sorbet.semantic_family dropped from query results

[jleibs]

This appears to be a consequence of:

• https://github.com/rerun-io/rerun/pull/8499/files

This breaks the new send_dataframe functionality introduced in:

• Introduce new API to send a dataframe to Rerun #8461

Code to repro:

from __future__ import annotations

import tempfile

import rerun as rr

with tempfile.TemporaryDirectory() as tmpdir:
    rrd = tmpdir + "/tmp.rrd"

    rr.init("rerun_example_test_recording")
    rr.set_time_seconds("my_index", 1)
    rr.log("points", rr.Points3D([1, 2, 3]))
    rr.save(rrd)

    recording = rr.dataframe.load_recording(rrd)
    assert recording is not None

df = recording.view(index="my_index", contents="points").select().read_all()
print(df.schema.field("/points:Position3D").metadata)

Prior to https://github.com/rerun-io/rerun/pull/8499/files

➜ python dump_tags.py 
{b'sorbet.path': b'/points', b'sorbet.logical_type': b'positions', b'sorbet.semantic_family': b'Points3D', b'sorbet.semantic_type': b'Position3D'}

After https://github.com/rerun-io/rerun/pull/8499/files

➜  python dump_tags.py 
{b'sorbet.path': b'/points', b'sorbet.semantic_type': b'Position3D'}

[jleibs]

I checked on the 0.19 / 0.20 releases.

Even though the code existed to set semantic_family:

rerun/crates/store/re_chunk_store/src/dataframe.rs

Lines 276 to 281 in d44b701

	archetype_name.map(|name| {
	(
	"sorbet.semantic_family".to_owned(),
	name.short_name().to_owned(),
	)
	}),

This repro code shows it wasn't being set there either. We just didn't notice.

[jleibs]

Can confirm this can be worked around by specifying include_indicator_columns = True on the view. This makes sense for a pre-tagged-component world.