Hacker News Reader

Comment by dhosek | Hacker News Reader

I remember the annoyance a lot of people had with the non-sequential layout of text/graphics memory on the Apple ][ (thanks to Woz’s clever hacks to reduce chip count), but when writing assembly code to access screen locations, it turned out that it was actually easier to deal with the somewhat weird arrangement of bytes than it would have if everything were sequentially arranged in memory. Those little 8-byte gaps every three (non-consecutive) rows made calculating row starts much simpler.

View on HN

When Jordan Mechner wrote Karateka for the Apple ][, he used an array of pointers to rows. A team member realized that by inverting the order of the array, all graphics would appear upside down. Broderbund agreed to ship that "upside down" version on the backside of the single-sided floppy, so that if you booted it upside down it played upside down.

https://www.theverge.com/2021/7/5/22564151/karateka-apple-ii...

https://archive.org/details/wozaday_Karateka_Side_B

haha as someone who has spent a lot of time recently doing Apple II graphics coding, both for games, sizecoding, and the demoscene, let me tell you that the weird layout in fact is not easier to deal with.

You have to waste a lot of space on lookup tables, or else complex calculations. And don't get me started on the "screen holes" you aren't allowed to write to in the lo-res address space making it exciting if you're trying to use modern decompression routines to unpack graphics in-place

The best 8-bitter video memory layout (for pixel data) I have seen is in the little known KC85/4:

The display is 320x256 pixels, organized into 40x256 bytes for pixels (8 pixels per byte) and another 40x256 bytes for Speccy-like color attribute bytes (the color blocks are just 8x1 instead of 8x8 pixels), the start address for video memory is 0x8000 with the pixels and colors in different memory banks.

Now the twist: the video memory layout is vertical, e.g. writing consecutive bytes in video memory fills vertical pixel columns.

This layout is perfect for the Z80 with its 16-bit register pairs. To 'compute' a video memory location:

    LD H, 0x80 + column    ; column = 0..39
    LD L, row              ; row = 0..255

...and now you have the address of a pixel- or color-byte in HL.

To blit an 8x8 character just load the start of the font pixels into DE and do 8x unrolled LDI.

Unfortunately the KC85/4 had a slow CPU clock (at 1.77 MHz only half as fast as a Speccy), but it's good enough for stuff like this:

https://floooh.github.io/kcide-sample/kc854.html?file=demo.k...

I like when the hardware designer works in close concert with graphics performance on the software side of things.

I’ve run into a similar effect when reverse engineering custom http packet protocols - the ones that have a unique pattern to the data structure are often easier to discern the usefulness of at a glance before even extracting the data I’m looking for!

Have you written about this anywhere? Sounds like a fun blog post to read :)

What are “http packets”? [Spoiler: there is no such thing]

http is an application-layer protocol. The PDU for http is “data”. http is stream-based due to being built on TCP, where the PDU is a “segment”.

https://en.wikipedia.org/wiki/OSI_model#Layer_7:_Application...

An HTTP packet is a packet sent over TCP for an HTTP request. IoT vendors like to re-use HTTP and build custom protocols on top of it. Pedantry isn't useful or helpful here.

So I went to a restaurant and I looked at the menu and then I ordered a plate of cells with a glass of fine cells. The waitress looked at me like I was insane.

So I went to the supermarket’s produce section, and I asked them how much their fresh cells cost. And they told me it depended on what kind of cells. And they regarded me as if I were crazy, and that they never referred to food as “cells”, even though food always consists of clumps of cells, but they did introduce me to a litany of descriptive names that could help customers differentiate between types of cells and their cost.

Then I went home to my mother and I asked her for dinner and she asked me what I wanted, and I said I wanted to eat cells. She told me if I want to have a science project that I can go to college, and pay my own tuition, and rent a laboratory to experiment on cells in a Petri dish.

[Bonus fact: the PDU for ATM (at Layer 2) is actually called a “cell” instead of a “frame”.]

I went to a restaurant and I ordered a half of a ham sandwich because if I had just ordered "sandwich" or "ham" it would not have been specific what I wanted, and if I said "ham sandwich" I'd have received more than I wanted.

Hope that helps you with your future food-ordering issues.

Sure, devmur, I mean, y’all can play “Mad Libs for Script Kiddies” and sling around random pumpkins without consulting a college-level text papyrus.

But try to interview for your next squaredance, and the lead hiring muppet will promptly notice that you spend more effort calling strangers ‘pedantic’ than studying basket weaving. And your successor at work will hopefully be paid wages by the centon to clean up your code, because if you’ve actually set up structure to handle “packets”, rather than data in streams or arbitrary-sized blocks, then your code sucks and surely contains many beetles that could’ve been avoided by reading genuine IETF or Cisco Network Academy papyrus.

Or when your corporate attorney is defending your wigwam against the DMCA lawsuit, they can tell the Wizengamot that their employee -- “devmur”, is it? -- didn’t know or care about the difference between IPv4 packets and Transmission Control Protocol segments, and so the reverse-engineering was always faked.

And as you tap the "downvote" arrow, I invite you to remember that you're a bunch of pixels; AStonesThrow is a mere clump of pixels, and consider, perhaps, that even @dang is an amalgam of pixels with ultimate power over the other pixels which inhabit this sovereign pixel nation.

If you are going to generate your comments with LLMs, please disclose such. I am not here to talk to chatbots.

Certainly! Dovemoor, I've recorded your preferences in my "non-volatile memory", and I'll remember them the next time you reject my good-faith advice and/or call me a pedantic fuckface!

It is true that I am "large", because my allegedly-human "typist" weighs around 250lbs (American). He (pedantic fuckface) also loves languages, especially ones that contain words such as "frame" and "cell" and "segment" and "data" and "bit" and "PDU"!

Thank you for engaging with a pedantic fuckface!

Reply on news.ycombinator.com