miniblog.

LLMs are surprisingly good at reducing crash samples. I've had success with "this project crashes my static analysis tool with the following command, try to shrink the repro whilst preserving the crash. Commit each smaller version to a branch".

I'm beginning to think this last sentence was a mistake. Lots of software fixes crashes in new versions. Perhaps I should have a "N releases since last fix for a crash" count?

Tried fuzz-testing difftastic this morning (using cargo fuzz), and didn't find any crashes. I guess that's a good thing? I was slightly disappointed. Judging by the output, I think the tree-sitter parsers were exercised much more heavily than the tree diffing logic.