edge
7 Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
| 3c179e3338 |
fix(release): ship server/commands, and check that the bundle is complete
v1.0.0 installed and then died on every boot:
module "uo" failed to load — {"stage":"register","reason":"Cannot find
module './commands/guild.command'"}
`server/commands/` arrived with the Teams cutover (
|
|||
| 2fa4d87a40 |
feat(shard): ingest guild rosters and departures (protocol 4)
Protocol 2 gave the guild board a member *count* and nothing else, so the Guilds page could say a guild had 155 members but never who they were, and findGuildForActor deliberately answered only for leaders because membership for rank-and-file was not in the feed at all. Protocol 4 puts it there. `shard_guild_members` holds one row per member per guild, keyed on (guild_id, serial). `guild.roster` replaces a guild's rows; `guild.leave` removes one. A guild.remove now clears the membership too, so a disbanded guild does not leave orphaned rows behind. The chunking needs explaining. A roster over the shard's per-frame cap arrives as several frames carrying seq/more/total. The sidecar reassembles them for its own GET /guilds board, but the live WebSocket feed and the /history backfill both carry the individual frames — so this ingest sees them unreassembled. It copes without buffering, because a table expresses what the sidecar's single JSON column could not: the frame carrying seq 0 clears the guild first, and every frame then upserts its own rows. Upsert rather than insert because the /history backfill replays stored frames on every reconnect, and a redelivery has to be a no-op rather than a duplicate-key error. The cost is a sub-second window during a multi-frame update where the table holds part of a roster; buffering to close it would duplicate the sidecar's reassembly for a projection that is already only as fresh as a 60s sweep. On visibility: both kinds are mapped to the existing `guilds` feature. Without that mapping rule 2 fails an unmapped kind closed to admin-only, which would have quietly kept rosters off the public page forever. Mapping them is safe because a roster is the first frame carrying locked fields inside an ARRAY of actors rather than one nested actor, and the projection walker already recurses into arrays and matches acct/webId by suffix — so a member's account name is stripped below admin by exactly the rule that already strips guild.leader.acct. There is a test for that specifically, because the difference is a public page listing character names versus one publishing 150 account names. `acct`/`web_id` are still stored, since that is what lets a linked member be matched to a site user; they are just never projected below admin. guild.leave is appended to the event log, as the departure counterpart to guild.join and for the same reason — it is what a "so-and-so left" feed reads. guild.roster stays out: it is board state like guild.update, and it is the one fat frame on the wire, so logging it would put a full membership snapshot into shard_events on every membership change. The PUBLIC_KINDS guard test caught the addition, which is what it is for; its expected set now carries a v4 group alongside the v3 one. Refs: docs/website/TEAMS.md Part 12 Phase 1 Co-Authored-By: Claude <noreply@anthropic.com> |
|||
| a0c24456c7 |
ci: retry npm registry reads before failing a check
The frozen-manifest job read ETIMEDOUT from the registry installing the client deps, after it had already cloned core at the pin and proved core's own manifest regenerates — a red X that meant nothing about this PR. There are five `npm ci` calls across the three jobs and the runner is shared, so this will recur. npm's own retry, turned up at the workflow level so every install gets it. Co-Authored-By: Claude <noreply@anthropic.com> |
|||
| 5cdcf0fbb6 |
feat(release): ship an OpenAPI fragment, a frozen manifest and a bundle (phase 3, slice 5)
The three artifacts that make this module installable and checkable, closing
phase 3's extraction. Nothing about what the module serves changes: the same 72
URLs, the same behaviour.
**The OpenAPI fragment (MODULE_API.md §2.8, §6.1a) was never built, on either
side.** The 417 `#swagger` annotations came across in slice 1 and went nowhere,
and core's /api/docs.json merged nothing — so every route this module serves was
in no spec at all, which is core's standing rule ("never ship a route that isn't
in the spec") being broken by the extraction rather than by a route.
`server/scripts/swaggerFragment.js` generates it. The prefixes are DERIVED: the
script runs the module's own `register()` against a recording api and asks
`require.cache` which file each router came from, so a mount prefix exists in one
place — `server/index.js` — and not in a table beside it. The 31 schemas moved
here from core's swagger.js, namespaced `Uo…` because core wins every key
collision in the merge; `Error` and `ValidationError` stay referenced by core's
names, since they resolve in the merged document.
**The frozen route manifest (§5.3)** is derived too, and by subtraction: CI
clones core at the ref pinned in ci/core-ref.json, generates its manifest without
this module and then with it, and the difference is what this module serves. That
buys the half of §5.3 that matters most for free — a module that shadowed or
displaced one of core's routes shows up as a REMOVAL, not merely as an addition
elsewhere. The same job checks the fragment against ground truth: every route
must have an operation and every operation must be a route.
**The release workflow** publishes `module-uo-<version>.tar.gz` plus a manifest
carrying its sha256. The version is declared in module.json rather than computed
from commit subjects, and the workflow never writes to a branch — it tags and
publishes — so `main` needs no push exception. The bundle is assembled from an
include list, because an exclude list ships whatever it forgot.
Four annotation defects, inherited from core and never visible until something
generated a spec from these files: two `requestBody` literals a brace short (the
route documented with an empty body), and two descriptions whose inner quoting
swagger-autogen cannot survive — it re-quotes `"` and a backtick to `'` before
evaluating, so either inside a single-quoted description ends the string early
and the annotation is dropped. It reports each one and then prints Success in
green, so the generator now captures its diagnostics and makes them fatal.
Also fixed while writing it: passing one shared `doc` to swagger-autogen six
times. It renders components.schemas from an EXAMPLE object and writes the result
back into what it was handed, so each pass re-wrapped the last and the fragment
came out at 484 MB.
- 409 server tests (+21), 40 client tests unchanged
- swagger-fragment.json: 69 paths covering all 72 routes
- routes.manifest.json: 72 routes; core's own surface unchanged, 0 removals
- verified end to end by assembling the bundle exactly as CI will, unpacking it
into a real core and regenerating the manifest
Refs: docs/website/MODULE_SYSTEM.md §2.7.1, MODULE_API.md §2.8, §5.3, §6.1a
Co-Authored-By: Claude <noreply@anthropic.com>
|
|||
| e4af7dd9a8 |
test(client): check what the chunk registers, and fix two defects in the checks
Three things, all about checks that had never met a real chunk.
`checkExternals.js` rejected slice 3's build outright, naming a fragment of
minified JSX as an imported specifier: a button reading "Approve and import"
puts the token immediately before a quote, and no regexp can tell that from a
statement. Same wall the server's `checkImports.js` hit, answered the same way —
a character walk. A mask rather than a rewrite, because a real import has its
keyword outside a string and its specifier inside one.
Writing the test for that false positive found the false NEGATIVE underneath
it: the pattern required whitespace after `import`, so it could not see
`import{useState}from"react"` — the one shape a minified build actually emits,
and the most likely way for a missed alias to reach production. It has never
been able to see it.
`registration.test.js` is new: stand up a fake `window.__rg` with a recording
registry and the real React, import the BUILT chunk, and read back what it
asked for. No DOM, because nothing renders. It holds the agreement that rots
quietly — every nav row points at a route this module actually registered —
rather than restating both lists.
CI now builds before it tests, because both of those read `dist/entry.js` and
skip without it. Run the other way round they are green and asking nothing.
Co-Authored-By: Claude <noreply@anthropic.com>
|
|||
| 5d7668d5ea |
feat(module): the bundle skeleton (phase 3, slice 0)
The first real module. It registers nothing, deliberately: what slice 0 proves
is the delivery path itself, end to end, before a single UO file moves into it.
Server half: module.json, an entry point that takes (ctx, api) and registers
nothing, a test suite built on a fake ctx, and scripts/checkImports.js -- the
MODULE_API.md §5.1 boundary check. Client half: the Vite library build, four
shims re-exporting react / react-dom/client / react-router-dom / jsx-runtime
from window.__rg, an entry that verifies each is identity-equal to core's copy,
and scripts/checkExternals.js. 29 server tests, 9 client tests, both new.
Verified against a real core: the module loads, mounts its zero routes, runs to
`started`, and is published by /api/v1/public/modules. Its chunk serves from
the entry's directory with `Cache-Control: no-cache` while the module's server
source, module.json and package.json all 404. In Chrome, under the enforced
`script-src 'self'`, the chunk evaluates and reports all four shared
dependencies OK, with zero CSP reports and no console errors.
Three findings, each of which had produced a green build that was wrong.
MODULE_API.md §3.6 shows `external` alongside the aliases and they do not
compose. Rollup asks `external` BEFORE Vite's alias resolver runs, so a
specifier in both is marked external and never aliased -- the chunk then ships
bare `import "react"`, which no browser can resolve without an import map, and
CSP forbids one. Built cleanly and emitted exactly that; checkExternals caught
it. So: alias only, `external` empty, and vite.config.js grows a resolution-time
guard that fails the build if a shared dependency resolves into node_modules.
That guard was wrong twice before it worked. Written against Rollup's `load`
hook it never ran -- `load` is first-wins and an earlier plugin had already
claimed the module -- so a deliberately-broken alias produced a 24 kB chunk with
react-router welded in, and a green build. And its forbidden-package list was
derived from the alias list "so the two cannot disagree", which meant deleting
an alias also deleted the guard against what that alias prevented. It states the
contract now, and a test asserts the aliases stay inside it.
checkImports failed on its own documentation the first time it ran: the comment
naming require("../../etc/passwd") as an example of what to catch, and index.js
explaining why the module must never require("express"). A boundary check that
cannot survive being described is one people stop writing comments around. It
strips comments and template literals with a character walk rather than a
regexp, because a URL in a string contains a comment opener and a comment
contains quotes -- and it has its own test suite, since a check never shown to
fail is a check nobody knows the state of.
Co-Authored-By: Claude <noreply@anthropic.com>
|
|||
| 4a0d8f0873 |
ci(module-uo): gate pull requests into main on server and client checks
Mirrors RunicGateway/website's pr-checks.yml, since this module is two npm packages shaped like that repo's server/ and client/ and is loaded into that repo's process: same Node 20, same npm ci + test + build, same concurrency cancel and the same `PR Checks / *` status pattern for branch protection. The repo has no module code yet — the API contract is settled in Phase 1 and Phase 3 is what extracts the UO half of website/ into here — so each half's gates are conditional on its package.json existing. Before the code lands the jobs report green with a notice explaining why; the moment a package.json appears they arm themselves with no edit to this file. That is the same guard the installer repo used through its own planning phase, and it beats both alternatives: leaving the repo ungated, or red-Xing every governance PR. Landing it now also gets the status contexts reported once, which is what makes them selectable when the `main` branch-protection rule is configured. Deliberately not here yet, because there is nothing for them to act on until Phase 3: the release workflow that publishes module-uo-<version>.tar.gz and its sha256, the zero-internal-imports check, and the module's own frozen route manifest. Each lands with the code it checks. Co-Authored-By: Claude <noreply@anthropic.com> |
