31da86980051ca6b0c515ae1c5ea8d7deb24c33d
3
Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
  mr.zeroandClaude Opus 5
|
6285d93790 |
Stores can be removed, and added from an address you type
Two gaps that were the same gap: the store list could only ever grow, and it could only grow from what the registry happened to offer. **Removing** uninstalls what the store installed, then deletes the store itself, in that order. The order is the whole of it: `state.json` is the only record of which payloads, icons and menu entries belong to a store, so deleting the home first would strip the one thing that knows — leaving files nothing could ever identify, least of all a later install of the same store into the same folder. The confirmation says how many titles will go, because that is the part nobody would otherwise expect. The token goes too; a credential for a store that is not here is a secret kept for nothing. The window names a *store*, never a path: the home is resolved against what a disk scan actually found before anything is deleted, and `removeHome` refuses anything else. That is the only guard between a bad argument and `rm -rf`, so it has a test. **Adding** moved to a + beside Refresh — both are actions on the whole store rather than on one of them, and the full-width button under the list read as a third store — and the picker now takes a catalog address as well as a listed one. A bare host is enough and the name comes from the address; nothing else about installing changes, which is why the typed path hands the same record to the same method instead of growing a second one. The picker also has a Cancel now: opening it with a store installed used to replace the grid with no way back. `make storetest` is new, and it earned itself immediately. Removal is the only code here that deletes a directory tree, which the smoke test cannot cover — it runs against the real machine and would have to delete a real store to prove anything. Two bugs on the first run: - `http://` was accepted and became a store called *http*. The trailing slashes were stripped before the scheme was checked, turning `http://` into `http:` and then into `https://http:`, whose hostname parses as "http". The URL is rebuilt from the parsed form now, which also settles the trailing slash in one place. - `STORE_ROOT` only *prepended* to the search path, so a "sandboxed" run still listed the real stores — despite the README saying "instead of the real one". Harmless while a sandbox could only add; not harmless now that it can delete. It replaces the search path. The self-test needed two changes, both of which are it working: the store row is a wrapper now, so clicking `.store-row` did nothing at all, and the footer icon check counted exactly two named controls when there are three. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> |
||
|
mr.zeroandClaude Opus 5
|
526c67b069 |
Registry-only stores, a quieter window, and a CI that builds
ci/woodpecker/manual/woodpecker Pipeline was successful
A store no longer needs a repository of its own. The engine's built-in defaults already cover the host-to-asset mapping, the install modes, the platforms and the behaviour; what they cannot know is identity — a slug, a name and a catalog URL — and that is exactly what a registry record carries. So `storeRepositoryUrl` is optional: a record with a name and a catalog is a complete store, the id falls back from the repository name to the catalog host (`teletypegames.org` becomes `teletypegames`) to the display name, and the client writes a three-section config. Given a repository it still reads it, and that file stays the authority on how the store behaves; a repository without a config.json is treated as no repository at all. Measured end to end against a local registry serving one record with a null repository: the engine and the core downloaded, engine 1.1.0 accepted the written config, it listed the same ten titles the configured store does, and a hosted title synced into a sandbox with its menu entry written. The "Install all" button is gone, and with it the string it used. Titles are installed one at a time from their own cards. No footer. The window carried a bar at the bottom at all times — a toggle and a line of absolute paths — for something most sessions never need. The log is still there, folder buttons included, behind a quiet switch at the bottom of the side menu; it takes no room until it is opened, and an arriving line does not open it, because the store logs on every refresh and a window that unfolds panels by itself is worse than one that keeps quiet. Three faults that every automated count had passed, found by photographing the setup screen: the store badge rendered as an empty pill with no store open; the gate's picker showed as an empty dropdown stub, because an explicit `display` beats the browser's own `[hidden]` rule; and the gate went up while the empty-catalog line stayed on screen underneath it. The last was a design fault — whether the gate is up was a call on a view rather than state, so the two could disagree. The setup screen is now a field in the state store, and that one field decides which of the gate and the grid is drawn. The window test's gate assertion was wrong too: it demanded a store picker, which only appears when the registry offers more than one store, so one store — the ordinary case — failed it. CI builds the packages this machine cannot. `.woodpecker.yaml` runs the checks on every push and, on a tag or by hand, builds the Linux packages in `electronuserland/builder:22` and the Windows ones in `:22-wine`, then attaches them to the release with scripts/ci-upload.sh. The pipeline lives here rather than in the update server's `/build/config` extension, which serves game-platform pipelines publishing into the site's catalog — a different product with a different target. macOS stays a local build: Apple's toolchain and its signing exist only on a Mac. Both build steps verify what they produced, because a half-finished Wine build leaves a 162 KB stub named like the real installer and `ls` is happy with it. The Linux step was rehearsed locally in the same image (AppImage 128 MB, deb 100 MB); the Wine step cannot be rehearsed on Apple Silicon, where 16 KB host pages break Wine's 4 KB assumption, so the runner is where it is proven. The size check was tested against both outcomes. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> |
||
|
mr.zeroandClaude Opus 5
|
3d63c8a0b0 |
TypeScript, in layers, with a strict linter
The client was one main.js, one preload.js, three files in lib/ and one renderer
script. It is now a typed application whose imports point inward: domain (models,
ports, errors) knows nothing about Electron, Node or Python; application orchestrates
it through those ports; infrastructure holds the adapters — the Python CLI, HTTP, the
filesystem, Electron itself — and main, preload and renderer sit on top as hosts.
STRUCTURE.md is the map, and the deliverable as much as the code is: every layer, every
pattern in use (ports and adapters, repository vs gateway, service, DTO and mapper,
composition root, controller and router, single flight, observer streams, state store
with unidirectional flow, passive view, coded error hierarchy, frozen constant tables,
untrusted-data readers) and the naming rules — files, classes, and a verb vocabulary
for methods where find/require/read/list/apply/render/handle each state a contract.
Two properties fell out of the move, and they are why it was worth doing:
- The catalog can be driven with no window and no Electron at all. The smoke test
assembles the same services against the same ports in a plain Node process; it used
to be a script that reimplemented the bridge.
- The window never receives a filesystem path. A title crosses the bridge without
one, and launching is asked for by name, resolved in the main process from the
store's own state. Verified with a fake launcher: an unknown name answers false, a
native title resolves to its menu entry, a hosted one to its catalog URL.
Types are mandatory, including where inference would manage: explicit return,
parameter and property types, strict plus noUncheckedIndexedAccess,
exactOptionalPropertyTypes, noImplicitOverride and noPropertyAccessFromIndexSignature,
typescript-eslint strictTypeChecked and stylisticTypeChecked, exhaustive switches, no
any, no non-null assertions, and no casts on foreign data — engine stdout and the
registry go through readers that turn unknown into typed values. naming-convention
enforces the patterns rather than trusting them.
Two rule conflicts had to be decided rather than papered over. typedef and
no-inferrable-types disagree about `fallback: string = ''`: the annotation wins, since a
signature states its types. erasableSyntaxOnly is off, because it forbids constructor
parameter properties, which are how dependencies are declared here.
The preload and the renderer are bundled by esbuild into one file each: a sandboxed
preload may not require its own modules, and a module script over file:// is blocked by
the page's own origin rules. tsc compiles the rest. The package ships build/** and
package.json — 111 entries, no sources, no toolchain.
New targets: build, typecheck, lint, lint-fix, and check — typecheck, lint, then both
test suites, cheapest failure first. Every script that runs the app builds first, so a
stale bundle cannot be tested.
Nothing about the window changed: same side menu, same categories, same switcher, same
two languages. make check is clean, both test suites pass with one store and with two,
the packaged 1.3.0 bundle drives the real store, and the window was photographed before
and after — the two are the same picture.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
|