Repository navigation
fix: ten more small fixes from release testing - #5186
Conversation
With `rerun-open web` the bridge opened a browser tab and logged only the gRPC address. Rerun prints the page URL at a log level its SDK hides. The bare `http://localhost:9878` shows a viewer with no data, so a closed tab could not be reopened from the log. The bridge now logs the full URL, `?url=` query included, whenever it serves the web viewer.
The first run of an arm blueprint cloned its description repository (62 MB for the xArm) without a log line, so the terminal sat silent and the run looked hung on a slow network. `GitAssetCache` now logs the repository, the ref and the cache path before it clones, and the duration after. The M20's sparse fetch and the data repository clone of an installed package were silent in the same way and log the same two lines.
PointLio logs `odom_freq: 30.0` and publishes odometry at 10 Hz. The estimator puts out one pose per lidar scan, 10 Hz with the Mid-360 driver, and `odom_freq` only skips poses that come faster than it. A value above the scan rate changes nothing. A value of 5 gives 5 Hz. The config comment, the `--help` text of both fields and the `--odom-freq` help of `pcap_to_db.py` now say so. The help of `pcap_to_db.py` called it the odometry rate.
Arms whose URDF has no collision meshes logged "Cannot set `.show_collision`, since no collision meshes were loaded" at startup and on every display mode change. For such an arm the scene draws a second copy of the visual meshes as a stand-in. That copy loads no collision meshes, and the scene still set its `show_collision` flag. The scene no longer touches that flag. The single-arm OpenYAM model has collision meshes since #4980. `dual-openyam-planner-coordinator` still showed the warning.
`get_data` called `tar.extractall` without a filter. Python 3.12 warns about that on every extraction, and Python 3.14 changes the default. The call now passes `filter="data"`, so the behaviour is the same on every Python version and unsafe members (absolute paths, links out of the tree, device files) are refused. All 123 archives under `data/.lfs` pass the filter. The one visible change is that extracted files lose their group write bit.
…ssip off `go2-sim-nav` and the M20 sim turn zenoh gossip off on purpose. Zenoh then warns "Gossip is disabled: cannot de-duplicate data traversing region gateways" once per session. A clean `go2-sim-nav` start printed it 12 times: 4 from the native modules and 8 from the Python sessions. The log filter that Python sessions and native modules share now raises `zenoh::net::routing::hat::peer` to error. That warning is the only one the target has in zenoh 1.10.1. The experimental Rust recorder printed the warning too, so it takes the same filter.
`unitree-g1-groot-wbc` started with two Rerun warnings, "Dropping chunk without components" for `/world/odom/g1` and `/world/odom/g1/pelvis`. The static URDF logger put a `Transform3D()` with no fields on the robot root and on the base link. Rerun has nothing to store for those and drops them. The two entries are gone. The recording holds the same entities as before. Alfred uses the same logger and loses the warnings too.
The "Faster builds via the Cachix substituter" section read as if every nix-built native module came from the cache. CI publishes the modules in the `dimos` package, the ones `bin/native-modules --build` builds. The two example outputs under `examples/native-modules` are not in the cache, so a machine without them compiles the examples on their first run. The section now says so.
Newer nixpkgs warns "stdenv.isLinux is deprecated, use stdenv.hostPlatform.isLinux instead" at evaluation. The depth2depth_cloud flake printed it on every `bin/native-modules` run. The root flake uses the same attributes 36 times and would start to warn at its next nixpkgs bump. Both flakes now use the `hostPlatform` form. The root dev shell evaluates to the same derivation as before on x86_64-linux, aarch64-linux and aarch64-darwin. The depth2depth_cloud derivation differs only in its source, which holds the flake. The "stdenv.isDarwin" warning of the C++ builds stays. It comes from the external `lcm_extended` flake.
Under a host node that is not 24, `npm install` warned "EBADENGINE Unsupported engine" for dimcode, which wants Node 24. It read like a broken install. npm checks engines against the node that runs it, but the script installs a private Node 24 in that same run and that one runs the CLIs. npm cannot silence that warning alone, so the script passes `--loglevel=error` when it installs the private Node. Errors still show. With a host Node 24, as in CI, nothing changes.
[Medium impact] The PR reaches several startup, visualization, data-loading, and build workflows. Fix Python compatibility in archive extraction before merging. Findings
SummaryTen release-testing fixes improve startup messages, reduce expected warnings, and clarify configuration and build docs.
Reviews (1) · Last reviewed commit: "fix(build): install-test-clis prints no ..." · Reviewed by Greptile |
There was a problem hiding this comment.
go2 replay realtime (arm64)
Details
| Benchmark suite | Current: 37af1e6 | Previous: 1916f35 | Ratio |
|---|---|---|---|
peak memory |
1740.793 MB |
1740.68 MB |
1.00 |
peak threads |
374 threads |
377 threads |
0.99 |
network (transport) |
2728.328 MB |
2730.983 MB |
1.00 |
disk write |
1.832 MB |
1.586 MB |
1.16 |
instructions |
138.773 G |
138.613 G |
1.00 |
This comment was automatically generated by workflow using github-action-benchmark.
Codecov Report❌ Patch coverage is
@@ Coverage Diff @@
## main #5186 +/- ##
==========================================
- Coverage 83.73% 83.64% -0.09%
==========================================
Files 1864 1867 +3
Lines 181729 183001 +1272
Branches 14967 15165 +198
==========================================
+ Hits 152168 153075 +907
- Misses 25585 25959 +374
+ Partials 3976 3967 -9
Flags with carried forward coverage won't be shown. Click here to find out more.
... and 26 files with indirect coverage changes 🚀 New features to boost your workflow:
|
Why do we need this?
Ten small bugs found during release testing.
rerun-open webthe Rerun bridge opens a browser tab and never logs its URL. The barehttp://localhost:9878shows a viewer with no data, so a closed tab cannot be reopened.get_dataextracts tar archives without afilterargument. Python 3.12 prints a DeprecationWarning for that and Python 3.14 changes the default.odom_freq: 30.0and publishes odometry at 10 Hz. The setting is an upper limit. Point-LIO puts out one pose per lidar scan..show_collision, since no collision meshes were loaded" at startup.go2-sim-navprints the zenoh warning "Gossip is disabled" 12 times on a clean start. The blueprint turns gossip off on purpose.unitree-g1-groot-wbcstarts with two Rerun warnings, "Dropping chunk without components".bin/native-modulesrun.bin/install-test-clisprints npm "EBADENGINE Unsupported engine" warnings on a host whose Node is not 24. It reads like a broken install.What is this feature?
filter="data". All 123 archives indata/.lfspass it.--helptext now say thatodom_freqandpointcloud_freqare upper limits.dimospackage and that the examples compile on their first run.stdenv.hostPlatform. The C++ builds still print anisDarwinwarning. It comes from the externallcm_extendedflake.How to manually run this
Which issue(s) does this PR close?
Fixes
Checklist