feat: 6849 loading screen by data by alejandro-jimenez-dcl · Pull Request #8046 · decentraland/unity-explorer

alejandro-jimenez-dcl · 2026-04-03T06:36:31Z

Pull Request Description

What does this PR change?

Replaces count-based scene loading progress with byte-weighted tracking so the loading bar reflects actual download volume rather than asset count. Implementation is gated behind a new feature flag (byte-weighted-loading-progress) for safe remote rollback.

Closes #6849.

Key pieces:

New SceneByteProgressTracker aggregates per-entity byte progress, credits finishes/deaths, and smooths output frame-rate independently.
LoadAssetBundleSystem issues a HEAD request (new GetDecompressedContentLengthOp on the web request controller) to learn each asset's decompressed size, then reports streamed bytes via IProgress<float>.
GatherGltfAssetsSystem drives the tracker each frame and reports the smoothed value to AsyncLoadProcessReport.
When the flag is off, ContentLength stays at -1 and the progress reporter is null; the tracker math collapses to count-based behaviour, matching today's loading screen.

Other notes:

Output is capped just below 1.0; only the explicit conclude path reports 1.0 to avoid auto-closing the loading screen mid finalize-wait.
Added SceneByteProgressTrackerShould (11 cases) covering register idempotency, credit balance, monotonicity, unknown-size weighting, in-progress reset, and dt-gated smoothing.

Test Instructions

Dont forget to test LSD as well.

Test Steps
1. Launch the client with the byte-weighted flag off (default for remote users), enter a scene, and observe the loading screen.
2. Relaunch with alfa-byte-weighted-loading-progress on (editor defaults to on; remotely toggle via the feature flag service or --feature-flag arg), enter the same scene, and observe the loading screen.
3. Expected: the bar progresses smoothly to completion in both runs and the loading screen closes at the same point. Difference between the two modes should be barely perceptible, the on-path may look slightly smoother on scenes with mixed asset sizes.
4. Try a few more scenes (small, large, mostly-cached) with the flag on and off.
Additional Testing Notes
- Cached scenes: progress should still reach 100% promptly, no stalls.
- Slow network / partially failing assets: bar should not get stuck or jump backwards.
- Watch for any premature loading-screen dismissal (would indicate the < 1.0 cap is leaking through).
Quality Checklist
- Changes have been tested locally
- Documentation has been updated (if required)
- Performance impact has been considered (extra HEAD per AB, only when flag is on)
- For SDK features: Test scene is included (n/a)
Code Review Reference

Please review our Code Review Standards before submitting.

github-actions · 2026-04-03T06:36:42Z

Windows and Mac build successful in Unity Cloud! You can find a link to the downloadable artifact below.

Name	Link
Commit	`f282fff`
Logs	https://github.com/decentraland/unity-explorer/actions/runs/26943320744
Download Windows	https://github.com/decentraland/unity-explorer/suites/72316455074/artifacts/7408668759
Download Windows S3	https://explorer-artifacts.decentraland.org/@dcl/unity-explorer/branch/feat/6849-loading-screen-by-data/pr-22722-f282fff/Decentraland_windows64.zip
Download Mac	https://github.com/decentraland/unity-explorer/suites/72316455074/artifacts/7408505595
Download Mac S3	https://explorer-artifacts.decentraland.org/@dcl/unity-explorer/branch/feat/6849-loading-screen-by-data/pr-22722-f282fff/Decentraland_macos.zip
Built on	2026-06-04T10:08:31Z

AnsisMalins · 2026-04-04T09:55:44Z

PR #8046, run #23942270238

Builds: Windows change, Windows baseline, macOS change, macOS baseline

Intel Core i3

Metric	Change	Baseline	Delta	Improvement
Samples	1945	2658
CPU average	46.2 ms	33.8 ms	12.4 ms	-36.8%	🔴
CPU 1% worst	77.0 ms	57.5 ms	19.5 ms	-33.9%	🔴
CPU 0.1% worst	187.9 ms	122.3 ms	65.6 ms	-53.6%	🔴
GPU average	41.8 ms	17.7 ms	24.1 ms	-135.9%	🔴
GPU 1% worst	51.0 ms	25.1 ms	25.8 ms	-102.8%	🔴
GPU 0.1% worst	55.6 ms	28.5 ms	27.1 ms	-95.0%	🔴

dalkia

Leaving some comments for possible simplifications

…uest controller

…en-by-data # Conflicts: # Explorer/Assets/DCL/Infrastructure/ECS/StreamableLoading/AssetBundles/LoadAssetBundleSystem.cs

Address PR review feedback and assembly cycle constraints. - Drop IStreamableLoadingProgressHandler interface, use BCL IProgress<float> directly. StreamableLoadingState implements it. - Move HEAD content-length lookup out of WebRequestController layer into LoadAssetBundleSystem. WebRequestController no longer references streamable loading concepts (avoids DCL.Network to ECS.Unity cycle). - Replace progressHandler param with (expectedContentLength, progressReporter) pair across SendAsync chain. - Extract byte-weighted progress bookkeeping from GatherGltfAssetsSystem into SceneByteProgressTracker (separately testable). - Use HashSet<Entity> for tracked entities instead of Dictionary; read ContentLength fresh from StreamableLoadingState at finish time. - Fix silent cancellation leak in polled SendRequest path: switch to UniTask.Yield(token) so OperationCanceledException propagates. - Drop redundant SetProgress(0) and SetContentLength(-1) init in LoadAssetBundleSystem; pool reset already handles it. ContentLength unknown sentinel is now -1 consistently.

SceneByteProgressTracker now memoizes ContentLength on registration in a Dictionary<Entity, long>, so CreditFinish/CreditDeath can credit the stored size without re-reading StreamableLoadingState. GatherGltfAssetsSystem only calls ReadLoadingState while an entity is still in LoadingState.Loading. Average-size estimation is replaced with a fixed UNKNOWN_ASSET_BYTES (1 KB) weight for assets that finished before their size was observed. This makes the unknown-slot contribution to effectiveTotal symmetric with the credit applied in CreditFinish, removing the prior drift between snapshot-time and compute-time averages.

Add BYTE_WEIGHTED_LOADING_PROGRESS so the HEAD round-trip and progress reporter in LoadAssetBundleSystem can be skipped at runtime. Flag off leaves state.ContentLength = -1 and progressReporter null, which collapses SceneByteProgressTracker math to count-based (every entity worth one UNKNOWN slot) for safe remote rollback. Tune the tracker so the on path looks better: - UNKNOWN_ASSET_BYTES from 1024 to 1 so cache hits stop skewing the bar away from real-download bytes. - Cap output below 1.0; AsyncLoadProcessReport.SetProgress(>=1f) auto-closes the loading screen and would fire mid finalize-wait window when in-progress credits saturate effectiveTotal. Only the explicit conclude path in GatherGltfAssetsSystem should report 1.0. - Lerp toward target each frame to absorb spikes from per-entity main AB completions and bursty UWR reports.

m3taphysics · 2026-05-21T09:15:37Z

PR #8046, run #26216237624

Builds: Windows change, Windows baseline, macOS change, macOS baseline

Framework 13 i7

Metric	Change	Baseline	Delta	Improvement
Samples	2700	2700
CPU average	33.3 ms	33.3 ms	-0.0 ms	0.0%	⚪
CPU 1% worst	33.4 ms	33.8 ms	-0.3 ms	1.0%	⚪
CPU 0.1% worst	33.7 ms	37.8 ms	-4.1 ms	10.9%	🟢
GPU average	7.8 ms	7.6 ms	0.1 ms	-1.9%	⚪
GPU 1% worst	20.5 ms	21.2 ms	-0.7 ms	3.4%	🟢
GPU 0.1% worst	24.0 ms	29.1 ms	-5.1 ms	17.4%	🟢

- Drop dead Math.Max from ComputeAndClamp lerp guard. - Make smoothing frame-rate independent: take deltaTime parameter (GatherGltfAssetsSystem passes Update's t). - Log warning on GetDecompressedContentLength HEAD failures instead of swallowing silently; cancellations stay silent. - Add SceneByteProgressTrackerShould with 11 cases covering register idempotency, credit balance, monotonicity, unknown-size weighting, in-progress reset, and dt-gated smoothing. - Expose SceneLifeCycle internals to DCL.EditMode.Tests.

m3taphysics · 2026-05-21T10:51:30Z

PR #8046, run #26220799532

Builds: Windows change, Windows baseline, macOS change, macOS baseline

Framework 13 i7

Metric	Change	Baseline	Delta	Improvement
Samples	2700	2699
CPU average	33.3 ms	33.3 ms	-0.0 ms	0.0%	⚪
CPU 1% worst	33.4 ms	33.5 ms	-0.1 ms	0.4%	⚪
CPU 0.1% worst	33.5 ms	34.8 ms	-1.3 ms	3.8%	🟢
GPU average	8.3 ms	8.2 ms	0.1 ms	-1.5%	⚪
GPU 1% worst	19.8 ms	21.5 ms	-1.7 ms	7.9%	🟢
GPU 0.1% worst	28.7 ms	34.1 ms	-5.4 ms	15.9%	🟢

m3taphysics · 2026-05-21T15:16:03Z

PR #8046, run #26233784698

Builds: Windows change, Windows baseline, macOS change, macOS baseline

Framework 13 i7

Metric	Change	Baseline	Delta	Improvement
Samples	2700	2700
CPU average	33.3 ms	33.3 ms	-0.0 ms	0.0%	⚪
CPU 1% worst	33.5 ms	33.7 ms	-0.2 ms	0.6%	⚪
CPU 0.1% worst	34.4 ms	36.7 ms	-2.2 ms	6.1%	🟢
GPU average	7.8 ms	7.9 ms	-0.1 ms	1.0%	⚪
GPU 1% worst	20.0 ms	20.5 ms	-0.6 ms	2.8%	⚪
GPU 0.1% worst	25.8 ms	28.2 ms	-2.3 ms	8.3%	🟢

m3taphysics · 2026-05-22T10:22:36Z

PR #8046, run #26281706672

Builds: Windows change, Windows baseline, macOS change, macOS baseline

Framework 13 i7

Metric	Change	Baseline	Delta	Improvement
Samples	2701	2700
CPU average	33.3 ms	33.3 ms	0.0 ms	-0.0%	⚪
CPU 1% worst	33.8 ms	33.7 ms	0.0 ms	-0.1%	⚪
CPU 0.1% worst	37.7 ms	37.3 ms	0.5 ms	-1.3%	⚪
GPU average	8.3 ms	7.9 ms	0.3 ms	-4.4%	🔴
GPU 1% worst	19.2 ms	19.1 ms	0.1 ms	-0.3%	⚪
GPU 0.1% worst	27.2 ms	24.2 ms	3.0 ms	-12.3%	🔴

sandrade-dcl

LGTM. Good job!

…en-by-data # Conflicts: # Explorer/Assets/DCL/PluginSystem/World/AssetBundlesPlugin.cs

m3taphysics · 2026-06-04T09:26:38Z

PR #8046, run #26942264212

Builds: Windows change, Windows baseline, macOS change, macOS baseline

How to read this table

Each build is measured 3 times. The values are the median, and (min–max) is the lowest and highest of those runs — a wide range means the metric is noisy and small differences are not trustworthy.
Δ is Change minus Baseline (a negative Δ means Change is faster).
🟢 faster / 🔴 slower — a real difference: larger than both 3% and the run-to-run range.
⚪ within noise — the difference is smaller than how much the build varies between its own runs, so it cannot be told apart from random variation. Treat it as no change.
Exceptions logged — exceptions found in the run logs; more than the baseline is flagged 🔴 even when frame times look fine.

Framework 13 i7

Metric	Baseline	Change	Δ	Result
Samples	2700 (×3)	2701 (×3)
CPU average	33.3 ms	33.3 ms	-0.0 ms	⚪ within noise
CPU 1% worst	34.2 ms (33.8–35.0)	33.6 ms (33.5–33.6)	-0.6 ms	⚪ within noise
CPU 0.1% worst	37.0 ms (35.4–49.7)	36.0 ms (34.0–36.4)	-1.0 ms	⚪ within noise
GPU average	7.6 ms (7.1–8.2)	9.7 ms (8.5–9.7)	2.1 ms	🔴 28% slower
GPU 1% worst	19.8 ms (19.7–20.0)	17.8 ms (17.6–18.8)	-2.0 ms	🟢 10% faster
GPU 0.1% worst	24.6 ms (24.2–24.8)	21.4 ms (18.4–25.9)	-3.2 ms	⚪ within noise
Exceptions logged	48	46	-2	🟢 fewer errors

m3taphysics · 2026-06-04T10:37:17Z

PR #8046, run #26945250405

Builds: Windows change, Windows baseline, macOS change, macOS baseline

How to read this table

Each build is measured 3 times. The values are the median, and (min–max) is the lowest and highest of those runs — a wide range means the metric is noisy and small differences are not trustworthy.
Δ is Change minus Baseline (a negative Δ means Change is faster).
🟢 faster / 🔴 slower — a real difference: larger than both 3% and the run-to-run range.
⚪ within noise — the difference is smaller than how much the build varies between its own runs, so it cannot be told apart from random variation. Treat it as no change.
Exceptions logged — exceptions found in the run logs; more than the baseline is flagged 🔴 even when frame times look fine.

Framework 13 i7

Metric	Baseline	Change	Δ	Result
Samples	2701 (×3)	2701 (×3)
CPU average	33.3 ms	33.3 ms	-0.0 ms	⚪ within noise
CPU 1% worst	33.8 ms (33.7–34.1)	33.5 ms (33.4–33.6)	-0.2 ms	⚪ within noise
CPU 0.1% worst	36.3 ms (36.2–38.1)	35.0 ms (34.0–35.9)	-1.4 ms	⚪ within noise
GPU average	8.2 ms (7.2–8.3)	9.6 ms (8.5–10.0)	1.4 ms	⚪ within noise
GPU 1% worst	17.5 ms (17.3–18.8)	17.9 ms (17.0–18.1)	0.3 ms	⚪ within noise
GPU 0.1% worst	23.5 ms (22.5–23.8)	23.1 ms (19.4–23.9)	-0.4 ms	⚪ within noise
Exceptions logged	44	44	0	⚪ none new

DafGreco

✔️ PR reviewed and approved by QA on both platforms following instructions playing both happy and un-happy path

Regressions for this ticket had been performed in order to verify that the normal flow is working as expected:

[ ✔️ ] Backpack and wearables in world
[ ✔️ ] Emotes in world and in backpack
[ ✔️ ] Teleport with map/coordinates/Jump In
[✔️ ] Chat and multiplayer
[ ✔️ ] Profile card
[✔️ ] Camera
[✔️ ] Settings

Evidence without flag ✅

20260604-1032-53.9187944.mp4

Evidence with flag ✅

evidenceloading.mp4

alejandro-jimenez-dcl added 3 commits March 18, 2026 09:43

feat: progress is now reported with byte-weighted tracking

b62f498

fix: wip

7fcd792

fix: update progress computation and handle unknown-size

20f96d2

alejandro-jimenez-dcl changed the title ~~Feat/6849 loading screen by data~~ feat: 6849 loading screen by data Apr 3, 2026

alejandro-jimenez-dcl added the force-build Used to trigger a build on draft PR label Apr 3, 2026