As expected, Revision 2025 was a blast again, and also as expected, I did not magically find enough time to finish the little entry I was working on.

I’m burning to check out all the amazing Amiga entries we had, but I’m also burning physically after all the buildup/teardown/infodesk work, so this will have to wait a bit…

I did, however, manage to do a belated release of a little birthday entry I did for JobJ’s demoparty event back in February: a numbers game. At the time, I was too stupid to send it to JobJ properly – who would have thought that Google Mail will filter out e-mails with a ZIP attachment containing a weird EXE file…? :)

Old school organizer Charlie was so kind and sneaked it into the test compo before the party (which included testing, recording, and scheduling the entry) – thanks a million for that!

With Revision’s deadlines looming and the party build-up commencing today, the time budget for a possible size-coded Amiga entry doesn’t look too good anymore. Well, that’s life!

I did, however, find a little time – between my last work day before Easter and tying up my luggage – for some compression optimization. Packing before packing! :)

Specifically, I was looking to shorten the decompression header in my bootblocks. I knew I had used some magic constants before, but that was at an earlier version of Shrinkler, and maybe I haven’t used it for salvador/ZX0 at all. So let’s take a fresh look!

What exactly do the registers contain when your bootblock gets executed?

Register	Kickstart 1.2	Kickstart 1.3	Kickstart 2.04	Kickstart 3.1
d0	00000000	00000000	00000000	00000809
d1	8000ffff	8000ffff	0000ffff	00000808
d2	00000001	00000001	00000001	00000001
d3	00000000	00000000	0000d688	0000ad48
d4	00000000	00000000	00001468	00001520
d5	00000000	00000000	00000000	00000000
d6	ffffffff	ffffffff	ffffffff	ffffffff
d7	00000000	00000000	00000000	00000000
a0	00001958	00001958	0000dfb8	0000ed38
a1	00c014e6	00c014e2	0000d54c	0000e2c4
a2	00c0185c	ffffffff	0000d530	0000e2a8
a3	00fe8b3a	00fe86ee	0000d54c	0000e2c4
a4	00001558	00001558	00003e4c	00002708
a5	00c014ba	00c014b6	00000a44	00000a74
a6	00c00276	00c00276	00001468	00001520
a7	00c014b6	00c014b2	00003e10	000026cc
pc	00001564	00001564	0000dfc4	0000ed44

Note: Of course, this is not an exhaustive list. These are merely the values for some typical configurations: Amiga 500 with 512KB of extra RAM for Kickstart 1.2 and 1.3, Amiga 500+ for 2.04, and a vanilla Amiga 1200 for 3.1

But it seems there are some specific fixed values we can rely on across different Kickstart versions!*

*) If we only need to support classic Commodore-made AmigaOS editions, that is. This will probably not work with AROS68k or AmigaOS 3.2.x.

Register	Contents
d2	Constant 00000001
d5	Constant 00000000
d6	Constant ffffffff
d7	Constant 00000000
a1	IORequest, as specified by the OS (bootblock sequence)
a6	exec.library, as specified by the OS (ibid.)
pc	Noteworthy: The 12 bytes before the entrypoint contain the bootblock header as it is stored on disk (i. e. 'DOS',0 followed by checksum and rootblock longwords

Nifty! Now let’s delete some code. With Shrinkler, we can save four bytes in the initialization code:

ShrinklerDecompress:
        ; Init range decoder state
        moveq   #0,d2           ; omit, use d5 instead
        moveq   #1,d3           ; omit, use d2 instead
        ...

And two bytes for ZX0:

zx0_decompress:
        moveq   #-128,d1        ; initialize empty bit queue
        moveq   #-1,d2          ; omit, use d6 instead
        ...

I love that each of the values 0, 1, and -1 is of use somewhere!

For both packers, the bootblock starts like this:

disk:   dc.b    'DOS',0
        dc.l    0               ; checksum gets inserted here
root:   dc.l    $20000          ; destination; usually 880=rootblock 
entry:  move.l  root(pc),aX     ; destination address
        lea     meat(pc),aY     ; start of compressed data
        pea     (aX)            ; rts will jump into unpacked code
        ...                     ; decompression code starts here

With this, my current bootblock stubs take up this much space:

Compression	Stub size	Room for compressed data
Shrinkler	168	856
Shrinkler, parity disabled	162	862
Savlador/ZX0	100	924

Note: It pays out to have build scripts ready for all compression schemes – with my current work-in-progress payload, the “winner” (in terms of best overall compression) fluctuates constantly!

Now, either I find some more bytes to spare during the party or I put some of that saved space to good use with an entry! :)

Then again, I already have my working gloves on…

New lore on “New Topaz” has dropped, i. e. the modernized sans-serif system font used in AmigaOS 2.0 and later.

During the “Inside Commodore Amiga’s History” discussion panel at VCF East 2025, ex-Commodore software engineer Peter Cherna talked a bit about the overall new look introduced in 2.0 (still called 1.4 when he started), and the new Topaz font in particular:

There was an agreement that we were going to have a sans-serif font – we still had the old Topaz in there – and then Kodiak Burns in California kind of forced the issue and picked a slightly ugly sans-serif font.

…

We found a font floating around on the not-internet at the time that was derived from Topaz. Basically, it was Topaz with all the serifs removed.

That’s how the weird 1.4 version that can be found in Kickstart 1.4 Alpha 18 came to be. Bob “Kodiak” Burns is the creator of the original Topaz fonts.

So I went and cleaned it all up and fixed a bunch of the letters and the lowercase “L” to make it not look like a vertical bar, and “I” has a little curl at the bottom. That was very contentious, a lot of people thought that it was wrong. I was very insistent because I wanted it to be distinctive for legibility.

Turns out it’s a big contribution to the personality of the new font as well.

Indeed! I remember being really excited of the modern new 2.0 look when I was a teenager, and New Topaz had a big part in that, with the new Amigas feeling way more professional and all. I loved working with New Topaz so much I even destested the old, “ugly” variant a bit – but not anymore! :)

The font that was “floating around” probably was a variant of clean.font. All the specimen I’ve seen on textfiles.com indeed have a vertical bar for the lowercase L.

Here’s a screenshot of Workbench 1.4 Alpha 15 running on Kickstart 1.4 Alpha 18. This seems to be an inbetween snapshot: The lowercase L already has its little curl, but the lowercase I hasn’t, and a lot of letters still look wonky.

Anyhow, don’t forget to watch the whole panel on Amiga Bill’s channel. They even have an original Amiga breadboard prototype on stage!

Weird: Google has notified me that search hits for “joys of feeling stupid” have gone through the roof! 250 search result clicks in a month is a lot, at least for this humble website.

I mean, it was nice blog post, but it is hardly what people are looking for when googling that, I suppose?! Even weirder: “joys of feeling stupid” doesn’t seem to be a comtemporary viral video, song, or article. What gives?

I don’t really know, but that’s okay. Fitting, even – the mystery of “joys of feeling stupid” making me feel… you know. :)

Update: Mystery solved!

A helpful visitor of this blog (thanks, beni!) pointed me to the source. It’s from an interview (in Turkish) with Beliz Günel, senior research scientist at Google DeepMind. Around 3:45, she mentions her mentor who…

…made me read one article, I remember it very much: “Joys of feeling stupid” I liked feeling stupid like that.

Now I want to dig up that article, too, of course! :) Since Günel graduated at Stanford, maybe it’s this article hosted at stanford.edu?

• The importance of stupidity in scientific research
  Martin A. Schwartz, University of Virginia

Hacker News featured a link to developer.chrome.com: “Select fields can now be styled” Interesting, let’s click this link! This is what I got:

Yes, that’s the whole page. Somewhat cool, just: Das Element and nothing else. Maybe the start of a Rammstein song? :) But I did expect to read more on the topic…

That “switch to English” button already gives a hint who’s at fault. The original English version of the page looks intact:

So, a “helpful” auto-translate service strikes again! This time, it’s not focused on mis-translating technical terms, though, but rather messing up the HTML markup… Danke! :)