Fly the camera to a new framing instead of cutting to it - #13
Merged
Merged
Conversation
The attribution was wrong in four ways, all of them found by reading CAreadme.txt in the library rather than the summary page. Name both licence versions. Every part that ships carries CC BY 4.0 or "CC BY 2.0 and CC BY 4.0"; only the OMR models are CCAL 2.0, which the same document defines as a deprecated spelling of CC BY 2.0. Crediting only 2.0 named the wrong licence for most of the geometry on screen. Say the models were modified. CAreadme draws the derivative-work line in exactly the place packing sits on the far side of: a model that references library parts is not a derivative, one that includes their source "in any form" is. Every packed root now carries a notice saying what packing did and what it left alone. It goes under the header rather than above it, because LDrawLoader only treats "0 FILE" as the main model when it is on line 0, and a header is read positionally. Credit the model's author in the viewer, not just the gallery. A deep link to /build/galaxy-explorer showed no sign of who built it. Credit now rides on ModelData, so it travels with the model instead of being glued into the title string. Stop signing other people's work. A free build is the person's own model, so the exporter no longer stamps it "0 Author: LDraw Builder" or licenses it on their behalf. It owes the library nothing either: the file references parts and embeds none, which CAreadme names as the case that is not a derivative work. Two repairs the above needed. pack-palette passed `index` where packModel wants `resolve` and so could not run at all, which blocked repacking the palette; it inlines 642 part files and needs the notice as much as a model does. And disposeModel had no tests, which is worth fixing for its own sake: nothing observable goes wrong when teardown is incomplete, it just leaks a few hundred megabytes per model opened.
"Quality, Brand, and the LDraw Name" says LDraw "should not appear in a program's title and/or URL to protect the uniqueness of the LDraw trademark", and that LDraw.org cannot grant an exception because it licenses the mark from the Jessiman Estate rather than owning it. It suggests an LD prefix instead, which is where LDView and LDCad get their names. "LDraw Builder" at github.com/thebuilder/ldraw-builder was outside that on both counts. The README records the reasoning next to the trademark notice, so this does not get quietly reverted as a typo later. The packed-file notice deliberately carries no tool name, so it did not have to change here.
There was no LICENSE at all, which on a public repo means all rights reserved: nobody could legally fork it. A single blanket licence would have been wrong too, because the packed .mpd files under public/ carry other people's CC BY work and are not ours to relicense. So LICENSE is split. MIT for src/, scripts/, demo-models/ and the root config; then the CC BY data called out path by path, with which version applies and why it varies at 2023-03-05, the two OMR sets named with their authors, and a statement that the packed files are modified and how. It also says plainly that the MIT grant covers copyright in the code and licenses neither trademark. NOTICES.md covers the dependencies, generated by `pnpm notices` rather than written by hand. A notices file's only real failure mode is going quietly stale when a dependency changes, and this repo already generates and commits colors.generated.ts, omr-index.json and the packed models for that reason. Now a diff in NOTICES.md is the honest answer to whether the obligations moved. Three things a hand-written file would have missed: The inventory was machine-specific. Only this Mac's prebuilt binaries install here, so a notice generated on a Mac omits what a Linux deployment ships. The families are read out of the lockfile instead: 34 variants across three, of which 3 install locally. Siblings match on the stripped prefix rather than startsWith, or @img/sharp-* swallows @img/sharp-libvips-* and reports Apache-2.0 for ten LGPL packages. libvips is LGPL-3.0-or-later, the only copyleft in the tree, so it gets its own section instead of one row among thirty. It arrives under sharp, which serves next/image; the claim that nothing imports next/image is checked with git grep at generation time rather than asserted, so it cannot rot silently. caniuse-lite is CC-BY-4.0 and needs attribution like the parts library does. The rules live in scripts/lib/notices.mjs, pure and tested, with the CLI as the I/O half. That is how ldraw-pack and omr-index are already split, and it is what lets the audit see the coverage.
Opening a bag moves the work to a different part of the model, and the camera followed by teleporting: one frame here, the next frame there. A cut between two views of the same model does not read as the camera having moved, it reads as the model having jumped, which is exactly the thing a build video never does. So frameBox now hands the move to a flight that plays out over frames. The camera is carried as a spherical offset from the point it is looking at rather than as a position, so it swings around the model the way someone walking to the other side of a table would, instead of taking the straight line through the middle of it. The near and far planes and the orbit distance limits travel with it: handing OrbitControls the destination's minimum distance up front would shove the camera back out of a close framing it has not arrived at yet. Duration comes from how far the move actually goes, so shifting to the neighbouring subassembly does not take as long as being carried the length of the model. Even the long end stays under a step of playback, so the build is never waiting on the camera. The viewer always wins: a drag, a wheel notch or a nav key calls the rest of the flight off on the spot, so the two can never pull at once. Reduced motion still snaps, and so does the first framing of a model, where there is nothing to fly from and the camera is only being placed.
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
This branch was successfully deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Opening a bag moves the work to a different part of the model, and the camera followed by teleporting: one frame here, the next frame there. A cut between two views of the same model does not read as the camera having moved, it reads as the model having jumped.
Viewport.frameBoxnow hands the move to a flight that plays out over frames.easeInOutCubicthat was already there.minDistanceup front would shove the camera back out of a close framing it has not arrived at yet.setModeland free build's initial frame passinstant.Viewport.updateNavigation(dt)becameupdateCamera(dt), since moving the camera for a frame is now either advancing a flight or reading the keys.prefersReducedMotionmoved from a private function inSceneControllertolib/dom, shared with the viewport.Verifying
Checked in a real browser with agent-browser, since the in-app preview pane freezes rAF. Burst screenshots across a FRAME toggle on the gatehouse and across a bag change on the Saturn V both show a continuous dolly through four to six distinct intermediate positions, settling on the destination rather than arriving in one frame.
pnpm test(595 passing),ultracite checkandtsc --noEmitare all clean. Three existing framing tests now run the move out before asking where the camera ended up, via aframedDistance()helper that also removes a repeated cast. New tests cover the flight itself, interrupting it, and reduced motion.For the reviewer
This branch also carries three earlier commits that predate the camera work: crediting the parts library, the rename to LDBuilder, and the licence note. The camera change is the last commit,
Fly the camera to a new framing instead of cutting to it.