Overdue Last read · last published · next check
Last read 11 hours ago, longer than this feed's 1 hour schedule.
HotFlow is looking for its direction in terms of interface etc. because I haven't see anything doing exactly what it does but I'm getting useful and valuable profiling work done with it :-)
my quest for finding less cluttered (chrome, firefox) browser; and one which does not have any weird antigay or whatever political bias :-) i don't think vivaldi was bad but definitely was not for me. helium and zen are pretty strong contenders in the minimalistic space.
My kernel goals for upcoming weeks/months is to help to push trenchboot through. Ross has suffered even more than I did with SGX :-) And the feature does make sense and is totally legit.
@ zygoon Yeah, all this said I don't "hate" Canonical, it's all just what I see and how I understand it. I hope they do well. Just wanted to clear this up :-)
@ zygoon Library crates are like that yes but executables can be whatever. And if the concern is that there might come a day that some of the code might be outsourced to a library crate, even CLA would be a more sane idea IMHO :-) At least the end product have likely have shipped with less bugs.
@ zygoon I'm even cool not liking GPL. Otherwise I would not have any Apple products :-) I get Apple's motivations on what they are doing.
@ zygoon Sorry, I modified the phrasing in order to make it explicit that not throwing any hard truths :-) The most recent example that raises my concerns is uutils: 1. Not everything has to be GPL 2. There's no things wrong using Rust or doing a Rust rewrite. 3. Cheapest overall R&D cost on doing that would be: 3a. Use compatible license, which is GPLv3 and look up from the sources. 3b. Use GPLv2…
Canonical feels an adversary of GPLv3 and/or somwhat shady actor when it comes to free software.
E2E software was actually slop even before ChatGPT was born. The stuff that companies like Tieto or Accenture push: 1. used to suck a lot 2. does suck as of today 3. will suck forever Slop is a cosmological constant :-) If I look at any real product company i.e., sell something customers are expected love, not that much has actually changed, except a new automation tool.
have been cleaning up C code for caffeine commander and remembered one of its features: it does not use floating point numbers :-) i wonder how many tui programs make FPU a separate concern but i just cannot use it. it's weird and does not exist in kernel 🤷 i have a fixed point number library instead.
continental plates are moving: took the head of the Trenchboot patch set to my tree early morning. They sit on my for-next-tpm branch and once it is synced up with linux-next I'll send my v7.3 PR to Linus with the Trenchboots commits included.
I'm just looking at it and thinking how hot Dell can be :D
Impulse buy: Dell UltraSharp U2410 from 2010. You would not believe but this is one of the most desired displays for any sort of retro computing. Its speciality is composite inputs and direct support for Amiga's 15 kHz video signal.
HotFlow - a probablistic CFG profiler that I've written to help optimizing CPU inference for my tools Goosedump and ReadSeek. Inference is quite nasty and tricky to profile by measurement given maximization in scaling. E.g., when you optimize dedicated hardware for inference it's a minmax optimization to thenear proximity as possible of OOM as one can reach. Since I'm not really seasoned on doing…
and it is fully agnostic. i can feed it with full vmlinux.
that is actually the magic here: you can profile even stuff that you could not run. symbolic execution is cool i think :-)
i wrote a static profiler for myself with angr and capstone python modules. it calculates for each basic blocks average number of branching instructions executed and ranks basic blocks based on how frequently they are visited. why run when you can just count? 🤷 after playing with CPU inference quite many weeks this sort of first hand info i need in order to know what look for further... i used…
When doing https://gitlab.com/jarkkojs/tpm2-library I realized something about git subtree. Mono-repo merge is one single use case where you *have to use* git subtree command. Otherwise authority is lost and it makes the whole repo legal wise somewhat shady.
Another nice "unslop" surprise. I decided to merge my TPM2 crates to a mono repo: https://gitlab.com/jarkkojs/tpm2-library I thought that they must have at least 100 KSLOC or something (had never checked). They are actually 25 KSLOC. Not that bad for seven craates and full TPM2 tack.
I've now been finalizing Puu OS to be published at Gitlab and undoing various (especially Rust) workarounds. It was ~13500 KSLOC of build files. Now it is only ~8500 KSLOC :-) There was a lot of cruft gathered while iterating this build for 6 months. Now it starts to look sleek and clean :-) #buildroot
The reason why there is no ~/.config/gcc/config.toml and why I made landstrip a *parametrized* sandbox becomes obvious when I you hit your head againt the wall few times with Rust and Cargo :-) ~/.cargo/config.toml is IMHO an anti-pattern. #rustlang #buildroot
I moved ReadSeek and Goosedump to Gitlab only to get less AI attention because it was basically cost in time, not very useful stuff overall :-) Landstrip is still in Github and it got e.g., 3000-4000 SLOC PR recently and submitter had no clue of its internals.
@ jan voting period is still ongoing. It was up to end of this month.
Debian's AI policy is understandable, as AI neither maintains the code nor pays the cloud bills :-)
