<J225293829>
finaly my gears are rendering and not getting errors .. after 3h i got a 17mb 3mf file for rack and pinion ..
<teepee>
now the interesting question is, does it work with the fast-csg branch too? and how long does that take?
<OlivierChafik[m]>
teepee: oh blimey, sorry about that, will try and sort this now
<teepee>
no need
<teepee>
I just removed the first part of the condition
<OlivierChafik[m]>
hah ok :-D
<OlivierChafik[m]>
I've reduced my CGAL test case to 7 lines but the mxe docker env is driving me nuts
<teepee>
what happens?
<OlivierChafik[m]>
everything I think I have clean instructions to install an arbitrary version of CGAL to show it compiles w/ and w/o it, something goes wrong
<OlivierChafik[m]>
I'll try again w/ your -openscad docker image instead of the drier -ui one I used
<OlivierChafik[m]>
I could just tell them "go test this on win64 with your own scripts" but I can see how high that would go on their list :-D
<teepee>
I just tagged the container to openscad/mxe-x86_64-openscad:cgal-5.4 so it's not overwritten by newer builds
<OlivierChafik[m]>
I generally noticed a few cases of the file system playing tricks on me, like not picking up my changes, etc
<OlivierChafik[m]>
(inside docker I mean)
<OlivierChafik[m]>
I probably just need to sleep haha
<OlivierChafik[m]>
(I feel like I've always and will always be a docker beginner)
<teepee>
it's probably just missing some tools, it's really trimmed down, not even less or vi are installed by default
<OlivierChafik[m]>
cool thanks, that will definitely help w/ the repro
<OlivierChafik[m]>
yes I've never installed vim so many times in a day
<OlivierChafik[m]>
I'll pick this up again tomorrow, maybe throw some docker composer to simplify things
<OlivierChafik[m]>
(also, the only failure in the PR is now the lgtm c++ 🎉; hope I didn't leave too many oddities in the code)
<teepee>
oh, it should use the mxe-cmake I suppose
<teepee>
but yeah, there's more days ahead :)
<OlivierChafik[m]>
that cmake has been quick to complain about things like its g++ not managing to compile simple stuff
<teepee>
I'll check if I can make it run with an extra container
<OlivierChafik[m]>
thanks!!
<OlivierChafik[m]>
(I think maybe I've chronically underestimated the value of `. scripts/setenv-mingw-xbuild.sh -64` as you suggested - saw that a bit too late)
<peepsalot>
finding the intersecting point(s) of two paths boils down to determining the intersection of two line segments, and just do that for all possible pairs of segments
<J22529382928>
this seems to be an awful amount of operations.. as i have seen doing that is often slower than the 2D operation of scad itself .. does it make sense to make the calculation myself instead of using intersection() as i am extruding the points later anyway
<peepsalot>
no idea what you want to do. probably not worth re-implementing intersection within scad... but if you figure it out, feel free to submit a PR https://github.com/thehans/FunctionalOpenSCAD
GNUmoon has quit [Quit: Leaving]
GNUmoon has joined #openscad
<peepsalot>
2D boolean is already *very* fast, and actually done via the library which I just linked to (clipper).
GNUmoon has quit [Remote host closed the connection]
GNUmoon has joined #openscad
<J22529382928>
i just getting "polygon not closed" errors - which shouldn't be possible as they are overlapping but something is causing errors when scaling these in extrusions maybe due to the undercut
<peepsalot>
what is overlapping?
<J22529382928>
the polygons to make the intersection
<J22529382928>
so when linear_extrude(5) this is fine but with scale=[1,.85] it results in "not closed"
<peepsalot>
can you take a pic of F6 render on the 2D profile before extrusion? you can use root modifier "!" to single out
TheAssassin has quit [Remote host closed the connection]
teepee has quit [Remote host closed the connection]
TheAssassin has joined #openscad
teepee has joined #openscad
Alexer has quit [Ping timeout: 240 seconds]
Alexer has joined #openscad
little_blossom has quit [Ping timeout: 240 seconds]
little_blossom has joined #openscad
<peepsalot>
J22529382928: what is the scale of that, and roughly how many points are in the outline?
<J22529382928>
each teeth should have ~ 32 but there is an rounding (via offset ±) which adds some points
<J22529382928>
its about 1cm long .. interestingly i made a small change just moved a little so it starts between teeth - and now the problem seems to be gone.
<J22529382928>
peepsalot it seems the beveled part adds a lot of slices - not sure why https://imgur.com/vW8Jeoo
<J22529382928>
when scale is a list you get awfull amount of slices - does that make sense? linear_extrude(10,scale=[1,0.9])circle(10);
<peepsalot>
yes that's intentional, when scale is not uniform it will create slices based on $fs and $fa (or $fn)
<peepsalot>
you can also manually override slices with parameter "slices"
<J22529382928>
seems to me it getting much more slices than needed
<J22529382928>
ah the height is not taken into accout for the calculation
<peepsalot>
are you using $fn?
<J22529382928>
yes
<J22529382928>
without it looks better
<peepsalot>
well, that's why its better to use $fs and $fa instead for most cases. because it scales reasonably
walterwhip has quit [Quit: My MacBook Air has gone to sleep. ZZZzzz…]
<J22529382928>
it also causes issues as it changes the position of vertices depending on size like you have a tube then the outer circle can have a different number of fragments then the inner - which causes problems in printing
<J22529382928>
well from 6h down to 6min render - the slices within the beveled edges - Ü, interesting that such a drastic change got through the tests when submitting the PR
<J22529382928>
probably no test is checking for that
<J22529382928>
vertices from 200k for 6teeth down to 8k for 15teeth
<J22529382928>
is it known that 2022.2 has massive Z-fighting when show edges is on? https://imgur.com/vRZsIE7
<Scopeuk>
On the mimalloc windows build stuff do we know how much of the performance build we can get by just swapping out the cgal allocator to mimalloc https://doc.cgal.org/latest/Manual/devman_memory_management.html? As thats done through a compile time overide rather than link time shenanigans for general overide it should also work with a static build
<Scopeuk>
Hmm going through the history that was the starting point
Guest42 has joined #openscad
Guest42 has quit [Client Quit]
<teepee>
J22529382928: it does not like CyclGetriebe :(
<J22529382928>
it?
<J22529382928>
fast csg
<teepee>
yes, it reverts to nef_union for at least one of the bigger steps
<teepee>
I think that means that it classified one of the temporary meshes as not closed
<J22529382928>
i wonder why .. i even tried an offset which should close any gaps without success
<teepee>
not sure, that could be the feature itself
<J22529382928>
but the linear_extrude with 3 slices helped (or $fn=0r
<J22529382928>
what about the Z-fighting should i open an issue for that (didn't found any existing)
<teepee>
I'm not sure that is actually z-fighting
<teepee>
well, at least not the classic one between coincident faces
<J22529382928>
yeah right there should only be one face .. first i thought it is my model but happens with any difference
<J22529382928>
of extruded circles
<teepee>
it's the edge fighting with the face
<J22529382928>
convexity also doesn't change a bit
<J22529382928>
also happens with cylinder()
<teepee>
yeah, just hide the edges :)
<J22529382928>
and in intersections
<teepee>
there's probably some way to fix that on the shader / OpenGL side
<J22529382928>
.oO yeah best way to circumvent these problems is switch of the internet and PC - and take a walk in the woods .. all bug reports there are from a different nature
<J22529382928>
.oO we fixed all bugs - also we removed all features and buttons and options and preferences and settings ..
<teepee>
yes, also waaaay more secure
<J22529382928>
i really had a client telling his network is secure when my colleague said " if your car is always in the garage you didn't have accidents on the street"
<J22529382928>
(every desk had two PC one internal and one with internet)
<Scopeuk>
J22529382928 thats standard in mid security enviroments
<Scopeuk>
Its known as air gapping
<J22529382928>
until there are ways to get data out via LEDs or RF etc.
<Scopeuk>
Hence mid security, high security will increase the gap and introduce additonal measure
<othx>
J22529382928 linked to YouTube video "What is Your Password?" => 1 IRC mentions
<teepee>
always good to keep in mind, even when thinking "can't happen to me"
<teepee>
unfortunately it can, it just takes a good timing and a bit of luck sometimes
<teepee>
so far I'm good, but I've seen that with other people who I know are cautious and it still happened
<teepee>
which is why I'm sometimes angry at actual official mails that are raising the same red flags
<J22529382928>
want access - take a blue overall and a ladder with you
<J22529382928>
was in a high sec datacenter the double door system was weighting you .. and the chief always need to carry a big PC to get in because she weighted not enough to trigger the system
GNUmoon has quit [Ping timeout: 276 seconds]
<teepee>
ouch, what a horrible system
<J22529382928>
she was super tiny
<J22529382928>
<45kg
<J22529382928>
<90 pound
snakedGT has quit [Ping timeout: 240 seconds]
<teepee>
not a good reason to be discriminated by a security system
<teepee>
what happens with a 2,30m big guy? just breaking through down to the next floor? :D
snakedGT has joined #openscad
<J22529382928>
you get much more trouble if you brought hardware with you and installed it (and use the wrong entrance)
<J22529382928>
you were trapped until security arrived
<Scopeuk>
People regularly mix up is a pain in the ass to get into with is secure
<Scopeuk>
The best security is almost transparent
<J22529382928>
not sure if "best" should be used in context with security
<J22529382928>
once i entered a building via metal detector and police searching your bag .. then in the first floor i found a kitchen with 25cm(10inch) knifes
<Scopeuk>
Sounds about right, if its anything like my work kitchen the back office them will be sharper than the edge. The point stands however
<Scopeuk>
Back of them**
<J22529382928>
honeypot knifes
<J22529382928>
like the plastic knifes on planes while there was a glas mirror on the toilett
<J22529382928>
there is some interesting classification for 3D printed weapons as they are not metal - and companies started to build working plastic combat knifes
<teepee>
oh, good, so just waiting for polititians to see that and produce crazy ideas again :(
<teepee>
meh, getting 2021.01 to build with the new MXE is quite annoying
<Scopeuk>
peepsalot from my understanding of the mimalloc docs there should be two dll's mimalloc and the redirect one, i suspect we may be at the bottom end calling into the static version of Microsoft's os calls instead of calling through the dynamic interface mxe 64 bit target vs mxe 64.static targets. I guess we'll see what they have to say for
<Scopeuk>
That should probably be os call bindings/interface
<Scopeuk>
peepsalot did you try running with just the allocator in cgal overidden to see how much difference that made? As its compile time rather than link time it should take on a windows static build
<Scopeuk>
Trying to wrap my head arround a plan b for getting speed on windows without someone spending years fighting the build chain
<teepee>
I think there's code for that in cgal.h ?
ur5us has quit [Remote host closed the connection]
ur5us has joined #openscad
<Scopeuk>
Yeh its guarded on pre proc if 0
<Scopeuk>
I should really get my build enviroment back up and stop be lazy over it
<peepsalot>
Scopeuk: the main performance improvement is from GMP memory functions, which i override anyways. i couldn't measure any difference with just the CGAL (c++ style allocator)
<peepsalot>
there's like 12 or so variations of build configurations you can test
<Scopeuk>
Ok, thanks
<Scopeuk>
I think i only currently have windows build on windows setup and working and thats somewhat edgecase, i have a linux box lying arround i can push into a build env i should really do that again
ur5us has quit [Remote host closed the connection]
ur5us has joined #openscad
<peepsalot>
Scopeuk: btw the redirect one is a pre-compiled binary that is just there in the github repo. it just gets copied out during cmake "install"
<peepsalot>
i guess on windows you can do all 16 combinations, this was my potential configuration grid on linux: https://bpa.st/BSSA
<Scopeuk>
Line16 tickles me for some reason, shall take a look
<peepsalot>
double it up to 32 if you want to compare static vs dynamic builds
<Scopeuk>
There is a tool i have used before on windows that walked the dynamic dependency tree for an executable
<Scopeuk>
I saw you raised a few issues on their tracker, pleanty for me to chew on
<peepsalot>
i don't know... feeling kinda burnt out on messing with it tbh. i'm guessing it was just never intended for a cross-compile scenario and that's where things are mostly going wrong
<peepsalot>
the static linking option built and ran on windows, but it suffered the same issue of no console output, which made it harder to confirm it was even working
<Scopeuk>
Yeh it looks like getting linux working was fairly cooperative but everything else does look to just be fighting
<peepsalot>
teepee: what was the original reasoning for tests running on msys2 build only, but artifacts upload for mxe only?
<peepsalot>
could we add executable artifact upload for msys2 builds so we could download and debug more easily?
<peepsalot>
er, i guess the mxe build being on linux would require wine-based testing... answering my own question mostly.
<peepsalot>
although wine tests might not actually be a bad idea after all
snakedGT has quit [Ping timeout: 256 seconds]
<teepee>
I guess MSYS2 happened mostly because it was available on the CI sites
<teepee>
there could be a github action just using the docker builds too I suppose
snakedGT has joined #openscad
<peepsalot>
teepee: i wonder if msys2 builds would actually be preferrable over MXE for packaging and releasing, maybe less compatibility issues with mimalloc etc due to being at least built on the same platform as target?
<peepsalot>
i didn't quite get dynamically linked mimalloc working on msys2, but it should only require some small tweaks to install/copy the dlls into the exe directory
<Scopeuk>
if we went over to build windows on windows could potentially switch fully over to the visual studio tool chain (via cmake of course)
<teepee>
peepsalot: considering we have not much control about what versions happen at any time, I'd be a bit concerned
<teepee>
with MXE we have at least some control over the build environment
<Scopeuk>
that did hit the vcpkg stuff, ignore my previous comment
<teepee>
yeah, that might be an option
<teepee>
I guess that would mean going all-in with github which is also a small concern, having no other option at all
<teepee>
in the last time a number of the CI platforms dropped out due to limitations
<InPhase>
rapha: Jack is going by J\d\d\d\d\d\d\d\d\d\d\d now. :)
<peepsalot>
rapha: you mean J22529382928 ?
<InPhase>
There were too many Jacks in the world, so they had to be given serial numbers.
<rapha>
InPhase: 11 digits, always?
<InPhase>
rapha: Well it has been gradually growing with time.
<J22529382928>
testing the limits
<J22529382928>
seems it stops there
<rapha>
InPhase: easier then to just go with J\w*\d+ then.
<peepsalot>
limits of what, your peers' patience?
<rapha>
J22529382928: just in case you're Jack, wanted to keep you updated on the efforts to restore the old Extra 300. Motor arrived today and the first version of the cowl that got printed fully is already not a bad fit at all! Since the walls are so thin, it also adapts itself to the fuselage shape almost perfectly! blob:https://imgur.com/e7a1b4d5-bad5-43a5-aa69-14cbd7dd2491
<J22529382928>
haha no this clients always adding number to my nick and i just wanna find out where this goes
<rapha>
needs to be made a little longer, that's it. also, need to decide for a color. thinking white or grey since it'll be tough to match the shade of orange.
<J22529382928>
look a good fit
<rapha>
it's amazing actually
<rapha>
and i never thought a PLA object with .35mm thick walls only would be so sturdy
<J22529382928>
aslong it is not a methanol engine
<rapha>
lolno
<rapha>
but a biest of a brushless
othx has quit [Ping timeout: 256 seconds]
othx has joined #openscad
<J22529382928>
a 3D super ellipse has good static - not good as a sphere but better than a rounded cube ( or sqround
<rapha>
interesting
<rapha>
but makes sense intuitively
<rapha>
are y'all here engineers / mathematicians / physicists btw?
<J22529382928>
using 3500kv with 3S lipo?
<rapha>
950Kv
<rapha>
10x10 prop
<J22529382928>
ah just remember this is much bigger then i thought (and it looks on the images)
<rapha>
but yeah, 3S ... but my smallest is 3.2Ah and it'd make the thing too heavy. will have to get a couple of 2.2Ah ones.
<rapha>
yeah, it's almost 1m wingspan.
<J22529382928>
you printing 0.35 with 0.4 nozzle? - and single walled?
<rapha>
well, pretty sure it's not actually .35
<J22529382928>
vase mode could make iteven more smooth
<rapha>
and it was 100% infill
<rapha>
oh, there's a "vase" mode? lemme try that then.
<linext>
measure the extrusion width
<J22529382928>
your slicer should have that.. and with .6 or .8 line width this should be sturdy enough
<linext>
layer height should be 50% to 80% of extrusion width
<rapha>
linext: don't care ... it came out really nicely either way :D
<linext>
that's the right way
<rapha>
J22529382928: Cura...
<J22529382928>
always wondered who came up with that .. i was printing .6 at 0.08layer
<J22529382928>
but the nozzle area must not be smaller than the layer×linewidth
<J22529382928>
(for bridging)
<linext>
i've done 25% before on a real thick nozzle
<linext>
it works but the resolution isn't that good
<J22529382928>
rapha in cura it is found under special modes
<linext>
vase mode is sweet
<linext>
i used to print in vase mode
<J22529382928>
i have a tool in my lib to make multiwall vase modes and also thick walls in vase mode
<J22529382928>
so i can make i tube with 2mm wall in vase mode
<J22529382928>
(in makes is a picture how it turned out)
TheAssass1n is now known as TheAssassin
<rapha>
holy shit that looks like a lot of math
<rapha>
also it looks like it might be REALLY useful as a guard for a fish tank filter / powerhead
<rapha>
as in, to keep fish and shrimp from killing themselves by swimming into the pump inlet
<J22529382928>
that probably works too .. i use it to drain my yoghurt - Ü
<J22529382928>
put it in .. and after empty the whey you have a nice cream
<J22529382928>
(putting the filter into yoghurt not other way around)
<J22529382928>
i think the amount of math didn't change for anything you do with a computer - only the perception of it
<rapha>
well, i tried reading the code in your library for the function you used to help me with that cowl. and that was a lot of math for me alright :)
<teepee>
if you have an scad, we can drop that in there, but just any image is fine
<teepee>
oh!
<teepee>
> Either git or ssh (required by git to clone through SSH) is not installed in the image. Falling back to CircleCI's native git client but the behavior may be different from official git. If this is an issue, please use an image that has official git and ssh installed.