CI on a headless ubuntu-24.04-arm runner caught what a desktop Pi could not:
the AppImage only started on a machine that already had a full desktop stack
installed. Three distinct causes, all from bundling Debian's builds of
libraries that Debian builds for a co-versioned system, which is the opposite
of an AppImage's situation.
1. Hard-linked backends. Debian's libSDL2 lists libpulse, libasound, libX11
and libwayland-client as DT_NEEDED rather than dlopening them, so the
loader demanded all four at startup; the CI job failed with
"libpulse.so.0 => not found". Debian's OpenAL does the same through
libsndio, which itself hard-links libasound. Built from source with
--enable-*-shared and ALSOFT_DLOPEN, both dlopen their backends, so the
image now runs on a Wayland-only session, a KMSDRM handheld with no X
server, or a box with ALSA and no PulseAudio.
2. A stray link. Debian's libtheoradec is linked against libcairo, which
drags in X11, xcb, fontconfig and freetype for a video decoder.
--disable-examples leaves it needing only libogg.
3. SONAME collision with the host. OpenAL dlopens ALSA, ALSA's config loads
its PulseAudio hook plugin, and that plugin pulls the host's libsndfile
into the process. libsndfile links libogg, libvorbis and libmpg123 -- the
same three we bundle -- and since the loader resolves a SONAME once per
process it bound to our bullseye copies. A bullseye libmpg123 has no
mpg123_info2 (added in 1.32), so the plugin failed to relocate, ALSA
config collapsed, and the game ran with no audio device at all. Building
them current means our copies satisfy the host's libsndfile instead of
starving it.
The general rule, now stated as an assertion instead of a comment: never
bundle a library the host's own stack may also load unless ours is at least
as new as theirs. build_appimage.sh fails if any shipped object hard-requires
anything beyond glibc, libstdc++ and the font stack, and CI re-checks it on
the extracted artifact.
Host requirements drop from "a working desktop" to glibc 2.29+, libstdc++,
libfreetype6 and zlib. Bundled libraries drop from 13 to 10: libcairo,
libpixman and libsndio are gone entirely.
Verified on a Raspberry Pi 5 (trixie, Wayland): boots, imports, plays, and
audio works -- SDL 2.30 now picks the native Wayland backend rather than
falling back to XWayland as bullseye's 2.0.14 did.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
scripts/build.sh's `linux` target only ever produces x86_64: it unpacks
LOVE's official love-11.5-x86_64.AppImage and re-fuses game.love into it.
There is no aarch64 equivalent to unpack -- LOVE 11.5 publishes win32,
win64, macOS, Android, iOS and exactly one x86_64 AppImage -- so arm64
desktop Linux (Raspberry Pi 4/5, Armbian, arm64 VMs on Apple Silicon) had
no artifact at all.
Compile LOVE 11.5 from the official linux-src tarball instead, inside a
Debian bullseye arm64 container, and assemble the AppImage from scratch.
Both pinned inputs (the LOVE source tarball and the AppImage type-2
runtime, on a dated tag rather than `continuous`) are SHA-256 verified on
the host, so the container runs with no network access.
Bullseye is the compile environment, not a claim about where the artifact
runs: glibc is backward but not forward compatible, so linking against the
oldest supported glibc is the only thing that makes one artifact work
everywhere. The binaries come out needing only glibc 2.29 / GLIBCXX_3.4.21,
covering Raspberry Pi OS bullseye through trixie and Ubuntu 20.04 onward.
The dependency walker copies in LOVE's own libraries and leaves the
driver-coupled, loader-coupled and font-stack libraries to the host. That
last category is not cosmetic: Debian's libtheoradec is linked against
libcairo, so a host cairo gets loaded into the process, and because the
loader resolves one SONAME once per process it then binds to whatever
libfreetype we bundled -- bullseye's 2.10.4 has no FT_Get_Transform, which
cairo 1.18 needs, and the game died at startup with a symbol lookup error.
Excluding the whole font stack makes the process self-consistent.
CI gets three path-gated jobs: an offline selftest on ubuntu-latest (pins,
the host-arch guard, the exclude list, the AppRun fusion contract), a real
build on ubuntu-24.04-arm that asserts the layout, that every bundled
object resolves under AppRun's LD_LIBRARY_PATH, and that the glibc floor is
still <= 2.31, and a release job that reuses the shared game.love payload.
None of it needs secrets or self-hosted hardware, so it runs on fork PRs.
Verified end to end on a Raspberry Pi 5 (Debian trixie, Wayland): the
launcher boots from the AppImage and renders correctly.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Ships a dual-NRO native launcher that checks GitHub Releases, updates both NROs from gen1recomp-*-switch.zip with matching NACP versions, and stays silent unless an update needs confirm.
The launcher spent ~9ms per frame building and drawing, and the Find Mods
tab could hang the window for minutes. Both had the same root cause: a
retained UI tree rebuilt every frame, and blocking curl calls made from the
draw path.
Replace the vendored FlexLove engine (28.5k lines) with src/ui/kit/ (Kit,
Theme, Layout, Loader). The kit caches Text objects and all measurement,
allocates nothing in the steady state, and draws flat. Build+draw is now
under 1ms at every window size and on every tab (POKEPORT_LAUNCHER_PROF).
Move every network call off the render thread onto a love.thread pool
(src/net/Fetch.lua): mod index fetches, per-mod release checks, find-tab
stats, thumbnails and mod installs. Mod indexes prewarm at boot so the
Find Mods tab is populated before it is opened.
Paginate every list -- mods, find, save slots, settings, release notes,
versions -- with the page size derived from the real viewport height, so a
500-mod index costs what a 10-mod one does. Scrolling is gone.
Anything that waits now raises a non-dismissable loader; per-row background
work shows an inline spinner instead. The in-app updater moves to the top
right beside the settings gear and pulses when an update is waiting.
Theme is black with white outlines, no gradients or glows, and solid
colour-coded embossed buttons with bold labels. The game tabs keep their
cartridge colours. Everything is 1.3x larger. The save editor shares the
theme, and adding an item there is now a searchable pop-up like adding a
Pokemon.
Also:
- Reset rebinds, in Settings and under Touch Controls. Rebinds are additive
(Input:applyBindings layers them over the defaults), so there was no
in-game way to undo one.
- Launch options: --game red [--slot N] / POKEPORT_GAME boots straight into
a game for shortcuts and frontends, falling back to that game's tab when
its ROM is not imported.
Fixes found while porting:
- Ellipsis and letterspacing truncated bytes, not codepoints, so a
multi-byte mod name crashed the first frame on a Japanese index.
Measurement no longer throws on malformed input either.
- The new font set missed UiFont's kana fallback, rendering translated
builds as tofu.
- Fetch workers idle in Channel:demand() and LOVE waits for live threads at
exit, so the process outlived the window; quitting mid-download also
waited on curl's 300s ceiling. Shut the pool down in love.quit and bound
its transfer timeouts.
- In one column the save-slot card drew below the fold, over the footer,
with no scrollbar left to reach it.
The two FlexLove engine tests guarded a scroll manager and an auto-height
propagation bug that no longer exist; replace them with a kit suite covering
page bounds, viewport sizing and UTF-8 truncation, and retarget the NX test
to assert the dependency is gone rather than that its perf guards are set.
The originals run GiveItem before printing the received texts, and when
the bag can't hold the TM they print a make-room line instead and leave
EVENT_GOT_TM* unset, so talking to the leader again retries the give.
The victory reward path added the TM straight into the inventory, so a
full bag went to 21/20.
Route the gym TM give through Bag.add, split the TM lines out of the
victory dialogue table into tmPre/tmDialogue/noRoom, and port the
beaten-leader middle branch that re-runs the ReceiveTM script. Saves
that already hold the TM without the flag count as received so they
don't collect a second copy.
Refs #797
* Resolve Find Mods stats from each mod's GitHub repo when the feed lacks them
A FIND MODS row now shows download/date stats even when its feed publishes
none: the row fetches the mod's own GitHub releases through the same
cached ModUpdate.fetchReleases the MODS tab uses (six-hour options cache,
so an installed mod's repo is instant). Feed-published stats still win
when present; otherwise one repo is fetched per frame -- the thumbnail
budget pattern -- so opening the tab never stalls for the whole listing.
ModUpdate.statsForReleases is the shared resolver.
* Fix crash opening the Find Mods tab: rename the stats cache field
The resolver stored results in self._findStats, which collides with the
method of the same name: self._findStats resolves through the metatable to
the function, so the or {} guard never fired and indexing it crashed the
launcher the moment the panel built. State now lives in _findStatsCache.
* Fix Find Mods crash: require ModUpdate in the find panel
buildFindPanel called ModUpdate.statsLine without a local require --
only buildModsPanel had one -- so opening the tab indexed a nil global.
* Retry Find Mods stats after failed repo fetches
A failed repo fetch (hourly GitHub API rate limit, transient network error)
was memoized as resolved, so a rate-limited first visit left those rows
empty for the whole session. Failures now schedule a 60s retry; a 404 is
still permanent so a renamed or vanished repo is fetched once.
* Add the MODS tab sort options to the Find Mods tab
lets a translation keep chosen characters on the rom tiles instead of the ttf.
needed for japanese: sizing the font for kana makes latin narrower, which
knocks the party menu numbers out of line.
A mod index can now publish three optional per-entry fields -- downloads
(total across every release), first_release and last_release (ISO days) --
which the FIND MODS listing shows in the same gold line the MODS tab uses.
The fields are additive by design: feeds that carry them stay readable by
every build that predates them (schema_version stays 1), and feeds that do
not render exactly as before. ModUpdate.statsLine builds the shared line;
the MODS tab reuses it. Parser, formatting, and parse coverage are tested.
The MODS tab now shows each installed mod's total GitHub downloads
(summed asset download_count across all releases), its first and latest
release dates, and a Sort row (Name / Popularity / Release date /
Last updated) persisted in options.modSort.
The launcher already fetched per-repo release lists for update checks, so
the data rides the same cached fetch: parseRelease keeps download_count
and published_at, writeCache persists them, and a cache entry written
before the fields existed is treated as stale and refetched once instead
of hiding the line behind an old cache.
The overlay only wrapped the five loaders the boot path needed, leaving a
silent-failure hole: any future state (or current code like Sound.lua's
widenMono, which re-reads the pika-cry WAV via love.sound.newSoundData
with the caller's bare path) could load a generated asset through an
unwrapped API and silently degrade on hardware.
NxAssetOverlay now wraps every read-side love function that accepts a
filesystem path (filesystem.read/load/lines/newFileData/getInfo,
graphics.newImage/newFont, image.newImageData, audio.newSource,
sound.newSoundData, font.newFontData), so new states and mods fall inside
the Blue/Yellow fallback with zero per-call-site work. Write-side
functions stay stock, proven by identity assertions in the fallback
suite. The static guard's forbidden-literal list covers the same APIs.
Co-authored-by: Cursor <cursoragent@cursor.com>
Three layers, all running without a ROM:
- nx_yellow_boot_test.lua: drives the real Yellow and Blue boot states
(TitleState, YellowIntro + IntroMovie pre-roll for Yellow, IntroMovie
direct for Blue, Sound.playPikaCry) against a broken-mount filesystem
where generated art exists only under yellow|blue/, and asserts no bare
assets/generated path ever reaches the raw love loaders. This is the
runtime complement to the static literal guard: data-driven manifest
paths and formatted paths (cry_%02d.wav) are exactly what a source scan
cannot see. love_stub gains Image:setFilter/getFilter so IntroMovie
constructs headless.
- CI path gate: switch-changes now also triggers on the NX runtime
(NxAssetOverlay, Platform, GameVersion, CacheFs) and the NX engine
suites, so src-side NX regressions rebuild the fused NRO instead of
slipping through with green headless-only checks.
- switch-selftest runs the three NX engine suites headlessly on the
fork-safe ubuntu runner, giving PR feedback before the self-hosted Mac
build. The content gate and switch-build.md docs were updated in sync.
Co-authored-by: Cursor <cursoragent@cursor.com>
The scattered per-call-site prefix rewrites were a parallel track that any
future newImage("assets/generated/...") would silently bypass. Replace
them with NxAssetOverlay: installed once from love.load on NX only, it
wraps newImage / newImageData / newSource / filesystem.read / getInfo so a
missing assets/generated path falls back to the active version's
blue|yellow copy. Call sites return to plain love loader calls, and
Assets.resolve goes back to being the platform-free mod-override point.
Two deliberate exceptions remain: the chip-audio worker (separate Lua
state) keeps receiving the prefix explicitly via audio.programPrefix, and
data/generated module loads keep using CacheFs.readActive.
A new guard test (tests/engine/nx_generated_guard_test.lua) fails CI on
any direct love loader call with a literal assets/generated path, so the
class of bug cannot regress by accident. scripts/test.sh --quick is
green across all tiers.
Co-authored-by: Cursor <cursoragent@cursor.com>
Capture resolve paths and newImage open results for Yellow/Blue art
triage without enabling switch-debug.txt.
Co-authored-by: Cursor <cursoragent@cursor.com>
Unlock love-nx SDL dock/undock resizing and sync via NxDisplay so
booting docked is not stuck on the conf 720p hint.
Co-authored-by: Cursor <cursoragent@cursor.com>
A shared imports/ inbox with Red+Yellow was starting Red from the Yellow tab; match by GameVersion.forSha1 for the selected game and document the tab-scoped rescan.
Co-authored-by: Cursor <cursoragent@cursor.com>
Wire tests/switch_transfer_docs_test.lua into switch-changes detection,
switch-selftest, and the ROM-free T0 lane so docs drift fails CI.
Co-authored-by: Cursor <cursoragent@cursor.com>
Ship stock engine chords only; community mods own their rebinds, and keys
2/3/5 are claimed by the engine before pipeline hotkeys run.
Co-authored-by: Cursor <cursoragent@cursor.com>
platform_nx_* and rom_importer_nx_* run with the love stub and must execute
in CI's ROM-free lane via tests/run_engine.lua, not only T3.
Co-authored-by: Cursor <cursoragent@cursor.com>
Players extract one zip at the microSD root for install and update; saves under pokemon-love2d/ survive merge. Drop the bare .nro from GitHub Release assets.
Co-authored-by: Cursor <cursoragent@cursor.com>
Split Import/Export paths into imports/saves/{red,blue,yellow}/ and
exports/{red,blue,yellow}/ so MTP destinations match each launcher tab.
Co-authored-by: Cursor <cursoragent@cursor.com>
bryanthaboi
MaxTomahawk
AverageConsumer
Andrew Quenehen
ratherDashing
johnjohto
Shane McGovern
Shane McGovern
Max 👨🏽💻 Coplan