2022-10-03 11:41:00 <joe9        > does anyone know Daniel of etherforth?
2022-10-03 13:09:31 <KipIngram   > I'm playing with a trial license for a program called PixInsight this afternoon.
2022-10-03 13:09:40 <KipIngram   > It's used for stacking astrophotography frames.
2022-10-03 13:09:50 <thrig       > a stack, you say
2022-10-03 13:09:56 <KipIngram   > Looks like quite a learning curve, but it also looks very powerful.
2022-10-03 13:10:03 <KipIngram   > :-)  Yes - different usage of the word.
2022-10-03 13:10:19 <KipIngram   > Stacking is combining a large number of similar frames.
2022-10-03 13:10:28 <KipIngram   > The "signal" adds up, but the "noise" averages down.
2022-10-03 13:10:35 <KipIngram   > Raises the signal to noise ratio.
2022-10-03 13:10:39 <thrig       > sounds like a stellar plan
2022-10-03 13:11:01 <KipIngram   > There are several steps involved, each designed to compensate for some particular imperfection in camera sensors.
2022-10-03 13:11:42 <KipIngram   > All the individual pixel sensors are a little different, and you make special frames called darks, flats, and biases that you use to equalize all that.
2022-10-03 13:12:05 <KipIngram   > Then the "lights" are your actual sky shots.
2022-10-03 13:12:12 <KipIngram   > Each type gets stacked.
2022-10-03 13:12:40 <KipIngram   > Supposedly they have a batch script that will just "auto-do" all this, but I'm working through it by hand just to get familiar with the flow.
2022-10-03 13:13:23 <KipIngram   > I'm using files I downloaded from a YouTube guy's website, just so I know I've got usable data to begin with.
2022-10-03 13:13:39 <KipIngram   > Orion nebula.
2022-10-03 13:14:34 <KipIngram   > If you really push this method, you can take astrophotos without fancy sky tracking gear - you just stack up hundreds or thousands of short exposures.  Each one short enough that Earth's rotation doesn't make the stars trail.
2022-10-03 13:15:04 <KipIngram   > I've got a little gadget for my tripod, though, that will rotate the camera, and it has a 15 degree per hour setting.
2022-10-03 13:15:13 <KipIngram   > That's "almost' right - that would give a full rotation in 24 hours.
2022-10-03 13:15:38 <KipIngram   > But what you really want is a full rotation every 23 hours 56 minutes, which is how long it takes Earth to rotate with respect to the *stars*.
2022-10-03 13:15:49 <KipIngram   > The extra four minutes in our day comes from our orbit around the sun.
2022-10-03 13:16:18 <KipIngram   > But that's 99.7% right, so I think I can use it to increase my exposure times by a fairly good factor.  I'll have to test to see.
2022-10-03 13:16:47 <KipIngram   > I'll be pretty happy if I can get to 30-second or 1 minute exposures.
2022-10-03 13:17:26 <KipIngram   > in theory if I aimed the thing perfectly at the north pole (which I won't, of course), it would buy me a factor of 360.
2022-10-03 13:17:55 <KipIngram   > So that would be order of six minute exposures.  I'm sure inaccurate aiming will steal some of that away, but I may make it to minute-ish.
2022-10-03 13:18:48 <KipIngram   > That "close enough" rotation will also make all of the frames really similar - I won't have to re-aim the camera every couple dozen short frames.
2022-10-03 13:19:23 <KipIngram   > I'm going to mount a little "red dot" polar finder on the side of the rotation gadget to use to aim it.
2022-10-03 13:20:13 <KipIngram   > And I bought a little $10 gadget called an "intervalometer" for my camera - you can program it with an interval, an exposure time, and a picture count and it will just run the camera through that sequence.
2022-10-03 13:20:50 <KipIngram   > So hopefully I'll be able to just aim the thing, frame the target, and hit "go" on the intervalometer and have it grap a couple hundred shots.
2022-10-03 14:33:11 <boru        > KipIngram; yeah, PI is my goto for processing data.
2022-10-03 14:33:47 <boru        > It takes a little bit of time to grok the basics, but you'll find yourself going back to the same tools time and again for your pipeline.
2022-10-03 17:10:00 <KipIngram   > Yeah, I'm pretty impressed.  A little pricey, but given all it seems capable of doing it feels worth it.
2022-10-03 17:21:50 <KipIngram   > I'm now trying the automated script - that's really quite low labor.  Just load the files and hit "go."
2022-10-03 18:21:08 <veltas      > remexre: My idea for typesetting was to generate either GNU troff's intermediate language or PDF
2022-10-03 18:21:16 <veltas      > Do it with as much FORTH as possible
2022-10-03 18:21:40 <veltas      > I can give more instruction if you want or explain myself more if there are specific questions
2022-10-03 18:23:52 <veltas      > Looking into esperanto atm what am I doing with my life
2022-10-03 18:24:37 <thrig       > why is the ATM set to use esperanto?
2022-10-03 18:24:40 <veltas      > It's the metric system of languages but I have this sick curiosity and the promise of it making it easier to understand Spanish
2022-10-03 18:25:00 <veltas      > thrig: I would have said ATM machine if I meant ATM, don't worry ;)
2022-10-03 18:25:34 <veltas      > It's the little touches I know you all have come to love from me
2022-10-03 18:31:00 <remexre     > veltas: ah, okay; I was gonna roll my own forthy-troff -> DVI-ish -> PDF pipeline, since I don't particularly want to be tied to a c-compatible foundation
2022-10-03 18:31:58 <thrig       > isn't postscript a stack language
2022-10-03 18:32:29 <veltas      > Go straight to PDF in Forth and don't use "C-compatible" stuff?
2022-10-03 18:32:36 <remexre     > yeah
2022-10-03 18:33:10 <veltas      > I seem to remember there is a PDF generation library written in Python that's relatively light, you should look that up because it's probably easier to read than the PDF spec
2022-10-03 18:33:16 <remexre     > v1 would probably just be to something like TeX's DVI, because I'd like to do my thesis in this :)
2022-10-03 18:33:16 <veltas      > But get the PDF spec as well of course
2022-10-03 18:33:29 <remexre     > oh, yeah, I spent a while looking for a free copy of the pdf 2.0 spec, to little avail
2022-10-03 18:33:33 <veltas      > Okay your project, your decision
2022-10-03 18:33:34 <remexre     > adobe republishes 1.7 tho
2022-10-03 18:34:12 <remexre     > hah, yeah, to my surprise, my advisor is totally down with the idea of me implementing my own typesetter, and his only suspicion is "isn't forth memory-unsafe"
2022-10-03 18:34:15 <veltas      > PDF is essentially the modern day "printer language" "document transmitting language" and a few other things that are less relevant
2022-10-03 18:34:36 <veltas      > He probably wants you to use Rust lol
2022-10-03 18:35:04 <veltas      > I actually really like Rust by the looks of it, I'll never be interested in writing a nice type-safe language because it already pretty much does what I would do
2022-10-03 18:35:25 <veltas      > Not that I've used it, but based on looking at it seems like they made a lot of good choices
2022-10-03 18:35:47 <veltas      > I know that might be very surprising to people who know how cynical I am, but it is what it is....
2022-10-03 18:36:02 <remexre     > I'm actually teaching him Rust this term, heh
2022-10-03 18:36:23 <veltas      > Maybe you can pass me your notes some time lol
2022-10-03 18:36:42 <thrig       > C -> 4 second compile time; Rust -> 4 minute compile time and worse security
2022-10-03 18:37:00 <veltas      > Or replace Rust with Forth there
2022-10-03 18:37:08 <remexre     > I've been professionally writing it for a few years; we're mostly going thru the book ( https://doc.rust-lang.org/book/ ) with some extra discussion of PL-researcher-relevant topics
2022-10-03 18:37:15 <remexre     > e.g. strict provenance
2022-10-03 18:37:28 <thrig       > I suppose I could try to write a gemini server in forth
2022-10-03 18:37:37 <veltas      > remexre: Okay never mind then
2022-10-03 18:38:03 <veltas      > What are you doing professionally with Rust?
2022-10-03 18:38:43 <veltas      > I personally think it's ready to replace C++ wherever it's used. And hopefully it getting into Linux kernel will catapult it into embedded which needs a shakeup
2022-10-03 18:38:45 <remexre     > "anything that needs to be in an unsafe language that doesn't have to be in C++ to link to important libraries"
2022-10-03 18:39:18 <veltas      > And Linus' extreme hate should hopefully galvanise the community a bit against some of the creeping utopianism that C++ contracted
2022-10-03 18:39:37 <remexre     > mostly stuff involving binary-level instrumentation and analysis, but that's b/c of the sort of work I do in general
2022-10-03 18:40:31 <veltas      > "binary-level" what sort of binary?
2022-10-03 18:40:42 <remexre     > usually amd64, tragically
2022-10-03 18:40:50 <veltas      > I am confused, because from my perspective everything is binary lol
2022-10-03 18:40:53 <remexre     > we might be doing something with ARM Morello sometime next year, which should be fun
2022-10-03 18:41:03 <remexre     > ah, binary analysis as opposed to analysis of source code
2022-10-03 18:41:21 <veltas      > Okay what sort of instrumentation and analysis? Like performance stuff for running programs?
2022-10-03 18:41:56 <remexre     > mostly security-relevant stuff; "what control-/data-flow is happening in response to what"
2022-10-03 18:43:26 <veltas      > Is this for profit?
2022-10-03 18:43:48 <remexre     > yyyyyyyyeea this is a question that gets a more specific answer lol
2022-10-03 18:44:32 <remexre     > I work for a private research lab, so it's usually publically funded projects, but my employer is a private company with private ownership (we're an ESOP, actually)
2022-10-03 18:45:16 <veltas      > I did some aviation work and one of the things needed for CERT-ified software is some way to demonstrate that the binary you get actually matches the source you reviewed
2022-10-03 18:45:24 <MrMobius    > i installed rust long enough to compile one project I was interested in and it was 600mb. i think it would be 200k or less in C. is that normal in Rust and is there a different way to compile to avoid that?
2022-10-03 18:45:54 <remexre     > yeah, research lab == our code can be utter crap, the deliverable in the contract is a paper describing our discoveries :)
2022-10-03 18:46:15 <veltas      > Guess what aviation code is utter crap
2022-10-03 18:46:16 <remexre     > but better approaches for certifying programs is exactly the sorta work I do
2022-10-03 18:46:47 <veltas      > Just giving you an idea, not sure if it's worth anything to you, you probably have your own ideas already....
2022-10-03 18:47:18 <veltas      > Probably a hard market to break into
2022-10-03 18:47:24 <remexre     > MrMobius: depends heavily on what it's linking to; it's way easier to depend on stuff in Rust than in C, so most projects have way more dependencies (and statically link to them); the standard library is also getting statically linked, so that's probably a decent chunk of it
2022-10-03 18:48:01 <MrMobius    > so enormous file sizes but nothing to worry about once you have the program?
2022-10-03 18:48:09 <thrig       > supply chain attacks would be one reason why a 4 minute rust compile is less secure than a 4 second C compile
2022-10-03 18:48:54 <remexre     > yeah, ^^ is a real thing; "don't depend on random crap without looking over it" remains necessary for secure software development no matter what environment you're in
2022-10-03 18:49:01 <veltas      > MrMobius: Given that they're integrating it into the Linux kernel, any embedded concerns (as this is) are probably solvable
2022-10-03 18:49:30 <thrig       > (the other reason being the C code has a nice pledge and unveil policy, and the only pledge in the rust code is for a code of conduct)
2022-10-03 18:49:32 <remexre     > yeah, I would also suggest double-checking that you compiled it with optimizations; cargo build --release for that, vs cargo build with no --release for the default build (faster, debuginfo)
2022-10-03 18:49:49 <veltas      > I doubt debuginfo weighs 600MB
2022-10-03 18:49:55 <veltas      > somehow
2022-10-03 18:50:12 <remexre     > yeah, no opts + debuginfo + static linking + lots of dependencies feel like they _could_, in sum, plausibly be the cause
2022-10-03 18:50:33 <remexre     > but without seeing it, can't say for sure
2022-10-03 18:50:37 <veltas      > I feel like the last in that list is the single factor of consideration
2022-10-03 18:50:55 <veltas      > Well actually static linking as well
2022-10-03 18:51:04 <veltas      > Both of those combined
2022-10-03 18:51:22 <remexre     > oh, having monomorphized generics (i.e. like C++ templates instead of like Java/Haskell/etc generics) also doesn't help
2022-10-03 18:51:38 <veltas      > Fanfuckingtastic my dudes
2022-10-03 18:51:49 <remexre     > though that you can do either way (which you can in C++ too, but it's a bigger refactor than in Rust; not my favorite thing to do either way)
2022-10-03 18:52:03 <veltas      > Is it 'supported' though?
2022-10-03 18:52:15 <veltas      > And by that I mean can it exist in industry
2022-10-03 18:52:19 <remexre     > doing non-monomorphized generics? Yeah, that's trait objects
2022-10-03 18:52:24 <veltas      > Okay nice
2022-10-03 18:52:52 <remexre     > they're a core feature of the language, but it's not like a --dont-monomorphize flag, it's a refactor to the code that may be annoying, depending on the code in question
2022-10-03 18:53:21 <remexre     > I suppose the C analog would be writing your data structures against void* vs using the BSD list macros
2022-10-03 18:53:33 <remexre     > the latter leading to better perf (usually) and a bigger binary
2022-10-03 18:53:34 <veltas      > At work once we had GCC building a small C project running to like 40GB of RAM usage
2022-10-03 18:53:49 <remexre     > yikes
2022-10-03 18:54:07 <veltas      > Because we had a (say it quietly) exponentially expanding macro that we started using with like 7 arguments instead of 5
2022-10-03 18:54:17 <remexre     > OOF
2022-10-03 18:54:29 <veltas      > Or 4 levels of nesting instead of 3, something that silly
2022-10-03 18:54:45 <thrig       > last I tried to build gforth gcc started taking more and more memory until I killed it
2022-10-03 18:55:01 <veltas      > I saw this and said to the project lead "okay this is really funny but no let's actually fix this now" and he agreed
2022-10-03 18:55:23 <veltas      > He had been putting it off before then, it was also making the build really slow
2022-10-03 18:57:48 <veltas      > I think it was actually my fault this even happened
2022-10-03 18:58:12 <remexre     > the 7 arguments or the exponential time?
2022-10-03 18:58:18 <veltas      > The exponential space usage
2022-10-03 18:58:35 <veltas      > It was also time usage but space seemed to diminish first
2022-10-03 18:58:44 <remexre     > ah, yeah
2022-10-03 18:59:24 <remexre     > seems like that's the tradeoff for most of my work stuff too -- we actually have less parallelism permitted for link jobs than for compile jobs because they can have a peak memory usage of nearly 5G for some of our programs...
2022-10-03 18:59:41 <remexre     > (I think this is largely LLVM's fault though, it's huge)
2022-10-03 19:00:43 <remexre     > oop, i've gotta go cook dinner; be back in 30min or so!
2022-10-03 19:02:47 <veltas      > I think I was doing something like #define ARR(...) .arr = (T[]){__VA_ARGS__}, .size = DIM((T[]){__VA_ARGS__})
2022-10-03 19:02:56 <veltas      > With nesting of ARR
2022-10-03 19:03:28 <veltas      > Looking at it, I probably should have used null termination instead, could have put that in the macro too
2022-10-03 19:03:43 <veltas      > The whole point of the macro was that it was too easy to forget to put the null terminator in manually
2022-10-03 19:03:54 <veltas      > Funny how things are obvious in hindsight
2022-10-03 19:04:10 <veltas      > Insight to save 40GB of RAM
2022-10-03 19:09:46 <veltas      > remexre: Does it do whole program optimisation or something?
2022-10-03 19:10:24 <MrMobius    > I shot myself in the foot too with assembly macros. I had a system set up where you could call functions with macros and it would handle passing arguments on 6502 using a python script to do flow analysis and allocate zero page efficiently. turns out it expands the entire macro every time you use it so a program of 20kb or so generated a 28MB listing every time :/
2022-10-03 19:12:56 <thrig       > for some reason memoization got invented
2022-10-03 19:14:56 <veltas      > Macros are never ideal
2022-10-03 19:37:13 <remexre     > veltas: we do LTO within our stuff but this is just for statically linking LLVM (without LTOing with it)
2022-10-03 19:53:09 <veltas      > Oh right linking LLVM