Carmack on the Goose Language: Data Stacks, Relative Refs
John Carmack called the Goose language neat and said parsing file and network data is often 100x slower than it needs to be. Here is what he said and what to do with it.
2 min read · Last updated: October 25, 2026
John Carmack praised the Goose language on October 23, 2026. He credited its automatic use of multiple data stacks and resizable fields, plus relative references and built-in binary structure validation. He also claimed that "parsing" file and network data is often 100x slower than it needs to be.
Came here from Instagram, LinkedIn, X or YouTube? Here is the full source list you came for. Everything below is built from these, nothing else.
The post
The discussion
Key takeaways
- Carmack's post went up at 6:49 PM on October 23, 2026 and showed 16.8K views when captured.
- He says multiple data stacks and resizable fields move a language from "lots of work can be done this way" to "basically everything can be".
- He links relative references and built-in binary structure validation to direct memory mapping of data structures.
- His headline claim: parsing file and network data is often 100x slower than it needs to be.
- The Hacker News thread had 26 points and 5 comments, so the discussion is early and thin.
What did John Carmack say about the Goose language?
He wrote two short paragraphs on X. The first reacts to the language: "That looks really neat!" He says the automatic analysis and use of multiple data stacks and resizable fields takes it from "lots of work can be done this way" to "basically everything can be".
The second paragraph is the one worth keeping. He says relative references and built-in binary structure validation "enable direct memory mapping of data structures and immediate use of network data." Then the line people will quote: "Parsing" file and network data is often 100x slower than it needs to be.
Why does it matter to a developer?
Carmack names a specific cost most of us accept without measuring: turning bytes into data structures. His claim is that this step is often 100x slower than necessary. That is his statement, not a benchmark, and the post gives no test setup.
If he is even partly right, the fastest parse is the one you skip. Data that can be used straight from a file or a network buffer never needs a separate parsing pass.
| Feature he named | What he says it enables |
|---|---|
| Automatic analysis and use of multiple data stacks | Moves the language from "lots of work" to "basically everything" |
| Resizable fields | Same sentence: broad coverage of what can be built |
| Relative references | Direct memory mapping of data structures |
| Built-in binary structure validation | Immediate use of network data |
What changes in practice?
Nothing changes today, and I will not pretend otherwise. The source is one post. It does not describe Goose's syntax, release state, tooling or benchmarks, so I am not going to guess at them.
What does change is a question you can ask of your own code: how much time goes into parsing, and how much of it is avoidable? Carmack's post is a prompt to measure that, not proof of a number for your stack.
What should you do this week?
- Read the original post and the Hacker News thread for the Goose project link and any replies that add detail.
- Profile one real path in your own service where you read a file or network payload. Write down what share of the time is parsing.
- Note whether that data could be used in place instead of copied into new structures. That is the idea his post points at.
FAQ
What is the Goose language?
The post I had does not describe it. It only says the language looks neat and names its features: multiple data stacks, resizable fields, relative references and binary structure validation. Check the Hacker News thread for the project itself.
Did Carmack publish benchmarks for the 100x claim?
No. The post states it as a general observation, that parsing is often 100x slower than it needs to be. Treat it as an expert's estimate until you measure your own case.
What are relative references and why does he care?
He says they, together with built-in binary structure validation, enable direct memory mapping of data structures and immediate use of network data. That is all the post says about them.
When did he post this?
At 6:49 PM on October 23, 2026.
More on how languages get built: programming languages and the languages used to build them and code was never the hard part. Try the free tools on Art of Code or browse /jobs.
Written by Adithya Guttha, Founder of Art of Code.
Sign in to like and save posts.
Keep reading
Discussion
Comments (0)
Want to join the conversation?
Sign in to commentNo comments yet. Be the first to comment!