Learning from the Amiga API/ABI
asm-basic-coder.neocities.org
asm-basic-coder.neocities.org
There was a single fixed location in the entire system (address 0x4 aka ExecBase), and everything an AmigaOS application required was 'bootstrapped' from that one fixed address.
All OS data structures were held together by linked lists, everything was open and could be inspected (and messed up - of course terrible for security, but great for learning, exploring and extending).
Everything I learned about after it was a huge disappointment, including Mach. Particularly because it demystified the OS. Just a bunch of of lists, and due to the OO nature, they were the same kinds of lists.
Here's what a node looks like: next, previous, a type, a priority, a name.
A task? A node. With a bunch of extra state.
An interrupt? A node. With a lot less extra state.
A message? A node. With an optional reply port if the message requires a reply.
Reply port? Oh, that's just a port.
A port? Yeah, a node, a pointer to a task that gets signaled and a list of messages.
How do you do I/O? Send special messages to device ports.
No "write() system call", it's queues at the lowest levels and at the API layer.
* Assigns. Basically aliases for paths, but ephemeral and can be joined together. E.g. the search path for executables is the assign C:. The search path for dynamic libraries ins libs:. I've added basic, superficial support for assigns to my shell. It's a hack, but being able to just randomly add mnemonics for projects is nice, and not having to put them in the filesystem as a symlink somehow also feels nicer even if it only saves a few characters.
* Datatypes. AmigaOS apps can open modern formats even if the app hasn't been updated for 30 years as long as they use datatypes: Just drop a library and descriptor file into the system.
* Screens. I'm increasingly realising I want my wm to let apps open their own virtual desktops, and close them again, as a nice way of grouping windows without having to manually manage it, and might add that - it'd be fairly easy, and on systems that don't support it the atoms added would just be a no-op. The dragging was nice to show off at the time, but less important. Ironically, given the Amiga was one of a few systems offering overlapping windows when it launched, screens often served as a way for apps themselves to tile their workspaces on a separate screen/desktop, and my own wm setup increasingly feels Amiga-ish - I have a single desktop with floating windows and a file manager, just like the Amiga Workbench screen, and a bunch of virtual desktops with tiling windows.
In terms of the API, one of the things I loved was more something that evolved: The use of "dispatch/forwarding" libraries, like e.g. XPK, that would provide a uniform API to do something (like compression) and an API for people implement plugins. So much of the capabilities of Amiga apps are down to the culture of doing that, and which Datatypes was an evolution of, that means the capabilities of old applications keeps evolving.
You can still have that Amiga feeling on old PCs by using AROS: https://aros.sourceforge.io/
On Macintosh, the whole GUI ran practically in the active app's event loop. The whole system could be held up by an app waiting for something.
Microsoft made the mistake of copying Apple when they designed MS-Windows. Even this day, on the latest Windows, which although it has had preemptive multitasking since 1995, a slow app can still effectively hold up the user interface thus preventing you from doing anything but wait for it.
When Apple in the late '80s wanted to make their OS have preemptive multitasking, they hired the guy who had written Amiga's "exec" kernel: Carl Sassenrath.
Could you explain what you mean here? If you were to make your event loop or wndprocs hang indefinitely it would not hang the Windows interface for the rest of the machine, it would just cause ANR behavior and prompt you to kill the program. As far as I can remember it's been that way since at least Windows 2000.
(edit: and to be clear, I did read the article and see what it said, but without more detail I'm not 100% sure what it really looks like in practice, and why it would be less likely for applications to have situations where they become unresponsive.)
I remember the Amiga had the checkered beach ball bouncing demo and others copied it, then on the Amiga they opened up several copies of the demo all multitasking and bouncing at the same time.
The only downside of the Amiga was the dreaded Guru Meditation Errors when memory went corrupt or something. IIRC AmigaDOS/Workbench had no protected memory features.
This was a limitation of the original MC68k CPU architecture. Though the Amiga operating system was indeed designed to leverage a single address space throughout, which made it significantly harder to retrofit memory protection after the fact.
AFAIK Windows is supposed to boost the CPU priority of the UI during user input, but apparently that doesn't work.
AmigaOS also boosted the CPU priority of the UI during mouse movement, except it actually worked.
PS: and instead of fixing the issue from the ground up in the OS (which admittedly is probably impossible) the VS team instead added a feature called 'Low Priority Builds':
[1] https://developercommunity.visualstudio.com/t/Limit-CPU-usag...)
[2] https://devblogs.microsoft.com/cppblog/msbuild-low-priority-...
That being said I preferred the Amiga.
The Amiga was a bargain in comparison, but it was not without its flaws, like all early machines. I had an A500 with a 1084 monitor, and the flicker at high res was bothersome to me. I later upgraded to an A3000 w/VGA monitor, and it was a vast improvement. I ran at 640x400 for everything at that point.
I think you are underestimating the price of "flicker fixers" at the time. I looked up the price of a Microway flicker fixer in an old Amiga World from 1988: Over $500. You also had to add an a VGA monitor: another $400.
Still way, way better than a 2-color mac at about the same price. And it wasn't even necessary, the Amiga 1000 was amazing without it.
The Amiga was ahead of its time in many ways, and the pre-emptive multitasking was fantastic, but claiming it was some paragon doesn't help anyone. If you wanted a fun home machine attached to a TV, it was great. Even a fun home machine attached to a monitor. If you wanted a business machine with a monitor, it wasn't the safest or best choice, if only due to a lack of software.
16 colors is still way better for games than 2 color black-and-white. The fact that the Amiga could achieve 4096 colors in a world where 16 colors was the norm, was astonishing. The resolution did not matter. The capabilities mattered, and the Amiga was far more capable in every way than the Mac. The mac had 1-channel 8-bit sound, the Amiga had 4-channel stereo 8-bit sound, and was capable of 14-bit sound. So go nitpick some more if you want to, I don't care, but I won't be responding to you anymore.
In what universe 512x342 was better that 640x400 ?
There was an add-on called a "Flicker Fixer" that cached the video signal and emitted VGA-signals at twice the pixel clock and horizontal refresh rate. The Amiga 3000 had one built in.
The ECS and AGA chipsets supported "Productivity mode" that emitted VGA signals but ECS supported only four colours in this mode. All games used the TV modes. "Multisync" monitors that could switch between VGA and television refresh rates were expensive, so few people had them.
Also remember the Amiga was competing with the Mac II line for most of its life. Yes, it was much more expensive... but we are comparing specs, and you could get Mac II displays that supported 256 colors out of 16 million (24-bit.) The Amiga didn't have 24-bit color until 1992.
Nope, at the time the flicker fixer was about $100. The monitor didn't cost that much either. I had both, I was not rich, I was a kid who saved up some money.
>Also remember the Amiga was competing with the Mac II line for most of its life.
We're talking about the original Macintosh computer released in Jan 1984, and the original Amiga 1000 released in July 1985. Don't move the goalposts.
The Mac II was not released until 1987, and it had a 68020 CPU. The system you should be comparing it to is the Amiga 2000 with 68020 card. The Mac II was released at a ridiculous price of $3700. It really makes the Amiga look like a bargain - the price for an Amiga 2000 with 68020 was $1495, you could easily buy two of them for the price of one Mac II.
>The Amiga didn't have 24-bit color until 1992.
I don't care? The topic was the original Mac vs the original Amiga 1000. Not wherever you want to take this conversation.
Fair enough on the Mac II. They were very expensive and a rip off! Also, early Mac OS sucked.
Technically the Amiga could display a rock solid hires picture but only on a special monitor that I personally never saw.
The priority on the Mac was to have a high quality monitor for black and white graphics. They put a lot of effort into drawing libraries to make the most of in-built display.
The result was that the Amiga was perfectly fine for playing games or light word processing but if you actually needed to stare at a word processor or spreadsheet for 8 hours a day you really wanted a Mac.
Apple in the 90's was cirlcing the drain, nobody wanted an overpriced black-and-white computer except die-hard apple fans, and Apple only exists today because Microsoft bailed them out. Too bad Microsoft didn't invest in Amiga instead.
There were even C64, Apple II, IBM PC-DOS, and Atari ST emulators for the Amiga.
And the earliest ARM machines ran rings around the Amiga because they had a custom-designed RISC CPU, so they could dispense with the custom co-processors. (They still cost a lot more than the Amiga, since they targeted the expensive education sector. Later on ARM also got used for videogame arcades with the 3DO.)
By contrast, there's a story about some Microsoft engineers taking a look at the Macintosh and asking the Apple engineers what kind of custom hardware they needed to pull off that kind of interface. The Apple guys responded, "What are you talking about? We did this all with just the CPU." Minds blown.
The designers of the Mac (and the Atari ST) deserve mad credit for achieving a lot with very little. Even though, yes, the Amiga was way cooler in the late 80s.
I know this first hand, because I got my first email address with CompuServe, running their software under emulation, while using my Amiga's dial-up modem. (I had to sneak the Mac ROM images from the computers at school...)
This was due to several factors, chief of which was that the SC2000 Amiga port was made under extreme time pressure and, probably, very low budget. Later patches alleviated that to some degree, but patching your game in 1993? Who did that? What you got on your floppies was usually what you were stuck with barring some extreme cases of negligence.
"no dynamic linking" (by implementing dynamic linking)
"no zombies" (as long as your programs aren't buggy)
I fail to see any meaningful distinction from what we have today. If it was more reliable, it was due to being smaller in scope and with a barrier to entry.
On modern Linux systems you can even do separate sets of memory permissions within a single process (and single address space), with system calls needed only at startup; see `pkeys(7)`.
https://www.man7.org/linux/man-pages/man7/pkeys.7.html
(note however that there aren't enough pkeys available to avoid the performance problem every microkernel has run into)