Fix handling of components that only vary by descriptors

[Wumpf]

Related

• Fixes Viewer queries incorrectly distinguishes by component metadata like archetype & field name rather than just component identifier #10903
• Aaaalmost all of Use ComponentIdentifier instead of ComponentDescriptor indexing into chunks #10460
  • scrolling over it it doesn't look like I've left opportunities out to not pass descriptors but oh my there's still more!

What

... by biting the bullet and pretty much everything that so far indexes of ComponentDescriptor instead index over ComponentIdentifier.
I'd like to believe this also improves performance, but I haven't proven it.

There's a lot of access that goes descriptor_THING().component now which strongly implies that we should generate component_THING() instead. But I've left that as an exercise for the reader.

Commit by commit if you're into this thing - every successive commit is curated to walk up the stack of crates

TODO:

• confirm the repro on Viewer queries incorrectly distinguishes by component metadata like archetype & field name rather than just component identifier #10903 behaves to our liking now

Needed follow-ups:

• Generate component_THING() methods (use them inside of descriptor_THING())
  • use it to tone down existing .component sillyness
• Use component_THING for queries in visualizers -> solve Use ComponentIdentifier instead of ComponentDescriptor indexing into chunks #10460 for good

[github-actions]

Web viewer built successfully.

Result	Commit	Link	Manifest
✅	f43c889	https://rerun.io/viewer/pr/11593	+nightly +main

View image diff on kitdiff.

^{Note: This comment is updated whenever you push a commit.}

[Wumpf]

The repro case for the original example has become a bit more pathological. but it has handled correctly.

import rerun as rr

rr.init("duplicate_tagged_components", spawn=True)
# These are two groupings of content with a variety of fields but at least one intersects
schema = rr.DynamicArchetype("mcap.Schema")._with_descriptor_internal(
    rr.ComponentDescriptor("data", "mcap.Schema"), 1.0
)
image = rr.DynamicArchetype("SomeCustomImage")._with_descriptor_internal(
    rr.ComponentDescriptor("data", "SomeCustomImage"), 2.0
)
rr.log("/camera/data", schema)
rr.log("/camera/data", image)

The first descriptor wins - we viewer only shows/receives mcap.Schema.
No warning is logged right now which is a bit unfortunate. But given how hard it is to hit this now I'm ok with that.

[Copilot]

Pull Request Overview

This PR addresses handling of components that differ only by their descriptors by transitioning the codebase from indexing on ComponentDescriptor to indexing on ComponentIdentifier. The change affects query APIs, cache keys, storage structures, and UI components throughout the viewer and SDK codebases.

Key changes:

• Query APIs now accept ComponentIdentifier instead of ComponentDescriptor for lookups
• Internal storage structures index by ComponentIdentifier with descriptors stored as values
• Cache keys updated to use ComponentIdentifier

Reviewed Changes

Copilot reviewed 152 out of 153 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
rerun_py/src/arrow.rs	Updates PendingRow component storage to use identifier-based indexing
examples/rust/extend_viewer_ui/src/main.rs	Updates component UI queries to pass identifiers
examples/rust/custom_view/src/points3d_color_visualizer.rs	Extracts component identifiers from descriptors for queries
Multiple descriptor snippet files	Updates test code to use new component access patterns
Viewport/view blueprint files	Converts descriptor-based lookups to identifier-based
Query cache and storage files	Major refactoring of indexing from descriptors to identifiers

Comments suppressed due to low confidence (3)

---

_{Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.}

[grtlr]

Wow, this is already looking much nicer 🤩!

All of my comments are minor and only suggestions / notes to myself. Thank you for going through the effort of starting to make these changes.

[grtlr]

Now I wonder if rustc will monomorphize the from_iter for IntMap into the same function as this (probably not).

[Wumpf]

Don't think so either. You're wondering whether the from_iter shortcut might be harmful?

[grtlr]

The deletion above is haunting me, but I guess that was just an early exit? Looking at the diff, we could have always just passed descr along.

Or did that line just become trivial because of @IsseW's recent refactor?

[Wumpf]

I believe recently we had to add a brunch of descriptor lookups because of the chnage to component target, and those are gone now again because the descriptor is available in a more obvious fashion

don't see why it would be needed since we're asking the same store on the latest_at query then for the existance

[grtlr]

Yes please (but not now, let's get this merged)!