Patrick Dubroy
@dubroy.com
Programmer & researcher, co-creator of https://ohmjs.org. 🇨🇦 🇩🇪 🇪🇺
Co-author of https://wasmgroundup.com — learn Wasm by building a simple compiler in JavaScript.
Prev: CDG/HARC, Google, BumpTop
Co-author of https://wasmgroundup.com — learn Wasm by building a simple compiler in JavaScript.
Prev: CDG/HARC, Google, BumpTop
A few quotes about Steve Jobs from "Valley of Genius" by Adam Fisher.
January 11, 2026 at 5:26 PM
It's actually official supported (from the pre-X times), you can find instructions here: www.tweetarchivist.com/how-to-downl...
I just took that and wrote some scripts to do some additional processing on it (resolving short links, etc.)
I just took that and wrote some scripts to do some additional processing on it (resolving short links, etc.)
www.tweetarchivist.com
January 10, 2026 at 7:15 PM
It's actually official supported (from the pre-X times), you can find instructions here: www.tweetarchivist.com/how-to-downl...
I just took that and wrote some scripts to do some additional processing on it (resolving short links, etc.)
I just took that and wrote some scripts to do some additional processing on it (resolving short links, etc.)
I believe it can! But it was a one-time thing, and I tried gallery-dl first, and it worked.
January 10, 2026 at 6:16 PM
I believe it can! But it was a one-time thing, and I tried gallery-dl first, and it worked.
Oh dang, that looks nice! I think I saw it a while back but forgot about it.
If I had remembered, I would have probably just used that :-)
If I had remembered, I would have probably just used that :-)
January 9, 2026 at 5:09 PM
Oh dang, that looks nice! I think I saw it a while back but forgot about it.
If I had remembered, I would have probably just used that :-)
If I had remembered, I would have probably just used that :-)
My web site is built with a custom-written static site generator (basically just a ~700 line Python script + Jinja templates).
I extended it to be able to able to parse the download and render the downloaded archive (with lots of help from Claude Code).
I extended it to be able to able to parse the download and render the downloaded archive (with lots of help from Claude Code).
January 9, 2026 at 2:52 PM
My web site is built with a custom-written static site generator (basically just a ~700 line Python script + Jinja templates).
I extended it to be able to able to parse the download and render the downloaded archive (with lots of help from Claude Code).
I extended it to be able to able to parse the download and render the downloaded archive (with lots of help from Claude Code).
Do you ever (or would you consider) recording/streaming your investigation into such an issue? Would be really interesting to see imo.
January 7, 2026 at 5:45 PM
Do you ever (or would you consider) recording/streaming your investigation into such an issue? Would be really interesting to see imo.
(This is basically the same as the distinction between synthesized vs inherited attributes in attribute grammars.)
January 5, 2026 at 12:08 PM
(This is basically the same as the distinction between synthesized vs inherited attributes in attribute grammars.)
One diff I've noticed is that in compilers, most computation is bottom up — generally, a change in a leaf means that all the ancestors are dirty.
Whereas in UI, there's a lot more top-down information flow (props).
Whereas in UI, there's a lot more top-down information flow (props).
January 5, 2026 at 12:06 PM
One diff I've noticed is that in compilers, most computation is bottom up — generally, a change in a leaf means that all the ancestors are dirty.
Whereas in UI, there's a lot more top-down information flow (props).
Whereas in UI, there's a lot more top-down information flow (props).
Nice article, I was looking at this recently as well: bsky.app/profile/dubr.... I've also been working on fully incremental pipelines on top of the incremental parsing support in @ohmjs.org.
PL folks —
For @ohmjs.org, I’m looking at how to support fully incremental pipelines (eg parsing → codegen, all incremental). I know about Salsa (github.com/salsa-rs/salsa); what else should I be looking at?
For @ohmjs.org, I’m looking at how to support fully incremental pipelines (eg parsing → codegen, all incremental). I know about Salsa (github.com/salsa-rs/salsa); what else should I be looking at?
GitHub - salsa-rs/salsa: A generic framework for on-demand, incrementalized computation. Inspired by adapton, glimmer, and rustc's query system.
A generic framework for on-demand, incrementalized computation. Inspired by adapton, glimmer, and rustc's query system. - salsa-rs/salsa
github.com
January 5, 2026 at 12:00 PM
Nice article, I was looking at this recently as well: bsky.app/profile/dubr.... I've also been working on fully incremental pipelines on top of the incremental parsing support in @ohmjs.org.
"a nice chewy problem"
January 5, 2026 at 10:11 AM
"a nice chewy problem"
Yeah, perhaps. Still think it would be really interesting to dig into the specifics of all the different cases! Will let you know if I decide to investigate and write that post :-)
December 21, 2025 at 7:10 PM
Yeah, perhaps. Still think it would be really interesting to dig into the specifics of all the different cases! Will let you know if I decide to investigate and write that post :-)
Isn't the specific issue there a megamorphic call site?
Whereas with tagged unions, you more often have a switch on `obj.kind` and then monomorphism within the individual cases, no?
And I wonder if the `obj.kind` access would go through a shared transitional map, and thus be fast? But not sure.
Whereas with tagged unions, you more often have a switch on `obj.kind` and then monomorphism within the individual cases, no?
And I wonder if the `obj.kind` access would go through a shared transitional map, and thus be fast? But not sure.
December 21, 2025 at 6:14 PM
Isn't the specific issue there a megamorphic call site?
Whereas with tagged unions, you more often have a switch on `obj.kind` and then monomorphism within the individual cases, no?
And I wonder if the `obj.kind` access would go through a shared transitional map, and thus be fast? But not sure.
Whereas with tagged unions, you more often have a switch on `obj.kind` and then monomorphism within the individual cases, no?
And I wonder if the `obj.kind` access would go through a shared transitional map, and thus be fast? But not sure.
Thanks. Yeah I’m familiar with the concepts (hidden classes, poly/megamorphism, etc.) but would love a concrete demonstration of how this plays out with typical usage of tagged unions.
December 21, 2025 at 5:32 PM
Thanks. Yeah I’m familiar with the concepts (hidden classes, poly/megamorphism, etc.) but would love a concrete demonstration of how this plays out with typical usage of tagged unions.
Do you know of a good post that describes the performance implications of tagged unions? I’d be interested to read that.
December 21, 2025 at 3:27 PM
Do you know of a good post that describes the performance implications of tagged unions? I’d be interested to read that.