Why OCaml Maintainers Rejected a 13,000‑Line AI‑Generated Pull Request

Background

OCaml provides two main compilers: ocamlc for bytecode and ocamlopt for native executables. Developers typically use ocamlc during development and testing, while ocamlopt is used for production builds. Native debugging support is limited compared with bytecode tools.

AI‑generated DWARF support

Developer Joel Reymont needed DWARF debugging information for the native compiler. He used Anthropic’s Claude Code to generate the required changes, supervising the AI over several days without writing code himself. The resulting modifications were submitted as a pull request to the OCaml repository:

https://github.com/ocaml/ocaml/pull/14369

Authorship and copyright concerns

The AI‑generated files listed Mark Shinwell of Jane Street Europe as an author. Shinwell is a contributor to OxCaml, an open‑source project that adds DWARF support to OCaml. OCaml contributor Tim McGilchrist noted that many files appeared to copy OxCaml’s work. When asked, Reymont said the AI chose the attribution and he did not question it. Shinwell later asked the AI about copyright; the AI responded that no code was copied from OxCaml.

Maintainer response

OCaml maintainer Gabriel Scherer rejected the pull request, citing several issues:

Assigning copyright to real people by an automated tool creates legal ambiguities.

Lack of prior design discussion for the feature.

Difficulty reviewing a >13,000‑line change with limited reviewer capacity.

Potential long‑term maintenance burden.

Existing parallel DWARF work in OxCaml that is not yet ready for upstream integration.

Scherer reminded that the project has long suffered from a reviewer bottleneck, and large, low‑effort PRs risk overwhelming the review system. He closed the PR, stating that no maintainer was willing to take on the required work and that the project's governance model does not accommodate such contributions.

Implications

The episode shows that Claude Code can, under human supervision, produce complex functionality, but the reliability, quality, and legal status of AI‑generated code remain uncertain. It also highlights broader concerns about the growing volume of AI‑assisted contributions to open‑source projects, including nondeterministic outputs, hallucinations, and prompt‑injection risks.