equivalent of `source ~/etc/bashrc.$hostname` for host-specific stuff.