IRC channel logs
2026-06-08.log
back to list of logs
<aggi>real fascinating, nonetheless 8/16bit might remain relevant for hardware bootstrap itself <matrix_bridge><Jeremiah Orians> Well, yes. Diversity in options is what would ultimately make it impossible for the whole to be compromised. (As there only needs to be a single honest bootstrap to ddc against) <aggi>the configurator/steps build system crossed my mind again, which got another advantage <aggi>if a complete system-integration for a bootable development host got migrated to it <aggi>this keeps the whole base-package set (in my case the c-toolchain system profile) distribution-agnostic <aggi>it's not necessary then to choose between any of the complexity with guix/gentoo/etc.. to main those critical base system components <aggi>too this configurator/steps system is fully bootstrappable, and enables reproducible-builds <aggi>there is no need almost to transition to any other build/packaging system <aggi>i will however think about a simple dependency tracking mechanism <aggi>adding a depends metadata file to individual steps/ - noting compile-time depends inside there - could do the trick <matrix_bridge><Jeremiah Orians> An sqlite database file is probably the one which could enable multiple applications to spawn from a shared knowledge set <aggi>then hacking a little script which parses all steps looking for the depends metadata, and generate a manifest from this to build steps/ in order these are needed <aggi>because i've got ~500 ebuilds here, willing to move those into configurator/steps <aggi>well, yes, sqlite is an option too (i've already ported it for support with linux2/tinycc/static-linking) <matrix_bridge><Jeremiah Orians> And tcc should be able to build it without too much trouble <aggi>since you mention it, sqlite had been a little problematic, i couldn't use it's latest version <aggi>and it's things like this, why the re-integration for a full release is important <aggi>too, there's other relevant packages which got blocked, bind9 for example <aggi>meanwhile all distributions ship latest versions of say sqlite, bind9 already - which aren't compliant with the last remaining c-toolchain integration here <aggi>things like __thread are a burden, nptl is working with linux2.4 here, and i kept track of all builds which actually need it (it's optional hence) <aggi>but the __thread language extension inside C-standard remains problematic <aggi>and _all_ major distributions are trapped inside it already <aggi>which isn't a problem once you've got and don't bother with depending on c++/gcc/llvm/clang <aggi>but if free software shall remain portable towards and with different pure c-toolchains, then this battle is lost almost <matrix_bridge><Jeremiah Orians> The battle remains won, as long as a single chain remains <matrix_bridge><Jeremiah Orians> More links take time and more people working on it <aggi>thing is, i had needed a confirmation from some university in germany for example, or any HR office <aggi>that's why i applied as laboratory engineer at some, with an offer to prepare for some lectures <aggi>but it's all been denied, ignored <aggi>i've too succeeded to salvage some older TeTeX implementation, remaining portable with tinycc <aggi>since i tried to keep in mind a few pieces which i actually needed when studying <aggi>then again, why would anyone bother with a 20year old TeX implementation? <aggi>try explaining this to some HR office <matrix_bridge><Jeremiah Orians> No justification needed. Who does decides. <matrix_bridge><Jeremiah Orians> HR doesn’t give a fuck as long as the work is done and no shit is being given to stir trouble in an area <roconnor>24KiB of Flash Storage. I wonder how much bootstrapping that gets you. <roconnor>As I work through my Compiler Quest I keep wondering how tightly I should be constraining my memory use. Though lazy functional programming is ill suited for low memory devices to begin with. <siraben>my project over the last few days, mostly to learn how to try to scale up RE and doc writing with coding agents <siraben>so much capability in 1 MB of flash, and most of it is related to scientific/mathematical processing! <siraben>roconnor: I looked into that a bit and you can do things like reduce the GC size/trigger GC more often or do something more generational, but it starts eating into wall clock time <siraben>the biggest payoff would come from changing the representation of strings out of cons cells <siraben>but, that would also come at the cost of introducing a new primitive in the VM. Given how much storage reduction would be possible maybe it's not so bad <roconnor>Yeah, my parser if offline so getContents like immidately inflates the input by like a factor of 8 at minimum :] <roconnor>Anyhow I figure I just get things working first and worry about these sorts of optimizations later. <siraben>Also, those types of optimizations can be explored much more cheaply with LLMs now <siraben>getting from A to B still requires a lot of human effort! <roconnor>Also I have no GC at the moment so there is that. :P