Neal Wilson ✅ @neal@social.typica.us

Finished up some changes in progress. I think I'm maybe not too far off putting out a demo showing what I've been working on, though there's still quite a lot of work to do before I'll be confident in putting out a proper release.

The shop needed more orange and lemon peel. The place I get that from had a deal for free barbecue rub which the shop has no use for, but I can use it at home.

New video scripted. Will try to get that put together some time over the next week or so.

Star Trek prequel series pitch: "2309" It's all about Klingon blood wine and the making of its finest vintage.

The problem with the bug that I'm trying to track down is that it should be impossible.

There's a problem in not my code causing a program to crash only when a debugger is attached. This is making it rather difficult to find out where a completely unrelated problem is.

Made a few more tweaks on the drawing code. For high quality data and reasonable settings it looks pretty great now on both of the machines I've been testing this on. I might do some more tweaks later.

New drawing routine looks a lot better than the old one except in cases where the data going in is especially awful. Unfortunately, I know there are people who will use this with hardware setups that generate that sort of especially awful data so I'll have to decide how much I care about that.

The cat thinks I should skip dinner and pet the cat. I don't think this is a great plan, but will try multi-tasking as a compromise.

Estimated ship date on some hardware I ordered ages ago has been moved up a week, but is still a couple weeks out so I'll believe it once it's been delivered.

(the algorithm is nothing original, it was just faster for me to rework the details on paper than it would have been to check reference material)

I think I want to rewrite one of my drawing routines. The way I did it first has the advantage of being fast and using very little memory, but has the down side of relying on implementation defined behavior which means it looks fine on my laptop and kind of trash on the machine at my roaster. Sketched out a replacement algorithm that doesn't rely on implementation defined behavior, is likely to be a little more flexible on most systems, but does require more memory. Still plenty fast.

I think everything else on my to do list is more involved than I want to get into tonight so it's time to pack up, go home, and hug a cat.

One line fix for a purely visual issue that's been bothering me. That change should be the default behavior, but it doesn't seem to be.

Added another feature after today's coffee roasting. It'll be a little while before I need to use that so I hope it works. It compiles, at least.

Did some data integrity related work today, disabling an operation in conditions where it can't do something meaningful where previously the program just crashed. There's one chunk of data that I'm not sure how I want to handle it yet. I need either another integrity check or preferably just rework some things such that all possible states become valid. Ideally I could extend the all states are valid approach to all the data, but I can't do that right now for back compat reasons.

Two other changes also got in today, but they're very minor QoL tweaks that I could do in 6 or 7 SLoC.

There are some other possibilities here as well which might require a bit more work, but my hope here is that it takes a feature that's already been proven to improve batch to batch production consistency and makes it even more powerful.

Use case #2: You're doing an experiment where you want to vary a part of the roast in the middle, while keeping the start and end the same. Having a later translation point makes it easier to match the end without having to design the whole modified plan up front. (which you should, but we all know coffee roasters aren't great at actually doing things like that in the real world for the most part)

Use case #1: You're far enough off the plan at the start of the roast, think you're running parallel, but want an early translation to make sure you're on the right track before the auto-translation at the start of taste-able chemistry kicks in, making it easier to match the significant parts of the plan.