Canadian entrepreneur and developer. Building Grain DDL.
Joined March 2008
- Tweets6.6K
- Following2.6K
- Followers1K
- Likes4.1K
Pinned Tweet
Announcing Grain DDL: code generation by and for demanding engine developers.
Define your structs and data once. Write custom generators for protocols, save files, language interop, shaders, databases, more.
No macros, no metaprogramming. Generate reviewable code on every keystroke. 300kB compiler exe. Windows, Mac, Linux.
Closed beta applications are now open - link in replies.
Michael Labbé retweeted
I don't follow this explanation. Could someone at one of these labs complete the thought publicly, so I can understand exactly what is being proposed?
What difference would it make if your company "gets there first"? What specifically are you going to instruct your first-mover human-extinction-capable ASI to do that will prevent competitors from also getting to that same point shortly afterward, and therefore still having the human-extinction-capable ASI you claim you must prevent them from getting?
In other words, if the argument for Anthropic building ASI is that it's too dangerous to allow OpenAI to build ASI, what specific steps would Anthropic take using their ASI to prevent OpenAI from achieving ASI sometime thereafter?
A common response is “if they truly believe this, why are they still building it?” At OpenAI, many have not deeply internalized the civilizational stakes. At Anthropic, the stakes are well-understood, but they are locked in a race to get there first - they believe no one else will act responsibly, so they must do it themselves, despite the risk.
Some problems are more quickly identified with visualization than a step debugger. Streaming debug mesh data on every frame so you can pan around it and frame step in both directions is useful. Video is from when I was working on meshing Doom sectors, a fun side project.
Southpaw numkeys makes way more sense, even if you're right handed. I use this for data entry, Blender and tracking.
Cameron's text editor bundles TCC to compile plugins. I'm a fan of audacious tech direction.
Expo presentation number five! @starfreakclone is the creator of an editor called fred, which has a very powerful customization system and preserves the document’s entire edit history, even when you undo.
youtube.com/watch?v=xz5aPCRx…
I summarized my rationale for writing a DDL compiler that does not call free() during compilation.
reddit.com/r/C_Programming/c…
Thank you to all of the people who have signed up for the Grain DDL closed beta. I read all of the emails, but haven't had time to respond to all of them.
I am planning on doing a series of releases, adding testers as it becomes more capable.
If Microsoft lays off id Software's engine-focused engineers, it is an admission that they believe peak engineering competence is not a part of the way out of their problems.
The history of id Software is that their engine team is not constrained by cost center thinking.
id software updates
-Programmers, tool makers and other id tech engine focused staff have been almost all laid off
-Speculation that id tech might be retired over time
-Quake Champions team has been gutted, future uncertain
-Senior staff in many areas such as art and more were let go
-Individuals involved with QuakeCon let go despite the event being next month
Sources: @ScottApogee @dark1x
@S1Braw @business
I started a .plan file as a half-joke and a half-statement: Don't depend on closed platforms like Discord for information archival. Mirror important info as txt.
Today's talks about generating relocation tables to avoid complex pointer fixup code.
finger [email protected]
26 years ago I wrote a .plan file aggregator for the web. This was before RSS or blogs were mainstream.
The general wisdom is that blogs+rss were modernized .plan files. I disagree. Blogging became org controlled, per-project. Plans were per-developer. Wild west.
A sign of a great conference is that you leave it tired but inspired. The Handmade Expo was full of people giving great demos. I gave one early on in the day, demonstraining some of the things a data definition language focused on systems code generation can do.
The second presentation from the Handmade Network Expo is now live! @frogtoss has been building a real-time metaprogramming system with a custom IDE called Grain DDL. Zero dependencies, runs everywhere, small enough to check into source control.
youtu.be/qfEzUnZlMIo
Software needs to be understood in order to be trusted. I have decided to explain how I wrote Grain DDL in the official Discord. Join if learning about compiler design in C is interesting.
Or, finger [email protected]. Yeah, I'm keeping a .plan that mirrors Discord posts.
I thought this was a viral one off. Nope. I’ve been listening to @IommiWatts all day while I work. New fan here.
Michael Labbé retweeted
The first video from the Handmade Network Expo is finally going live in just 30 minutes!
@nicbarkeragain and @antovsky have joined forces in a game studio called Ostrich Effect, and today they're ready to show their first game: an RTS with a programming twist.
Link in thread ⤵
We are seeing unfortunate price hikes due to ai component shortages.
Grain DDL (in active development) can generate text using megabytes of RAM instead of hundreds of gigabytes. It doesn't use inference. It uses code generation templates -- you create templates that parametrically define what is generated. Expert driven, deterministic, and responsive over time to the changes in your data schema. No human language re-prompting messiness.
It generates results in under a vblank on my $200 refurb thinkpad with DDR4 ram.
There are ways to ripple schema changes to your systems code that do not involve an expensive, controversial stochastic process. It won't replace AI because it needs an expert in the loop. Instead, it is the reliable way for an expert to keep their data and code in sync.
The video of me presenting this from the Handmade Network Expo will be posted soon.
