Remote

dgkf

@dgkf@fosstodon.org

mastodon 4.7.3

data scientist and software engineer, passionate about open software for collective social good

building tools for productivity, expression & creativity - currently focused on #helix #HelixEditor #blender #blender3d

#r #rstats #python #julia #julialang for analysis

#rust for performance

i love #bikeTO, public transit, #vegetarian #vegan food, #green #socialism policy

he/him (they/them also fine)

256 Followers

317 Following

22 Posts

Joined June 02, 2023

Codeberg:

https://codeberg.org/dgkf

Blender:

https://projects.blender.org/dgkf

GitHub:

https://github.com/dgkf

dgkf @dgkf@fosstodon.org

· 6mo ago

Noodling on a theory I had at the start of the AI wave:

Aside from translation & image gen, AI software solutions are almost universally solving problems where a structured data model would solve the problem better.

Of course, it takes significant thought to model these problems as data. This isn't a claim that all AI problems should be solved by data modelling, but when we are building AI systems we should be cogniscent of which deterministic problems we're approximating in stochastic space.

dgkf @dgkf@fosstodon.org

· 10mo ago

Replying to

@LeafyEricScott@hachyderm.io we have those too :) https://codeberg.org/r-codeberg/woodpecker-ci I have yet to find a CI system that I don't have some minor grievances with, but my primary concern right now is enabling migrations. I'm happy to support both for the time being if it means more converts! It seems like the momentum right now is toward Forgejo Actions.

codeberg.org

dgkf @dgkf@fosstodon.org

· 13mo ago

Replying to

@geospacedman@mastodon.social @defuneste@fosstodon.org yeah.. momentum is a hard thing to change. I was also GitLab-curious when it was the best alternative, but was always a bit unsure of what to make of their enterprise/open-source split. With Forgejo I think the story is a lot less ambiguous. Still, it’s hard to get people to move away from where all the action is. I think real change is only likely if we can see some major R packages move. The r-lib org seems like a great candidate 🤔

dgkf @dgkf@fosstodon.org

· 12mo ago

Replying to

Thanks to everyone who is boosting this! Just an update, this project is moving forward as an ISC proposal!! For anyone who wants to see a pre-read, you can see our draft https://github.com/dgkf/2025-10-isc-proposal-debug-adapter Unfortunately, that means this project's ISC proposal is already spoken for. With some luck, we'll see it maturing soon!

github.com

dgkf @dgkf@fosstodon.org

· 10mo ago

Replying to

@_wurli@fosstodon.org I saw that! Good on them. I've been similarly burned by GitHub Actions. I think they run actions on forks until inactive for 3mo. I was frustrated by this egregious waste when I forked a repo that had long-running nightly scheduled integration tests and realized all 3.2K of its forks were probably doing the same. I can't endorse a platform that has such disregard for the environmental impact of computation. It's unethical.

dgkf @dgkf@fosstodon.org

· 10mo ago

Replying to

Yeah, I mean GH is still unavoidable for how ubiquitous it is. I still have a lot of work on there and need an account to collaborate. But here's to something better!

dgkf @dgkf@fosstodon.org

· 7mo ago

Replying to

@andrewnez@mastodon.social YAGNI working overtime on these committee names. All are 🔥

dgkf @dgkf@fosstodon.org

· 7mo ago

Replying to

Let's start with the positives. It's pretty repairable. Replacement electronics are easily sourced and the 3D printing STL/STEP files are public. It uses the programmable and open source QMK firmware and has a plethora of buttons. Assembly was relatively straightforward. One throughhole component to solder and otherwise just cleaning up some 3D printing mess and screwing a few things together. Probably about an hour of assembling a kit with good instructions.

dgkf @dgkf@fosstodon.org

· 12mo ago

Replying to

@sebffischer@mastodon.social yeah, I think the immediate benefit would be more for developers. I'd see it as something that would run at build time and throw an error if the type checking failed. It would be able to catch things like changes to the function signature of dependency functions that fail to type check while avoiding runtime overhead. Long term this could be used to optimize byte code.

dgkf @dgkf@fosstodon.org

· 13mo ago

Replying to

@defuneste@fosstodon.org @geospacedman@mastodon.social when there’s tension in the air (and here I mean the tension between financial and social interest), I think the best thing to do is to be loud about it and help bring other people to the realization that the current course isn’t in their interest. If the company you’re alluding to were a B Corporation, I’d hope they’d have the social interest in mind for core infrastructure decisions like this.

dgkf @dgkf@fosstodon.org

· 13mo ago

Replying to

And a couple other areas that I think need some love in the R ecosystem, for those looking for grant inspiration:

Package “features” https://github.com/r-lib/pak/issues/810
S7 traits and cleaner mutable S7 objects via extension package https://github.com/RConsortium/S7/issues/34
Ages ago I wrote a runtime type checker, {typewriter}. These days I think there’s more value in writing a static type checker that checks a package at build time. https://github.com/dgkf/typewriter

GitHub

Blueskies Feature Request: Package features syntax · Issue #810 · r-lib/pak

Just to set the tone, this really isn't a feature request but more of a proposal for something that I think could really benefit the R ecosystem. Since pak/pkgdepends has its own internals for reso...

dgkf @dgkf@fosstodon.org

· 11mo ago

Replying to

@mpadge @rOpenSci Beautiful! Yes, I'll absolutely give it a read. Thanks for keeping our project in mind! There's still plenty that can be improved in our guide, so we would be very welcoming to an influx of interested parties!

dgkf @dgkf@fosstodon.org

· 6mo ago

Replying to

@dcz@fosstodon.org I don't think we're disagreeing here. I tried to carve out other space in the little clause at the end for the speed to delivery. I also think that giving advice broadly falls under the category of translation. "translate my loose idea into the language of a structured argument". Its a bit more of an abstract definition of translation, but I think it's a better model for conceptualizing genAI than "reasoning".

dgkf @dgkf@fosstodon.org

· 20mo ago

Replying to

@michaelslade@mastodon.cloud @timnitGebru@dair-community.social [citation needed]

dgkf @dgkf@fosstodon.org

· 12mo ago

Replying to

@jrosell@mastodon.social interesting! How did I miss this base R function! It would be incredible if we could use that to build a type system - especially if it can be used by the byte code interpreter. Checking to see if a function uses NSE would allow some cool code path optimization.

dgkf @dgkf@fosstodon.org

· 12mo ago

Replying to

@jrosell@mastodon.social super cool - I didn't pick up on that syntax being a base feature. At least in quickr, it seems like it's really only used via NSE anyways, not actually using the base::declare() behavior. The docs suggest that this could be used by the byte code interpreter. It's not obvious to me how declare() expressions would be picked up and used. It seems to not actually do anything right now? https://github.com/wch/r-source/blob/15dfebbed34e85047e9e0c3932113882e8edad5c/src/main/eval.c#L9401-L9404

github.com

dgkf @dgkf@fosstodon.org

· 9mo ago

Replying to

@pat_dev@social.linux.pizza By "rolling out" I was referring to their hosted option. Up until recently they've supported forgejo actions only if you self hosted, but that's a pretty high bar for most folks I'd imagine. Now they have runners available to everyone. I think they refer to their hosted option as "beta" still. They're nearly compatible, but do require changes - most notably using fully qualified domains for action sources. I filed an issue with some enhancement requests https://code.forgejo.org/forgejo/runner/issues/1169

Forgejo: Beyond coding. We forge.

bug: using actions that themselves use actions with unqualified domain names cause jobs to fail

### Can you reproduce the bug on the Forgejo test instance? No ### Description I'm testing Forgejo Actions with a few common GitHub actions, namely the actions in [github.com/r-lib/actions](https://github.com/r-lib/actions), which are widely adopted actions for the R language. I'm using Codebe...