2025-03-17 05:09:10 <neauoire    > hey xentrac
2025-03-17 05:10:52 <neauoire    > can you remind me how to browse pavnotes without cloning it?
2025-03-17 05:11:32 <neauoire    > I'm working on a little UI in uxn for my self-hosted compiler, which is also written in itself.
2025-03-17 05:11:35 <neauoire    > https://assets.merveilles.town/media_attachments/files/114/175/793/414/222/895/original/4aa43c14781d97e5.mp4
2025-03-17 05:12:01 <neauoire    > lots of self-modification going on there ^__^
2025-03-17 05:51:33 <xentrac     > neauoire: well, in theory you could carry out the operations the Git client would carry out
2025-03-17 05:52:15 <xentrac     > GET http://canonical.org/~kragen/sw/pavnotes2.git/HEAD which points to refs/heads/master
2025-03-17 05:52:43 <xentrac     > GET http://canonical.org/~kragen/sw/pavnotes2.git/refs/heads/master which points to 91b4629a560b5656a5b760c1ded688db88db8317
2025-03-17 05:53:38 <xentrac     > GET http://canonical.org/~kragen/sw/pavnotes2.git/objects/91/b4629a560b5656a5b760c1ded688db88db8317 (since it's a loose object)
2025-03-17 05:56:45 <xentrac     > in e.g. Python zlib.decompress(open('b4629a560b5656a5b760c1ded688db88db8317', 'rb').read())
2025-03-17 05:57:08 <xentrac     > which says tree 2fe2dbf439e9a0ac5492d6fb88031a12990995e6 among other things
2025-03-17 05:57:48 <xentrac     > GET http://canonical.org/~kragen/sw/pavnotes2.git/objects/2f/e2dbf439e9a0ac5492d6fb88031a12990995e6
2025-03-17 05:59:30 <xentrac     > zlib.decompress(open('e2dbf439e9a0ac5492d6fb88031a12990995e6', 'rb').read()) gives you filenames and hashes
2025-03-17 06:01:30 <xentrac     > if you store it in a bytestring s, for example s[s.index(b' gsais-pong.md\0'):][:40] gives you b' gsais-pong.md\x00\xd64\x90\x16F"\xd9\xd7\xd13x\xf9nTh;\xa5p\xa3\xa910064'
2025-03-17 06:02:50 <xentrac     > and s[s.index(b' gsais-pong.md\0'):][15:35] gives you the hash, though in an inconveniently binary form
2025-03-17 06:04:04 <xentrac     > fixable with ''.join('{:02x}'.format(b) for b in s[s.index(b' gsais-pong.md\0'):][15:35])
2025-03-17 06:04:26 <xentrac     > which helpfully says d63490164622d9d7d13378f96e54683ba570a3a9
2025-03-17 06:04:59 <xentrac     > so you can GET http://canonical.org/~kragen/sw/pavnotes2.git/objects/d6/3490164622d9d7d13378f96e54683ba570a3a9
2025-03-17 06:06:03 <xentrac     > and then zlib.decompress(open('3490164622d9d7d13378f96e54683ba570a3a9', 'rb').read()) gives you the actual text of the document in question
2025-03-17 06:06:36 <xentrac     > so you *can* browse pavnotes2 without cloning it
2025-03-17 06:07:18 <xentrac     > but it might be more practical to clone it
2025-03-17 06:09:24 <xentrac     > the compiler dialog box looks very maccy, but I'd think the ideal UI for a compiler would usually be that you attempt to run your program and it either produces compile errors or runs
2025-03-17 06:12:06 <xentrac     > so the compiler as such is as invisible as the BIOS or the keyboard controller except when something goes wrong
2025-03-17 06:19:19 <xentrac     > your tastes may differ, of course
2025-03-17 06:21:59 <xentrac     > there may be advantages to not requiring the deployment environment to have the compiler, for example, or you might want to guarantee rapid startup
2025-03-17 14:12:49 <veltas      > In a lot of 8-bit code if you have some machine code in RAM and there's a small number of 8-bit globals it relies on, then you just use immediate values and overwrite the immediates within the instruction, instead of using separate globals
2025-03-17 14:13:34 <veltas      > I'm sure that's used in demos, but it's really just a useful way of managing many kinds of globals in 8-bit, probably seems hacky now
2025-03-17 14:14:16 <veltas      > In x86 I've seen reports that this is a lot slower, so I wouldn't do it on a modern CPU
2025-03-17 14:14:52 <veltas      > I think I remember hearing that gforth until not even that long ago was doing this for certain kinds of globals, and obviously had never profiled it
2025-03-17 14:20:43 <xentrac     > veltas: yeah, that's what the pong code is doing.  it only needs to run
2025-03-17 14:21:30 <xentrac     > uh,
2025-03-17 14:24:05 <xentrac     > it only needs to run about 160 instructions per screen frame, if I'm counting right
2025-03-17 14:24:44 <xentrac     > the rest of the time it's busywaiting on the VBI
2025-03-17 14:25:05 <xentrac     > so it is okay if it is a lot slower
2025-03-17 14:27:00 <xentrac     > it's a common technique also on neauoire's uxn virtual stack machine, where they *do* use it for speed, because uxn is invariably interpreted and so it doesn't invalidate any caches or anything
2025-03-17 14:27:49 <xentrac     > 120 of those instructions are erasing and redrawing the paddles
2025-03-17 14:29:33 <xentrac     > the instruction that updates the ball position is lea si, [bx+si+0x2]
2025-03-17 14:30:58 <xentrac     > bx is the vertical movement, the 2 is the horizontal movement, and the immediate 2 gets modified when the ball bounces left or right
2025-03-17 14:31:51 <xentrac     > so you get two additions in a single instruction
2025-03-17 14:49:39 <veltas      > Interesting stuff
2025-03-17 15:38:18 <xentrac     > I was sort of disappointed to find that the winning entry seemed to be almost entirely micro-optimization tricks rather than conceptual insights, but maybe I'll look through one of the larger entrants to see what kind of conceptual overcomplications he avoided
2025-03-17 15:39:29 <xentrac     > like, using the same code with a different color argument to draw the paddles and to erase them might be a thing that other entrants didn't think of
2025-03-17 15:39:58 <xentrac     > or using XOR to draw and erase the ball so the draw and erase code is the same
2025-03-17 15:40:47 <xentrac     > the rules for some reason prohibited clearing the screen every frame
2025-03-17 15:42:20 <xentrac     > contest context: https://www.hugi.scene.org/compo/compoold.htm#compo3
2025-03-17 16:15:46 <neauoire    > for things like flipping the orientation of a ball going back and forth across a field
2025-03-17 16:16:17 <neauoire    > it's pretty nifty to change the instruction on collision instead of checking for direction every frame
2025-03-17 16:16:40 <neauoire    > in a DVD screensaver thingy
2025-03-17 16:16:40 <neauoire    > https://git.sr.ht/~rabbits/uxn/tree/main/item/projects/examples/demos/dvd.tal
2025-03-17 16:17:46 <neauoire    > *instead of loading the direction register on every frame