CI and release builds use the newest Node 22.x (check-latest; the runner
image can lag behind 22.23.4), and the web UI image's frontend stage is
pinned to node:26.12.1-alpine. Node only runs at build time; no shipped
artifact contains it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01JoFXdUvVCDAG8xRjtp52qt
* Add regression cases for avatar framing, prefixed healthchecks and browser payload access
* Cover SMB permission classification and browser file staging regressions
* Exercise avatar position application and browser ELF selection in browser CI
* Document reverse proxy setup and exercise intact SMB folder moves in CI
* fix: browser intake, avatar framing and prefixed healthchecks
* test: inspect SMB metadata before moving a populated game folder
* fix: share browser payload routes with engine state
* fix: translate browser flows and validate prefixed Docker hosting
* test: isolate writable Samba data from daemon runtime files
* fix: close SMB readers before moving inspected collection folders
* test: capture Samba rename decision and fixture permissions in CI
* test: use an authenticated Samba account for move contracts
* fix: forward browser crop coordinates and remove CI timing races
The page the helper serves on the PS5 itself: client/src/console/ built by
vite.console.config.ts into one self-contained dist-console/index.html
(script and styles inlined, safari13, 253 KiB raw / 84.5 KiB gzip; the
build fails over 450 KiB raw or 150 KiB gzip). payload/Makefile gzips it
into build/webui_page.c and the helper serves it at GET /, with
GET /api/token for the per-start token.
- Shell: top bar (logo, console chip from /api/status), rail of the
implemented sections (Home, Files, Settings placeholders), hint bar
showing only verified bindings, reload banner on a 401.
- nav.ts: spatial focus (aligned neighbours first), the R11 default input
table (keyboard + Gamepad "standard", all unverified), D-pad repeat.
- api.ts: token fetched once, X-PS5U-Token on every POST, 401 → expired.
- strings.ts: cw_* keys from all 21 locales via build-console-strings.mjs
(vm-eval like the i18n gate), English fallback.
- CI: client npm ci before every make payload.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01RxCGb8Asp1A1AQYiv1kbgS
A release event runs on the tag's ref, and the github-pages environment
only deploys from main: every release showed a failed deploy ('Tag
v7.0.6 is not allowed to deploy to github-pages'). The deploy after
publish (from main) already covers new releases, and the daily run
picks up a hand-promoted one; the page reads releases live anyway.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Ub1mez9eaVtTF8pMEFXvA9
site/ is a static page in the app's own look (light and dark, the app's
screenshots): features, downloads for every platform with the visitor's
system first, the Docker command, news from the release notes, community
links. It reads the releases from GitHub's API live, falling back on
site/releases.json (scripts/site-releases.mjs), which the pages workflow
rewrites on every deploy: after each publish run, on release edits (promoting a
pre-release), on changes to site/, and daily.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Ub1mez9eaVtTF8pMEFXvA9
The logo was only applied when the Android project was first created, so an
existing one kept Tauri's placeholder, and the round icon was a small tile
inside a circle. scripts/release/android-icons.py draws the set from the icon
source (adaptive logo in the safe zone over white; full-bleed white tile and
disc for older launchers), and the post-init patch copies it on every build,
local and CI.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Ub1mez9eaVtTF8pMEFXvA9
Deletes tests/install-fallback-hw.mjs, the mobile audit probe and its allowlist,
scripts/requirements.txt, tests/lab/verify-pending.sh, bench/baselines/ and client/icons/
(nothing references them; the bundle icons live in client/src-tauri/icons). audit-scripts
lists fpkg-diff.py as a manual research tool and no longer counts a script naming itself as
a reference. SPEC.md no longer says zip/7z/RAR sources fall back to FTX2.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Ub1mez9eaVtTF8pMEFXvA9
- Bug report wizard (4 steps, time range, GitHub/Discord/zip), redacted on both sides.
- Engine event journal + app journal merged into one timeline; helper stderr mirrored from each PS5.
- Results that finish after a console switch land on the console they started on.
- Faster local gate and CI (nextest, parallel lanes, sharded C interop/ASan, cache restore/save).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Ub1mez9eaVtTF8pMEFXvA9
Cuts the vX.Y.Z tag scripts/release.sh would push, at the dispatched commit, for when a
release has to be cut without a local clone that can push tags. A tag pushed with
GITHUB_TOKEN does not trigger publish.yml's tag push, so the workflow dispatches it.
Claude-Session: https://claude.ai/code/session_01Vq9BzVoith72tgD87LGG2z
Co-authored-by: Claude <noreply@anthropic.com>
* chore(deps): Bump rustls from 0.23.43 to 0.23.45 in /client/src-tauri
Bumps [rustls](https://github.com/rustls/rustls) from 0.23.43 to 0.23.45.
- [Release notes](https://github.com/rustls/rustls/releases)
- [Changelog](https://github.com/rustls/rustls/blob/main/CHANGELOG.md)
- [Commits](https://github.com/rustls/rustls/compare/v/0.23.43...v/0.23.45)
---
updated-dependencies:
- dependency-name: rustls
dependency-version: 0.23.45
dependency-type: indirect
...
Signed-off-by: dependabot[bot] <support@github.com>
* Install: drop the staged-refusal retry guard and the stream reach pre-check
After the PS5 refused a package from its own storage (0x80b2116f), the app
would no longer run Install on that row again and blocked Retry with Stream
unless the engine had offered it. Install and Retry with Stream now always
go out.
The engine no longer probes whether the console can reach it before handing
over a stream URL. A console that never fetches still gets the firewall /
PS5UPLOAD_PKG_HOST_IP explanation from the post-failure path.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vq9BzVoith72tgD87LGG2z
* Helper: everything written on the console is 0777
Since 6.0 an upload kept the computer's file modes, so a game folder sent
from Windows landed as 0644 files. The PS5's app loader refuses game files
without world-execute, so those games would not launch until the user ran
chmod 777 by hand. The 5.x helper forced 0777 for this reason.
Uploads, on-console copies and the file manager's writes and new folders
now land as 0777. fs_chmod still sets exactly the mode it is given.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vq9BzVoith72tgD87LGG2z
* Fix launch refusals, folder permissions, abandoned upload space and overestimated free space
- A launch refused with 0x80940033 now says whether kstuff is missing; with kstuff running,
a folder game is made 0777 and launched again. Folder games get a Fix permissions button.
- The helper's own ps5upload folders are 0777, and older ones are opened up on start.
- The helper's job cleanup also deletes the half-written copy of an abandoned upload.
- An upload that runs out of room part-way teaches the engine how much of that drive the
console never allocates; later space checks and Volumes' safe figure leave it out.
- Queue rows that failed because the PS5 could not reach this computer offer the
Windows firewall fix; FAQ lists the ports and the 0x80B2116F firmwares.
- lf_slots_wedge: a bound that a healthy ASan run fits.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vq9BzVoith72tgD87LGG2z
* 6.2.0: games launch again, installs stop giving up early
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vq9BzVoith72tgD87LGG2z
* i18n: add the launch-refusal and fix-permissions strings to en, allowlist them for other locales
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vq9BzVoith72tgD87LGG2z
* ava1: a finished job counts the bytes still awaiting their Received
bytes_sent grew only on a Received the send loop read, and the loop stops at JobDone. A frame
still in flight then went uncounted although the job holds it: a list upload reported 19000 of
26000 bytes under the coverage build.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vq9BzVoith72tgD87LGG2z
* ava1: a lane stopped by its LaneDown never picks a frame afterwards
A lane task checked its stop flag only at the top of its loop. One that passed the check just
before LaneDown, and reached its pick after lane_death had swept the lane, charged a frame to
the dead lane: the frame sat in inflight with no lane to carry it, nothing queued, so the stall
rule never fired and the job waited forever (the soak test's resume hung under the coverage
build). The flag is now read again under the sched lock, which lane_death's sweeps take, so a
pick is either swept or sees the stop. The pick moves into pick_for_lane for a deterministic test.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vq9BzVoith72tgD87LGG2z
* ci(coverage): run the engine tests once for both reports, 60 min limit
scripts/coverage.mjs ran `cargo llvm-cov --lcov` and then `--html`, each a full instrumented
test run. One pass now takes ~34 min, so the second ran into the 40 min limit and the job was
cancelled (it was red on main too). The tests now run once (--no-report) and both reports come
from that run's profiles via `cargo llvm-cov report`; the limit is 60 min for headroom.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vq9BzVoith72tgD87LGG2z
---------
Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
Co-authored-by: Claude <noreply@anthropic.com>
* docs: 6.0.0 changelog, README, FAQ and contributor docs
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XR8fQMhp7Rdgq2e4U9oDff
* Release prep: version 6.0.0
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XR8fQMhp7Rdgq2e4U9oDff
* fix(ci): keep the leaked test pools reachable for LSan; skip the tiny-file floor under coverage
- ava1-ctest diag/mgmt_fs: the &'static Pool each test leaks for
AvaTransport::with_pool is now held in a static list, so LeakSanitizer sees
it (and the sessions on its threads) as reachable instead of reporting
~140 KB of indirect leaks the frame-name suppression could never match.
Verified under clang ASan+LSan in Linux Docker: diag 13/13, mgmt_fs 40/40.
- download.rs tiny-file throughput floor is skipped when cargo-llvm-cov runs
(CARGO_LLVM_COV set): instrumentation slows every call several times over.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XR8fQMhp7Rdgq2e4U9oDff
* fix(test): data_rust servers use the clients' fast timing
Five folder-host tests gave the client fast timing (dead after 500 ms) but
left the server on the default (ping every 2 s). On a slow runner a half-second
gap made the client drop healthy lanes and the job stalled until the 36 s
progress limit (both CI runs on #374/#377). The ordered-download tests already
set both ends; these now do too. The app uses one timing on both ends.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XR8fQMhp7Rdgq2e4U9oDff
* fix(test): the C lane socket-buffer test accepts Linux's wmem_max cap
Same cap as the Rust lane test: Linux CI gives the send side 416 KiB, so the
1 MiB floor failed under ASan in CI. 128 KiB is still far above a default.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XR8fQMhp7Rdgq2e4U9oDff
* fix(ci): gcc unused-result in the FTP selftest, serial ctest in make test, sender state on failure
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XR8fQMhp7Rdgq2e4U9oDff
* fix(ava1): flush a partial bundle after any reader message, not only a record
The last small-file record could arrive while a large file's chunk or root
was still queued behind it: the flush saw a non-empty channel, the next
message was not a record, and the records sat unsent until the receiver
cut the job as stalled (CI-only flake; 7/36 under local parallel load, 0/48 after).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XR8fQMhp7Rdgq2e4U9oDff
* fix(validate): run the C interop tests serially in make ci too
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XR8fQMhp7Rdgq2e4U9oDff
* fix(ava1): a resumed file goes in one-group pieces so a flapping link still makes progress
With kills every 2 s, the 4 MiB tail piece was cut before it was durable on every
attempt and the upload retried the same 28 MiB mark until it timed out (1 in 3
locally, CI coverage). Resumed files now send 1 MiB pieces: 5/5 in ~6 s.
Also: make docker-engine uses the repo root as context (as CI); governor
classification asserts are skipped under the sanitizers, where the C is the bottleneck.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XR8fQMhp7Rdgq2e4U9oDff
* fix(test): single-instance gate tests hold the C-server lock; relay test cap leaves margin
Coverage runs the C tests in parallel; the two gate tests stop the one global C server.
The relay order test's 4 MiB/s cap finished in 5.9 s on CI against its >6 s precondition.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XR8fQMhp7Rdgq2e4U9oDff
* fix(test): periodic-kill bound widened under coverage; tiny-file floor relaxed on CI runners
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XR8fQMhp7Rdgq2e4U9oDff
* fix(test): lane-kill test kills until a resend; periodic-kill liveness race skipped under coverage
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XR8fQMhp7Rdgq2e4U9oDff
* fix(test): a hung relay fails its test at the bound instead of hanging the CI job
The dest-ends relay test ran the relay under spawn_blocking: on a hang the timeout
panicked but the runtime then waited forever for the blocking thread (27 min on CI).
It now runs on a plain thread. The engine test steps get a 25-minute bound.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XR8fQMhp7Rdgq2e4U9oDff
* fix(test): dest-ends relay test in its own binary on small files; relay logs its silent retries
The test hung a CI job twice inside the shared relay binary (behind the 384 MiB relays'
lock) and never reproduced locally or in Linux Docker. It now runs alone, on 4x32 MiB
(still twice the relay buffer), ~3 s. The relay's open-dropped and retry paths now say
so on stderr instead of looping silently. Governor classification asserts also skip
under coverage, where the instrumented sender is the bottleneck.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XR8fQMhp7Rdgq2e4U9oDff
* fix(test): the partly-durable relay test caps its first attempt so it can be interrupted
On a fast runner the uncapped relay finished all 160 MiB before the first durable
report was observed. The first attempt now reads A through a 32 MiB/s proxy (~5 s);
the resume runs uncapped.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XR8fQMhp7Rdgq2e4U9oDff
* feat: one-time 6.0 upgrade clean-up that keeps every setting
Host (engine, once per data dir): pre-5.x files no 5.41/6.0 code reads (api_token,
ps5upload.ini, ps5upload_profiles.ini/.json, app-config.json) move to ~/.ps5upload/legacy-5x/.
App (desktop, once): the 1.x/2.x config.json/profiles.json move to legacy-5x/; the cached
ezremote-dpi.elf nothing loads is deleted. Client: the retired upload-streams key is dropped.
Console (once per console, on its first 6.0 helper answer): the FTX2 /data/ps5upload/tx and
/spool folders are deleted; a failure retries later. The health check no longer lists those
folders (its fix would have recreated them). Nothing current is moved or reset.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XR8fQMhp7Rdgq2e4U9oDff
* fix(test): the download periodic-kill race is skipped under coverage, like its upload twin
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XR8fQMhp7Rdgq2e4U9oDff
* fix(test): the tiny-zip floor only catches a stall on CI runners
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XR8fQMhp7Rdgq2e4U9oDff
* fix(test): 'other calls still answer during launches' is a state check, not a wall-clock bound
Both mgmt_t6 tests asserted a sub-300 ms answer while a slow Sony call ran; on a loaded
runner that separates nothing. They now assert the answer arrived while the slow call
was still in flight.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XR8fQMhp7Rdgq2e4U9oDff
* fix(ci): reword the 6.0 migration comments for check-no-ftx2
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XR8fQMhp7Rdgq2e4U9oDff
---------
Co-authored-by: Claude Sonnet 5.5 <noreply@anthropic.com>
* fix: installs the PS5 declined, Docker images that work, Payload Manager as a backup loader (5.41.0)
Installs
- A package the console serves itself (Upload & install, USB, the phone app)
is refused on FW 13.60 with 0x80B2116F (Sony's DbgGetPatchInfo/DbgCancelPatch
path, especially over an installed app). The unified install now re-serves
the console's copy from the engine (ps5:// read back through the helper) and
a refused link through the engine's link proxy. HW-verified on a 13.60 Pro:
stream, Upload & install fresh and over an installed app, USB.
- Installer 1.2.0: fixed past Last-Modified, logs to /data/ps5upload/installer.log.
- NET_REACH (frame 148/149): the console checks it can reach the engine before
a stream install and names a firewall/isolation cause in seconds.
- Staged-route refusals get their own reason and message (staged_refused).
- Folder-game register on 13.x: sceAppInstUtilAppInstallAll fallback,
/system_ex rw remount on EROFS, repeated uninstall for its two records.
- Removed the incorrect "no fake packages above FW 11.60" warnings.
Loaders
- itsPLK Payload Manager (:8084) fallback: starts the installer, recovers a
wedged elfldr (HW-verified) and puts the helper back when :9021 is dead.
Only ever uses its own ps5upload-* folders.
- Setup wizard skips payloads already running; "Send only ps5upload".
Engine / self-hosting
- Engine clients ignore proxy env vars; errors carry the real cause.
- Docker: writable /tmp and /data (also under a user: override), HEALTHCHECK
via --healthcheck, PS5UPLOAD_ALLOW_IP takes CIDR ranges, actionable 403,
bridge-network warning, engine/compose.yaml, scripts/docker-smoke.sh run in
the PR gate and before every release push.
- Desktop app with a remote engine browses the engine's disk; From this device
uploads first. Android engine listens on the LAN so the phone can stream.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* test: accept Windows' timeout as the probe's underlying cause
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* style: rustfmt the probe test
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* feat(elfldr): vendor elfldr with a read deadline, embedded in the engine
elfldr serves one connection at a time and its reads had no deadline,
so a client that went silent without closing (a dropped link, a machine
or console asleep mid-send) left it running but deaf on :9021 — users
had to load elfldr again. Measured on the Phat (FW 5.10): the stock
build answered nothing while a silent connection stayed open; with a
15 s SO_RCVTIMEO it dropped that connection and answered the next
request at once, and a payload sent with 10 s pauses still loaded.
third_party/elfldr is upstream 02cfe91 plus ps5upload.patch (GPLv3,
credited), with the tested ELF checked in so every engine build embeds
it; `make elfldr` rebuilds it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* feat(elfldr): the engine swaps a console's stock elfldr for the patched one
GET /api/ps5/elfldr/health tells healthy / stuck / absent from a request
elfldr answers without running anything. POST /api/ps5/elfldr/ensure
replaces the standard elfldr.elf with the embedded patched build when it
answers and isn't the one this engine installed (a reboot or autoloader
brings the stock one back), leaves any other loader on :9021 alone, and
returns once the new one answers. On the Phat: upgraded (pid 685,
answering at once), then 'current'; the helper was untouched.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* ci(publish): give each client build 75 minutes
The Intel macOS build ran past the 45-minute limit for v5.37.0 (about
33 minutes for v5.35.0 and v5.36.0), which cancelled it and skipped
create-release, so v5.37.0 was never published.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* test(engine): history tests get their own temp dirs
TmpDir named directories by pid and nanoseconds, but macOS time ticks in
microseconds: two tests starting together shared a directory and one
deleted it under the other (a_legacy_host_port_file_is_still_read read
nothing). A counter makes each name unique. The elfldr embed test reads
the embedded bytes directly instead of clearing an env var mid-run.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* feat(elfldr): the app keeps the patched elfldr in place and never sends into a stuck one
Before sending the helper, the app checks the console's loader: a stuck
one (took the connection, never answered) is waited out for 20 s — the
patched build clears in 15 — and one that stays stuck is reported
plainly ('load elfldr again') instead of a socket error. Once the helper
is up, the engine swaps the stock elfldr for the patched build: at once
after a fresh helper send (a reboot or wake brings the stock one back),
else at most every 10 minutes per console. Desktop, browser and Android
share the engine routes; strings in all 20 languages.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* docs(changelog): 5.37.1 — elfldr recovers by itself; 5.37.0 folded in
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(elfldr): swap it in when the helper is already up at launch
Only ensurePayloadCurrent asked for the patched elfldr, and a helper
that is already up when the app opens never goes through it — on the
Phat the stock elfldr stayed. The status poll's 'arrives at up' edge
(first sighting included) now asks too, after the same settle as the
log capture and never during a transfer.
Hardware (Phat FW 5.10, web build and Android app): stock elfldr
replaced ~10 s after opening; a silent :9021 connection held while the
helper was shut down — the helper came back through the patched
elfldr; with the stock elfldr wedged, the Android app sent nothing for
150 s (health: stuck), then redeployed and swapped elfldr once released.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(convert): the compress stage pays for reading and writing on one thread
kraken_image::compress runs the encoder on a pool, but one thread reads
the source and writes each compressed block, so the stage takes the
longer of encoding and read+write — not the slower of reading and
encoding, as the estimate assumed. With the Windows Battlefield 6 run's
speeds the model now gives 75 min for a 71-min build.
The estimate also reports the speeds it measured (rates), because the
small read/write samples can be fooled by caches: on a source read an
hour earlier the read sample hit the file cache at 8.6 GB/s and the
estimate came out at 25 min for a 54-min build. Calibrating from past
builds is the follow-up.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* docs(changelog): build-time estimate, stated as measured
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* ci(docker): 25 minutes for the PR image checks
Both PR gates build the whole image (engine, and the client for the web
UI one) in about 7-10 minutes; the engine gate took 9m55s and the web UI
gate was cancelled at its 10-minute limit on PR #338.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
The payload artifact keeps payload/installer/ps5upload-installer.elf as
installer/ps5upload-installer.elf; the image jobs still copied it from the
old dpi/ path and failed on the v5.35.0 tag.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
