tropism v2026.11.1: export that re-exports after an upgrade, and fewer false missing-dep reports in Rust
Contents
tropism is a dependency analyzer for polyglot repositories. It finds import cycles, manifest problems and duplicated packages, and it enforces the architecture rules your team wrote down in a tropism.toml, across Go, JavaScript/TypeScript, Rust, C#, Python, Ruby, Java, Swift and C++. It never runs a package manager and never executes the code it analyzes, so it works on a fresh checkout with no toolchain and no network. If you want a rule like “api must not depend on data” to fail a commit, in whatever mix of languages your repository uses, that’s who it’s for. I wrote about the idea in Architecture the Agentic Loop Can Actually Check.
v2026.11.1 was tagged on 31 October. It has 16 commits since v2026.10.1: three code changes, 12 dependency updates and the Homebrew formula bump left over from the previous release. The diff is 11 files, 800 insertions and 113 deletions. Two things in it are worth knowing about before you upgrade, and one of them only matters if you run tropism export.
Fewer wrong missing-dep reports in Rust
The missing-dep check reports a package that’s imported but not declared in the manifest. In Rust it could fire when it shouldn’t, in two cases. The fix is one commit that adds 115 lines to crates/tropism-lang/src/rust.rs.
The first is a use that names an item declared in the same scope. pub use helper as alias; next to fn helper, or use test_helpers::make; next to mod test_helpers;, both refer to local items, and Rust would reject a crate of the same name as ambiguous. tropism reported both as missing dependencies: one was a re-export alias in insight, another project of mine, and the other an integration test’s own helper module. Now an item declared in the same scope (a function, module, struct, enum, trait, const, static, type, union or macro) wins. Only the scope the use sits in counts, so an item inside a nested module still doesn’t hide a crate.
The second is a crate whose library name isn’t its package name. Cargo turns - into _ to get the library name, unless the package’s own [lib] name says otherwise, and that setting lives in the published package’s manifest, which tropism never downloads. So use md5::Md5 from the md-5 package looked like an undeclared md5. The fix is a short table of widely used crates where the two differ: md-5, sha-1, rust-crypto, rust-ini and xml-rs. It’s consulted only when the package is declared, so it can attribute an import but never invent a missing dependency. A crate outside that table can still trip the check.
This matters more than a single false positive suggests. The README rates missing-dep as “Good” and caps it at Medium confidence, and the pre-commit hook blocks on it. A check that blocks a commit has to be right when it fires.
tropism export: conformance records for software-analytics
tropism export writes JSON Lines records for software-analytics, which is a separate project of mine for collecting delivery data. The README’s invocation is:
Two commits change what it writes and when.
The first adds conformance records. Per the design notes, you get one conformance_snapshot per repository and commit, covering findings by check, rules configured and checks unavailable. You also get one conformance_finding per finding, keyed <repo>:<finding id>, so the same finding can be followed across commits. That’s the part I wanted: whether a rule violation was introduced, fixed or just carried along, over time rather than once.
The second keys the export state by tropism version as well as by commit. Before this, a repository whose commit hadn’t changed was skipped. But an upgrade to tropism can change the findings for the same commit, so skipping it would leave old results in the record after the tool changed its mind. Now the version is stored alongside the commit in <state>/export.json, and a new tropism version re-exports once. The visible effect is that the first export after you upgrade is bigger than the ones that follow it.
If you don’t use export, neither commit changes anything you run.
Dependency updates
The rest is maintenance, with no behaviour change that I know of:
- Rust toolchain 1.99.0
- tree-sitter 0.27.1 and tree-sitter-swift 0.7.4
- toml 1.1.7 and then 1.1.8, and serde_json 1.0.154
- insta 1.49.0, used in tests
- the Python Docker image tag, to 3.15
- two lockfile maintenance commits plus a
cargo update, and one grouped minor-and-patch bump
Of these, the tree-sitter bumps are the ones that could in principle move findings, since the grammars are how tropism reads Swift and the other source languages. The release notes don’t record any changed output from them, but I didn’t run a before-and-after comparison, so treat that as untested rather than as a guarantee.
Install or upgrade
On macOS and Linux with Homebrew, the README’s first install is:
After that first tap, brew upgrade picks up new releases like any other formula. The README explains why the first two lines are needed: the formula lives in the tropism repository rather than a separate tap, and Homebrew 6 refuses to tap an untrusted third-party tap, so the trust must be granted first and must use --formula. Older Homebrew has no brew trust and needs only the last two lines.
If you have bx installed, a tool of mine that downloads a GitHub project’s release archive for your machine and runs it, you can skip installation entirely:
Or take the archive for your platform from the releases page and put tropism on your PATH. Every release carries a SHA256SUMS. To build from source:
These are the instructions from the README at the v2026.11.1 tag. Newer releases have been tagged since, including v2026.11.2 with the tropism-mcp server and a no-admin Windows install, so brew upgrade today gets you a later version than this one.
Should you upgrade?
If you run tropism export, or you use it on Rust code and have seen a missing-dep you didn’t believe, yes. For everyone else it’s a safe monthly bump: nothing in the release notes removes a flag or changes the exit codes of 0, 1 and 2.