Stenberg Weighs Two Rust Rewrites of curl Against 14 Preventable Bugs in 81
Software / analysis
Stenberg Weighs Two Rust Rewrites of curl Against 14 Preventable Bugs in 81
Two repositories claim to be curl in Rust, one with 1,300 of curl's tests passing, and curl's lead developer says the language was never the obstacle.

Two projects now claim to be curl rewritten in Rust, and curl's lead developer says the language was never the obstacle. Daniel Stenberg, curl's founder and lead developer, wrote on Nov. 17 that agents converted about 185,000 lines of C into about 340,000 lines of Rust, by his own count, while the C project merged more than 300 commits in the same roughly three weeks.
His post, Actual curl rewrites in rust, names two repositories and gives a vulnerability count that makes the Rust case in numbers. The repositories' own READMEs add test claims Stenberg does not repeat.

The two rewrites side by side
The first, urlx by GitHub user jonwiggins, appeared a few months before Stenberg's post. Its README calls it "curl, rewritten in Rust" and ships a command-line tool, a Rust library and a C ABI layer meant to replace libcurl. The second, kov/curl, began on Oct. 20, 2026 and lives on a rust-rewrite branch of a curl fork.
| urlx | kov/curl | |
|---|---|---|
| Licence | MIT | curl's licence |
| Test claim | 1,300 of curl's tests pass, 92 skipped, 25 excluded | 1992 of 1996 runtests, 949 of 949 pytest on Linux |
| Size | 601 commits | about 2,300 commits, about 430,000 lines of Rust |
| GitHub stars | 25 | 5 |
Both are tiny by adoption. The star counts are as the pages showed them on Nov. 18, and stars are not usage.
The README for kov/curl says the code was written by Claude under the direction and review of the repository maintainer, whom it does not name. The urlx README does not say who or what wrote it, though the repository contains a CLAUDE.md file. Stenberg wrote that he has not been in contact with either set of authors.
The line counts disagree between sources. Stenberg counted about 340,000 lines of Rust from about 185,000 of C. The kov README says about 430,000 lines of Rust against about 200,000 of C. The post and the README may be measuring different things, and neither says how.
What the test numbers do and do not show
kov/curl's README says that "On every build configuration we measured, the rewrite passes every test" that the C build passes, apart from tests 96 and 558, which check C source file names. On Windows with Schannel it reports 1933 of 1948, where C passes 1934. It leaves out NTLM and 13 C unit tests that probe C internals.
Those are the projects' own measurements, run against curl's own test suite. Stenberg's point is that the suite is part of why conversion worked: the agents were trained on curl's well-documented code, and its extensive tests make conversion feasible. A passing suite shows the behaviour matches the tests. It does not show the project survives contact with users.
urlx lists HTTP/3 as "Full (HTTP/3 untested)." The kov README gives its lowest transfer-rate ratio against C as 0.81, for 50 parallel HTTP/3 downloads on aarch64, and a Linux x86_64 range of 0.86 to 1.11.
Stenberg's memory-safety arithmetic
The strongest number in the post concerns the Rust argument. Of 81 vulnerabilities confirmed in curl in 2026, Stenberg counted 14, or 17.3 percent, as mistakes a memory-safe language would likely have prevented. A Rust curl would still have had 67.
- Likely prevented by memory safety14 vulnerabilities
- Would remain in a Rust version67 vulnerabilities
Source: Daniel Stenberg, Actual curl rewrites in rust, Nov. 17, 2026
A commenter named Félix questioned the 67. Stenberg replied that nobody can know the exact count for another language, and that he stands by the verifiable count of past C-related vulnerabilities. That is a fair limit: the 14 is a classification of known bugs, and the 67 is the remainder, not a prediction of what a Rust codebase would have produced.
What a rewrite does not copy
Stenberg's case is that writing the code once is not the hard part. curl's value, he wrote, comes from years of iteration, user feedback, a working community and decades of ABI and API stability. Recreating the current state is doable, he said, but sustaining it is the hard part.
The numbers from his own project illustrate that. curl averages about 11 commits a day. In the three weeks he describes it merged over 300 commits, more than 200 of them bugfixes, with another month bringing about 200 more bugfixes. kov's DIVERGENCE.md already lists deliberate behavioural differences from the C build.
Stenberg does not believe in rewrites into any language. He favours what he called "constant and never-ending polishing, iterating and refactoring." He wondered whether the two projects are more than proofs of concept, and why their authors spent money and effort on them.
On cost, he wrote that the second project claims 8,600 US dollars, citing a Claude artifact. A commenter, Joonas Kuorilehto, wrote that the figure is an estimated API-equivalent cost and the work apparently ran on a subscription. The kov README as fetched carries no cost figure, so the 8,600 rests on Stenberg's account of a source we could not read.
The review burden of machine-written code is the subject of the Harvard coding-agent paper we covered, and curl's maintainers face a version of it on the receiving end. A second open-source maintenance story on the same theme is the ONNX Runtime repository that returned a 404. The next signal to watch is whether either rewrite tags a release that tracks a curl version number.
Sources
- 03kov/curl
More in Software
- 01OpenAI's WebRTC Voice Stack Draws a 30 ms Objection That Its Own Engineers Never AddressedLuke Curley says WebRTC drops audio older than 30 ms. OpenAI's engineers wrote about port ranges and routing, which is a different problem.
- 02Superpowers Passes 322,000 GitHub Stars as a Set of 15 Prompted Skills for 16 Coding AgentsJesse Vincent's MIT-licensed framework tells agents to brainstorm, plan and test first. Its clone count is large, and its evidence for better results is one author's view.
- 03GitHub Disables 144 Picofly Repositories After a Nintendo Notice Names OneThe November 6 notice is the first on GitHub aimed at Switch modchip firmware, and it asked for forks to be taken down without listing a single one.
- 04Chrome's Gemini Nano Download: What Google's Docs Say, and the Numbers They Leave OutGoogle's page on built-in model management gives triggers, a 30-day rule and full-download updates, but no file size, no disk threshold and no consumer off switch.