The desktop graphical client had our store's config URL compiled into it. That is
backwards: which stores exist for a catalog is something the site knows, and a
client should be able to ask. `GET /api/stores` answers, and the client picks from
what comes back.
A `Store` row is three fields — name, catalog_url, store_repository_url — with an
ActiveAdmin panel, a Blueprinter serializer in the catalog's camelCase, and a
service+controller pair following the events and members shape. `db/seeds.rb`
creates our own record, so a fresh database serves a working registry.
The endpoint is **public**, which is the point: the client runs on someone's
laptop before any store exists and has nobody to log in as. It exposes three URLs
that are public anyway.
This deliberately does not live in WarpEngine. The engine serves one catalog and
has no business knowing who ships stores for it; a registry of stores is a
property of this site, not of the catalog software. Anyone mounting WarpEngine can
keep their own list, or none.
16 model and controller examples pass. Worth noting for the next person who runs
them: the specs need RAILS_ENV=test, as the README says — without it rspec runs in
the development environment, host authorization rejects Rack::Test's hostname, and
every request spec fails with a 403 and an HTML body that looks nothing like a
routing problem.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The desktop store now has a window, and it is the only one of the three that
does — so the download link and the sentence explaining it appear on that tab
alone. On Windows this is the way in, since the shell installer needs a shell
the platform does not have.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The page had two device tabs, both for hardware most visitors do not own. The
desktop store makes the third one the likeliest answer, so "This computer" is
added and selected by default, and /desktop redirects to it the way /batocera and
/retroarch already do.
Device names now come from i18n rather than the component: "Batocera" and
"RetroArch" are product names either way, but "This computer" has to be
translatable. Every key the page uses was checked to exist in both locales.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The Batocera store had /batocera to itself. A RetroArch store now serves the same
cartridges on every other machine, and copying the page would have duplicated
everything the two have in common — the framing, the platform table, the engine
links — so they share one page and a Batocera/RetroArch chooser. Only the install
command, the CLI and the uninstall line change with the tab.
/batocera and /retroarch both redirect here with the matching tab preselected,
so old links keep working and the name someone guesses after reading "RetroArch
store" lands somewhere useful. `?device=` makes a link to either half shareable.
The page also documents uninstalling, which it never did, and the five command
blocks share one CommandBlock component instead of repeating the copy button.
The i18n `batocera` block becomes `stores` in both locales; every key the page
uses was checked to exist in both.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The engine's 180 specs were never run by CI: the mirror repo is
generated by a subtree split, so a pipeline committed there would be
overwritten on the next sync, and the monorepo's own pipeline only
mirrored and published. A broken engine could reach the gem registry.
The test now gates both — it runs on the same trigger the mirror does
(a change under libs/ruby/warp_engine), and the steps after it only run
if it passes. The specs need MySQL, hence the service, and a prepared
test database: on a fresh database maintain_test_schema! reports the
engine's own migrations as pending instead of loading the schema.
Verified by running the step as written in a clean ruby:3.2 container
against a fresh mysql:8 — 180 examples, 0 failures.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The step assumed one project shape: plain sources concatenated with cat
and Phaser pulled from a CDN. trickster-tiles is the other shape — Vite
plus TypeScript, built with `tsc && vite build` — so `cat src/*.js`
found nothing to bundle and the build failed. The earlier guard on the
syntax check fixed only the first symptom of that mismatch.
A project with a build script now runs npm ci && npm run build and is
packaged from its own dist/, which already contains index.html and the
hashed assets. Plain projects keep the existing path untouched.
The bundler's base has to be relative, since games are served out of
/file/<name>-<version>/; the step fails loudly if the build leaves no
dist/index.html rather than shipping an empty zip.
Verified in the phaser-builder image against trickster-tiles: a 344 kB
package with index.html at the root and ./assets/ references.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The org reorganization moved the toolchain repos to build/ but left
their container images in internal/: a package namespace does not
travel with the repo and gets no redirect, which is the only reason
the internal org was still alive.
All eight images now live under build/ — the six unchanged ones copied
layer-for-layer, the Ebitengine and Bevy ones rebuilt for the ARM
target. The internal org can be emptied.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Batocera and the ES-family distributions run on ARM as much as on
x86_64 — Raspberry Pi, Odroid, the retro handhelds — and a linux_x64
binary installs there but will not start. There was no Linux ARM asset
kind at all: KINDS had linux_x86 and linux_x64 and mac_arm64, but
nothing for 64-bit ARM Linux.
Registering the kind is deliberately separate from producing it: a
platform service only includes BuildLinuxArm64 once its pipeline builds
the artifact, otherwise /api/builds would report it missing for every
release. Hence Ebitengine and Bevy only. Godot needs a Linux arm64
export preset in each game repo first; LÖVE fuses an upstream AppImage
that ships x86_64 only; TIC-80's export command has no ARM target.
Ebitengine needs cgo on Linux, so binary_build gained a cross-compiler
argument — and unsets CC for native targets, otherwise a build after
the ARM one silently picks up the cross gcc.
Verified by cross-compiling both demos in the rebuilt images: each
produced a genuine AArch64 ELF (e_machine 183), not a silent fallback.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The push prompted for credentials and then died with "404 page not
found". Both are the same fault: RubyGems normalizes the keys in
~/.gem/credentials — dots become __ and a trailing slash is appended —
so a key written as the host URL can never match the --host value.
Finding no key, gem push falls back to signing in against the RubyGems
sign_in endpoint, which Gitea does not implement; that is the 404.
A named key is stored verbatim and is matched by --key, so the lookup
succeeds. Verified against Gem::ConfigFile with both formats.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The publishing pipeline was already complete — subtree split to the
mirror, gem build and push on a warp_engine-v* tag — but it had never
been triggered, and after the org reorganization three of its targets
were stale: the mirror push URL and the rubygems registry namespace
(twice).
The gemspec's allowed_push_host has to match the --host that `gem push`
receives, so it now carries the full registry URL rather than the bare
forge host, plus source and documentation links.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The client was split into a reusable engine and a store definition, so
the page is now about our store: new repo and wiki links, the CLI path
the installer actually writes, and a closing section pointing at
warp-engine-batocera-store for anyone who wants a store of their own.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The engine ships an ActiveJob based job (PipelineSyncJob), but a host's
application.rb does not necessarily require active_job/railtie - apps/api does
not. With eager loading off (development, test) nothing noticed; in production
WarpEngine::ApplicationJob blew up with "uninitialized constant
WarpEngine::ActiveJob", which is exactly what `rails zeitwerk:check` in
RAILS_ENV=production reported. An engine that ships jobs has to pull in the
framework it needs, so lib/warp_engine.rb requires the railtie.
Pre-existing on 0.1.0 as well; found while verifying that the 0.2.0 changes do
not break the portal. All three api-test steps are green now.
apps/api Gemfile.lock follows the 0.1.0 -> 0.2.0 path gem bump.
Two seams the hosts needed, both backward compatible.
Storage: artifacts are served through WarpEngine::Storage.adapter instead of
raw filesystem calls. The default :local adapter keeps the previous behaviour
byte for byte, including the path traversal guard. A host can now set
config.storage_adapter to any object answering file?/directory?/locate and
serve builds from an object store - FileService and /api/download both honour
a Location.redirect, so a signing adapter turns them into redirects.
DownloadService#create still returns an absolute path (nil when missing) for
existing callers; #locate is the new entry point that can also return a
redirect. Ingestion (upload, extraction, file manager) stays local for now.
Publish: PublishService emits ActiveSupport::Notifications
("warp_engine.publish") with platform/name/version/software/release, so hosts
can react to a new build without hanging callbacks on the models.
WarpEngine.instruments_publish? lets a host feature-detect and keep its
fallback for older engine versions.
mr.zeroandClaude Opus 5