The hamilton repository, the DAG framework for defining dataflows as Python functions, saw 24 commits in the last week. Every single one was a Dependabot dependency bump. No feature work, no refactors, no behavior changes. For anyone tracking the library itself, there is nothing to act on. For anyone running the Hamilton UI or VS Code extension, the bump list is worth a skim because several of these packages sit on network facing paths.
One week, four lockfiles ¶
The diffstat reads like a supply chain audit: 26 files changed, 1023 insertions, 878 deletions, essentially all of it lock churn. The commits split across four dependency surfaces:
/ui/backendonuv.lock, 7 commits touched it, includingdjango,aiohttp,urllib3,certifi, andopentelemetry-api/ui/frontendonpackage-lock.json, 6 commits, includingvite,undici,moment, anddompurify/contrib/docsonpackage-lock.json, 8 commits, led by shell-quote and joi and brace-expansion/dev_tools/vscode_extensionon its own lock, 3 commits, plus afast-uribump that jumped two minor versions at once in 50d5cd8
The top changed file list makes the story obvious. contrib/docs/package-lock.json took 8 commits, ui/backend/uv.lock took 7, and ui/frontend/package-lock.json took 6. Lockfile maintenance is now a standing cost of shipping a UI alongside a Python package.
The ddtrace jump is the big one ¶
The largest single bump is ddtrace in the UI backend, 4.8.2 to 4.15.6, in e8e266c. That is seven minor versions of a tracing library that instruments bytecode. The diff excerpt shows why it was not a one line change: ddtrace now pulls a ddtrace-internal package, and the lockfile gained per Python version resolution markers splitting bytecode 0.17.0 for anything under 3.15 from bytecode 0.19.1 for 3.15 and up.
The diff also shows the lockfile was restructured around Python 3.14 versus 3.15 markers, and a wheel list appeared for ddtrace covering arm64 macOS and manylinux targets. Anyone who has run uv against a lockfile with stale resolution markers knows that failure mode. This commit keeps the backend installable across the 3.10 through 3.15 range the lockfile declares.
If you run the Hamilton UI backend with Datadog tracing, expect your instrumentation to shift under you in a mostly invisible way. Tracing libraries patch at runtime, so a seven minor version jump is the kind of change worth a smoke test in staging before it reaches your image.
Network facing packages in the batch ¶
Several bumps touch code with request handling or content responsibilities. On the frontend, dompurify moved from 3.4.13 to 3.4.17 in 6df9fe7, which matters because the UI renders user visible DAG content. ip-address jumped 10.4.0 to 10.7.2 in 4352469, skipping a major.
On the backend, aiohttp went to 3.14.4 in 2926ff9 and urllib3 to 2.8.0 in a5a7212. Both libraries have had CVEs in the last two years that required prompt bumps. The Django patch release, 5.2.16 to 5.2.17 in ee83777, is the kind of point update you want merged without ceremony.
None of these commit messages say “security fix”, and Dependabot bumps usually land before the advisory text is public. The pattern is still recognizable: keep them current and the risky weeks disappear.
The VS Code extension stays exposed ¶
The dev_tools/vscode_extension surface got three bumps, but the notable one was brace-expansion, a package with a history of regex complexity issues, bumped again in 72ed393. Interestingly, the same package was bumped separately in /contrib/docs in f1a89f9 as 1.1.18 to 1.1.21. Two directories, two lockfiles, two independent rounds of the same fix.
This is the cost of a polyglot monorepo. Each subproject gets its own advisory surface, and Dependabot has to play whack a mole per manifest. It also means reading the changelog by directory is the only way to know your blast radius if you consume only one of these subprojects.
What to watch ¶
- The ui/backend lockfile now carries Python 3.15 resolution markers on
bytecodeandddtrace. If you pin the backend to an older Python, verify your resolvedddtraceversion against the marker table rather than assuming one version. vitemoving 8.2.0 to 8.3.4 in the frontend is a minor version jump for a build tool that occasionally breaks plugins. Rebuild the UI once before deploying.- Zero human authored commits in a 7 day window usually means a release or feature burst either side of it. Watch
mainfor the core library to catch up, since the framework code itself was untouched this week.