All the assembly code, and the C harness which loads the textures etc, was hand done by me, and period correct. That's where the fun is!

Outside of that, there are so many anachronisms. I actually built the test geometry as a quake map, and processed it using Quake's BSP/lightmap tools. That's 1996 vintage.. except I'm using a newer Quake editor (Trenchbroom, 2013) and modern forks of the tools. All natively on a modern system with a 4K monitor. Building and non-performance testing of the DOS code happens in a Windows XP (2001) VirtualBox (2007) using its' Shared Folder feature.

I might have time to pick this up again soon. My post-processing of the Quake .bsp presently only handles axis-aligned 64x64 surfaces with some hardcoded texturing assumptions. I'll want a full .bsp-to-isometric-data "renderer" - either written from scratch or somehow derived from Quake's source. But this is all just to enable the 286 assembly code noodling, which is what I want to spend time on!

So I'll probably use Claude to assist with the tooling side. By "assist" I mean "do as much as possible to produce the file format I specify". It doesn't seem much worse to me than all the other modernisms I'm already relying on outside of the core task.

I could still read the text! And it's useful to read texts and understand 95%, or 80%, or even 50%! IMO "fully understand what you're reading" versus "can't understand what you're reading" is a harmfully discouraging dichotomy. You can often gloss over such things rather than struggle with them.

EDIT: I recall reading that the Netflix client can continuously select between multiple content caches. I'm guessing they do this because it's a quality-of-service and capacity win over making a "best guess" at the start of a session, and sticking with it. It should also enable transparent recovery from a broken or slow cache node. If you test in a busy place with multiple caches, the loss of one needn't be a big deal.

I'm guessing dBase was itself built using Microsoft's compilers, so pulling in other code from the same compiler was plausible. Not a product choice re: the Turbo compilers, just a disjoint path dependency between two acquired product lines (Ashton-Tate's dBase, and Wizard C).

This is the reason that Clipper, a third-party dBase compiler, could only link against .OBJs from particular versions of Microsoft C. Clipper compiles to p-code whose interpreter is implemented in Microsoft C; "linking" a Clipper program is actually linking a Microsoft C program with a static array comprising the Clipper compiler's output. So you can mix in a .OBJ expecting that version of the Microsoft C runtime library. Not a .OBJ expecting the Turbo C++ runtime library.

Sometimes a "punch in the face" is useful. Makes you re-evaluate things. Being punched in the face all day, every day, is just tiring.

"Six paediatric health-care professionals were recruited to swallow a Lego head."

2. Separate entity spun up to focus on Thunderbird only. Now you can support Thunderbird development directly.

3. New separate entity is now in the business of extensible AI clients?

EDIT: I went back and read the launch announcement [1]. I'll concede it does say "will also allow us to explore offering our users products and services that were not possible under the Mozilla Foundation" which could mean anything, really. And this development was funded by a Mozilla grant, importantly not by Thunderbird donors. I'm still struggling to not see this as a distraction from the core mission. I wish they'd spun up a new entity instead.

[1] https://blog.thunderbird.net/2020/01/thunderbirds-new-home/

Conceivably LLMs might be good at answering questions from an unorganized mass of timestamped documents/tickets/chat logs. All the stuff that exists anyway without any extra continuous effort required to curate it - I think that's key.

Their stuff isn't running on top of Linux on the Pi.