<InPhase>
teepee: I think the 100% cpu issue has some sort of loop with a mem leak in it, because it really climbed up there in memory when I forgot I had that window open.
<InPhase>
But slowly.
<buZz>
there's paying openscad customers? :D :D
<buZz>
you could make it a patreon
<teepee>
nope, backporting to ancient distros would need to be a fully paid thing
<teepee>
sponsor/donations are great for various expenses but backporting things is pretty much a full time job once you hit a certain age threshold
<InPhase>
Yeah. That's a more serious time investment for little return. Anyone who thinks they absolutely need it can either figure out how to do it, or compensate someone to do that larger amount of extra work.
<InPhase>
We already target a pretty nice platform base among the supported list.
<InPhase>
Although I wouldn't object if someone wanted to volunteer to support the ancient stuff, as long as it doesn't result in choices that impair upgrading to newer libraries, standards, and features.
<InPhase>
Those latter things being why it's important to keep the list periodically pruned of the old things, at least from the "target" list. I don't mind an "okay, great, still works" list being larger. :)
<teepee>
mostly driven by OBS anyway, they dropped the repo, so we can't get new upstream packages
<InPhase>
Speaking of reasonable... I'm attempting to unreasonably render a 1 square meter patch of 1 mm thick interwoven printable thread.
<InPhase>
I think Manifold is going to actually succeed at this.
<InPhase>
I'm sadly going to have to turn off the feature after I'm done playing with it, because it's unpredictably unreliable, but my goodness is it ever fast.
<Guest8422>
This may seem unreasonable, but are there any forks of openscad that use a operand stacking interface instead of code? I am quite terrible at writing & editing code.
<InPhase>
Guest8422: Like BlockSCAD3D
<teepee>
blocks-cad
<teepee>
and I just saw something that sounded like that an hour ago... let me check
<InPhase>
Guest8422: Like, click on the Snowman, and then "Create My Own"
<InPhase>
teepee: A whole second version of the same idea?
<teepee>
hmm, those mastodon links don't make much sense for copy&paste
<teepee>
unbelivable
<teepee>
although I doubt it's the 2nd
<teepee>
there's also some appliation thing, but I don't remember the name. nodescad or something?
<InPhase>
:( Manifold ate my thread render. So close. Only 4 minutes to create... nothing.
<Guest8422>
Wow .. that's amazingly fitting to what I asked isn't it?
<teepee>
graphscad
<InPhase>
Guest8422: Honestly I think it's a terrible approach for general use. But it's really nice as an educational tool, and it fits well into people brought up on that MIT scratch notion.
Guest8096 is now known as Sauvin
<InPhase>
Guest8422: A bonus is that they all seem to be targetting the idea of having exportable normal-form .scad code, so that people can transition away from it.
<teepee>
it would be great similar to those FPGA editors
<teepee>
where you can get complicated code pieces as nodes
<InPhase>
teepee: Yeah, I used that for a project...
<teepee>
only low level nodes gets messy fast
<Guest8422>
I have memory issues. The approach is not the best, but with vanilla openscad I limit myself to a very small portion of what it is capable of because I have to consult the cheatsheet for everything constantly.
<InPhase>
teepee: It's one of the few areas where I accept the graphical approach, because it looks like circuits. But it does indeed get out of control fast.
<teepee>
like if you have a node that's a project box and you just need to connect connectors and stuff
<InPhase>
teepee: Although Xilinx ISE has some reasonable features for abstraction by building your own subcomponents, so you still essentially get "functions" with the graphical approach. So you CAN scale it.
<InPhase>
teepee: But generally text code is better.
<teepee>
uh, and firefox claims the download is dangerous
<InPhase>
teepee: Because it's an http download on an https site.
<teepee>
ah, ok
<InPhase>
Which is in fact dangerous in theory.
<teepee>
binary seems to be from 2017
<InPhase>
Ah, that particular one looks more like a graph and less like MIT Scratch.
<Guest8422>
still, it's a better start than trying to compile and export things out of .werkkzeug3
<teepee>
I still don't see any license statement
<InPhase>
teepee: Yeah. I don't think the author knows about those.
<teepee>
but I guess if anyone would want to pick it up it would be easy to ask
<teepee>
that's not going to run on python 3
<InPhase>
teepee: Do you think we have any sort of accidental maximum number of facets for rendering?
<teepee>
I don't know, but it's possible I suppose
<InPhase>
teepee: My square meter of fabric reported 24.9 million facets before it failed to render. I'm trying to figure out if this is the same unstable bug, or we are discovering new limits.
<InPhase>
I don't think I ever pushed one this far before.
<teepee>
that's a lot :)
<InPhase>
My half meter by half meter of fabric with 6 million facets rendered fine.
<teepee>
did you see that multi layer chain mail print?
<InPhase>
Ravo: I turned it down to 60mm by 60mm there, because if you don't have the new master branch build, the size scale I just rendered is quite unreasonable.
<teepee>
I wonder if I bookmarked that
<InPhase>
Ravo: There's a reason I never did it before. :)
<Ravo>
That explains why it rendered faster than I could time it.
<Ravo>
A legitimate render on that is taking far longer than the preview. I must be on an ancient version or something.
<InPhase>
There are some massive performance boosts on the way relative to the last release.
<InPhase>
But, they are not all ready yet.
<Ravo>
The latest release on Fedora is version 2021.01. If those numbers mean what I think they mean - God help me.
<InPhase>
That's our last release.
<InPhase>
We're slightly overdue.
<Ravo>
It's OK. I do a lot of stuff with Wings3D as well, and they have a similar release schedule.
<InPhase>
There's so much high quality development going on at once that it's hard to know where the right cutoff point is.
TheAssassin has quit [Remote host closed the connection]
<Ravo>
I asked my local copy of llama and it said openscad can only handle 1024 facets. I still have that carpet mesh open at the same time. I thought AI was the future?
TheAssassin has joined #openscad
peepsalot has joined #openscad
snaked has joined #openscad
<kintel>
InPhase It sounds like we're increasing that world record by a couple of orders of magnitude with the new Manifold developments
<InPhase>
kintel: Yes... It's flaking out on me at 24.9 million. But I think it's the new feature being unstable rather than a limit.
<InPhase>
It's only about 4 minutes at that scale.
<InPhase>
And only about 14GB of RAM.
<InPhase>
I could happily push it much higher if I could get it to actually preserve the result.
<kintel>
I feel once we have a good CSG evaluator, the maximum is really a function of number of intersections, not number of faces. CGAL is a degenerate case at it's build out of infinite intersecting planes AFAIU
<teepee>
correlates a bit with what ochafik wrote that he saw things break with $fn = 1000
<kintel>
..but I haven't had the chance to read up on the Manifold algorithms..
<buZz>
whats llama?
<InPhase>
kintel: Actually, speaking of which... I should be able to do this woven thread at extremes with lazy union.
<Ravo>
14gb of RAM? Can the polygons not be pushed to the GPU?
<InPhase>
kintel: These are all non-intersecting polyhedrons. It should just... work.
<InPhase>
kintel: I loaded it just because I remember it being super slow at scale in the past.
<kintel>
Lazy union is cheating as it's not doing anything ;) Now it's just a matter of memory and GL rendering
<buZz>
ah llama is the facebook AI
<Ravo>
Large Language Model Meta AI, but the source was leaked and now just about anyone can run a trimmed down version locally.
<InPhase>
kintel: A bit. :)
<buZz>
Ravo: by 'leaked' you ment 'published on their public git' ?
<kintel>
buZz: I've implemented GLES2 shaders now in my offscreen demo, if you're still up for testing
<buZz>
ha, sure :)
<buZz>
*grabs phone*
<buZz>
4x arrow up , enter
<Ravo>
Well, it's been fun but it's getting late. I've learned a little about bolts and successfully written 7 lines of scad code in the course of 2 hours. I'm calling that good enough.
Ravo has quit [Quit: Client closed]
<buZz>
:)
<teepee>
oof, menger sponge is < 2 seconds on a raspi4
<buZz>
something wrong? :P
<buZz>
thats with that new CGAL-replacement?
<buZz>
kintel: it doesnt crash, last msg is 'Using GLSL 100 es'
<kintel>
buZz Should output out.png
<buZz>
and out.png has pretty colors
<buZz>
:D
<buZz>
two squares with hexagon around it in rainbow
<buZz>
oh, -one-
<buZz>
somehow feh went to tiling view :D
<kintel>
cool, thanks for testing! This means I've got GLES2 under control
<buZz>
you want/need the out.png? or a md5 of it?
<kintel>
nah, I trust the description. The background is random anyway, to detect re-renders
<buZz>
md5sum ends on '81660'
<buZz>
ah
<buZz>
well, amazing!
<buZz>
gles2 openscad is such a long dream of mine
<buZz>
even if offscreen rendering is needed for it, i'm excited
<kintel>
Next step (which will take a while) is to request a GLES2 context in OpenSCAD and see what breaks
<kintel>
Doing offscreen first as I don't need to fight with Qt ;)
<buZz>
:)
<kintel>
(and I can do all dev in an ssh session)
<buZz>
well, i'm happy to test any gles2 stuff you produce
<kintel>
Speaking of rendering, does anyone know if there are GPUs/drivers in the wild which _only_ supports GL_EXT_framebuffer_object (as opposed to the newer ARB version which is also part of OpenGL ES2 and OpenGL 3+) ?
<kintel>
"old" being the operative word. Trying to remove legacy stuff which makes the codebase too expensive to maintain
<buZz>
ah hmhm
<buZz>
so you want a -modern- driver that doesnt do ARB stuff
<kintel>
The ARB extension is written against OpenGL 1.1, so the EXT extension is ancient
<kintel>
ARB stuff is fine. OpenGL 2 + ARB extensions is still ok. It's the GL_EXT_framebuffer_object which is the issue as it's not compatible with GL_ARB_framebuffer_object
<buZz>
kintel: maybe time to put some telemetry in openscad?
<buZz>
to see what renderpaths people actually use
<buZz>
if there's a lot of legacy involved
<kintel>
It wouldn't hurt, I just haven't wanted to push for that yet considering our user base being above average privacy concerned + we need some backend sink.
<buZz>
yeah , but just make it opt-in
<buZz>
instead of opt-out
<buZz>
backend sink can just be a hardcoded twitter account :D lol
<buZz>
jk
GNUmoon has quit [Remote host closed the connection]
GNUmoon has joined #openscad
guso78 has quit [Ping timeout: 260 seconds]
J2331 has quit [Quit: Client closed]
TheAssassin has quit [Remote host closed the connection]