The TI-89T is an amazing platform that's underloved.

The Nspire is more than capable of running a Z80 emulator at full speed. You can do it in C, which is a lot easier than assembly. And you won't need a specially modified OS. TI was being just plain lazy with the whole separate OS thing.

When reading the asm output for GlaßOS, I noticed that it uses a standard way of passing parameters to functions.  Push variables, then use IX like hell.  Yes, SDCC has a drink..,er, IX problem.  I know that index registers add an extra byte or so to op sizes, and are slow.  I can see a lot of tios asm programmers not liking that variables, even one byte, aren't passed by register but by the stack.  What about IY? Its dedicated for the interrupt, just like IX.

Passing arguments via the stack is standard for C. Though, you're right, it should try to optimize more.

Reading data directly from the archive with the files feature and {} is a little slower than reading the data from RAM with {}. So, it's probably best to copy an entire room or chunk or something from the archive (using Copy()) into safe RAM (r1-r6) and read from there using {}. In other words, buffering data from the archive into RAM is often faster than always reading from the archive. On the other hand, not buffering is a lot easier. You'll need to learn to balance the two.

By the way, the OS enforces an 8 K limit on nostub assembly programs (for no good reason, of course), so if your program is larger than 8 K, you might as well make it an Ion program.

So boot 1.03 goes through a longer, more complicated (and supposedly secure I guess) process? Fun...

You missed the news frenzy that resulted when we learned about 1.03? TI switched to 2048-bit RSA. The new signature is tacked onto the end of the 2.55MP image; it's been there this whole time. Also, they actually managed to write an even slower validation routine. They're rolling 1.03 out for the new school year. This move prompted an intense frenzy of research that resulted in us discovering how to reflash the boot code, thus completely negating the change.

400th post and I dont care

Neither do I care about how many posts you have.

TilP can handle OSes just fine. I strongly recommend going USB for it, since it's much faster. However, if you have a silverlink it's worth a try since they're a little easier to keep connected.

It really doesn't matter. Either way, he'll still need to wait six or more minutes for it to validate.

I found out about Rockbox before getting my current MP3 player, a Sansa e200 which I've had for at least four years. So, as a consequence, I've had Rockbox on it pretty much since I got it. I've used Rockbox almost exclusively because Sandisk's firmware doesn't support WAV, FLAC, or Ogg Vorbis, amongst numerous other deficiencies.

The Z80 doesn't support using an MMU. There's absolutely no way for the kernel to actually enforce anything. I think most people have also generally been of the opinion that relocation tables (needed for multitasking to be used with all programs) would waste more memory than they're worth.