miniblog.

There are *so many* ways that reading a text file can fail. Maybe it doesn't exist, it's a broken symlink, it's actually a directory, it's not the encoding you expected, or perhaps you just don't have the correct permissions. Reporting good errors is surprisingly labour intensive.

One interesting consequence of the rise of LLMs: there's more demand for tools that handle untrusted input. Arbitrary HTML+JS can be safely run in a browser. Lean can check an arbitrary proof. These work really well with an LLM that can be wrong, but sometimes gives exactly what you want. Are there other tools in this family?

I've been experimenting with an 'evaluate up to cursor' mode for my PL project. I love evaluating self-contained snippets in Lisp, this generalises the idea. The interpreter remembers the arguments when you run tests, then can re-use them when you say 'eval up to here'. What do you think?