Migrating from 0.31 to 0.32

"Data loaders" renamed to "importers" data-loaders-renamed-to-importers

The file import system previously called "data loaders" has been renamed to "importers" to avoid confusion with the widely-used ML/PyTorch "dataloader" concept and to better describe what the system does: importing external file formats into Rerun.

The old names are deprecated but most of them still work for this release.

Rust API rust-api

All old names are available as deprecated type aliases and will be removed in a future release.

The Cargo feature flag data_loaders has been renamed to importers.

C/C++ API cc-api

The old names are still available as deprecated aliases.

Python API python-api

The old name is still available but deprecated.

External importers external-importers

Executables on $PATH with the rerun-importer- prefix are now the canonical way to register external importers. The old rerun-loader- prefix continues to work but will log a deprecation warning.

Before:

rerun-loader-my-format

After:

rerun-importer-my-format

Lenses API (Rust) lenses-api-rust

The Lenses API has been restructured and simplified.

Entity path filtering moved to Lenses entity-path-filtering-moved-to-lenses

Entity path filtering is no longer part of the Lens itself. Use Lenses::add_lens_with_filter:

let lenses = Lenses::new(OutputMode::DropUnmatched)
    .add_lens_with_filter(EntityPathFilter::parse_forgiving("sensors/**"), lens);

This makes applying lenses to individual chunks more ergonomic.

New builder API new-builder-api

Lenses are now created through Lens::derive(), Lens::scatter(), and Lens::mutate():

// Before
Lens::for_input_column(EntityPathFilter::all(), "component")
    .output_columns(|out| { /* … */ })?
    .build()

// After - derive lens (1:1 row mapping)
Lens::derive("component")
    .to_component(component_descr, ".field")
    .build()?

// After - scatter lens (1:N row mapping)
Lens::scatter("component")
    .output_entity("/target")
    .to_component(component_descr, ".field")
    .build()?

// After - mutate lens (modifies component in-place)
Lens::mutate("component", ".field").build()

To output columns to multiple entities from a single component, multiple lenses can be registered for the same input component.

rerun rrd compact renamed to rerun rrd optimize, has profiles and new defaults rerun-rrd-compact-renamed-to-rerun-rrd-optimize-has-profiles-and-new-defaults

rerun rrd compact is now rerun rrd optimize.

A new --profile argument has been added to opt to known good values. Two profiles are available: live (optimized for the live Viewer workflow, same as previous defaults) and object-store (optimized for querying and streaming from object-store-backed storage, e.g. a catalog server).

By default, the object-store profile is now used. Use --profile live to keep the previous behavior.

DatasetEntry.register requires a sequence of URIs (Python) datasetentryregister-requires-a-sequence-of-uris-python

DatasetEntry.register no longer accepts a single URI string for recording_uri. Pass a sequence of URIs instead, and prefer batching many URIs into a single register call rather than calling register repeatedly in a loop (which is much slower).

Old single-string invocations still work at runtime but emit a DeprecationWarning.

- dataset.register(url, layer_name="base")
+ dataset.register([url], layer_name="base")

layer_name is unchanged: pass a single string to apply one layer to all recordings, or a sequence matching the length of recording_uri.

URDF importer transform entity urdf-importer-transform-entity

The URDF importer now loads the static transforms of the model to the /tf_static entity by default. This replaces the model-dependent entity path of previous versions, and improves consistency with ROS data.

A custom entity path can be now also configured in the UrdfTree API in Python and Rust, if desired.

MCAP metadata and statistics mcap-metadata-and-statistics

In MCAP to RRD conversion, metadata records, statistics, and recording info are now saved at dedicated reserved entity paths instead of RRD properties (__properties).

Metadata records are saved under __mcap_metadata, and MCAP statistics and recording info are saved under __mcap_properties.

rerun.recording deprecated in favor of RrdReader rerunrecording-deprecated-in-favor-of-rrdreader

The rerun.recording module — Recording, RRDArchive, load_recording, load_archive — is deprecated. Use rerun.experimental.RrdReader instead, which natively supports multi-store RRDs (multiple recordings and blueprints in one file) and lazy loading.