This repository was archived by the owner on Aug 6, 2026. It is now read-only.
feat: game-declared registration mode via a config.lua global - #132
Merged
Conversation
Registration posture was only reachable through the release sys.config, so every engine-hosted game silently ran the `open` default. Games can now declare `registration = "open" | "oauth_only" | "closed"` in the same config script that carries `guest_auth`, applied by asobi_lua_config:apply_registration_mode/1 at boot. Omitting the global leaves the app env untouched so a self-hoster's sys.config still wins; an unrecognised value is logged and ignored rather than downgrading the posture to `open`.
This was referenced Aug 4, 2026
Taure
added a commit
to widgrensit/asobi
that referenced
this pull request
Aug 4, 2026
…pshot (#353) * fix: atomize spawn-template base_state and zone.spawn override keys Ported from widgrensit/asobi_lua#131 (af391ff), which landed after the asobi_lua snapshot merged into this repo. Entities created by game.zone.spawn kept the binary key shape Luerl decodes, because asobi_zone_spawner merges a template's base_state (and any override table) straight into the new entity while only x/y/type are stamped as atoms. Every atom-keyed consumer in asobi_zone then read the wrong shape - snapshot_entities' `persistent` check, the crossing clauses, asobi_spatial - until some later zone_tick happened to round-trip the entity through atomize_entities/1 and fix it up. Convert base_state and override keys at the same boundary #270 uses for tick results, so a spawned entity has the same key shape whether or not a tick has elapsed since it was created. * fix: validate game.broadcast event names for immediate script feedback Ported from widgrensit/asobi_lua#134 (1bc28be), which landed after the asobi_lua snapshot merged into this repo. #303 already rejects an out-of-shape broadcast event name at the socket boundary, so a script that trips one of those rules lost the event with only a server-side log line to explain why. game.broadcast now asks the same guard before the fan-out and returns `{ error = "..." }` at the call site. Reworked from the asobi_lua original: in a single repo the two halves can share an owner, so asobi_ws_handler exports event_name_binary/1 and asobi_lua_api calls it instead of restating the length, charset and reserved-name rules. That removes the "both lists must move together" hazard the split-repo version had to carry. * test: real-Postgres CT coverage for the game.storage.* write path Ported from widgrensit/asobi_lua#133 (3354e5e), which landed after the asobi_lua snapshot merged into this repo. asobi_lua_api_tests mocks asobi_repo for every game.storage.* case, so the suite could only ever confirm the assumption baked into the mock. #296 (update_all/2 rejecting a raw jsonb map, second set silently dropped) shipped past exactly that. asobi_lua_storage_SUITE drives game.storage.set/get/player_set/player_get through Luerl against the Docker Postgres, asserting both the Lua-visible result and the row in the database: set -> set -> get, version bump, and both directions of the global vs player-scoped namespace isolation. Verified the suite fails against both defect shapes: dropping the `player_id IS NULL` predicate from storage_find/3 fails global_write_does_not_clobber_player_row, and restoring the unscoped update_all/2 write fails set_twice_updates_the_row plus player_write_does_not_clobber_global_row. Setup uses asobi_test_helpers:start/1 rather than the standalone pgo/kura_migrator boot the asobi_lua original needed, and ci.yml already passes extra-services-compose, so no CI change was required. * feat: game-declared registration mode via a config.lua global Ported from widgrensit/asobi_lua#132 (4b4bf8e), which landed after the asobi_lua snapshot merged into this repo. Registration posture was only reachable through the release sys.config, which no engine-hosted game can edit, so every hosted game silently ran the `open` default. A game can now declare `registration = "open" | "oauth_only" | "closed"` in the same config script that carries `guest_auth`. Reworked for ADR 0006. The asobi_lua original called application:set_env(asobi, registration, Mode) from the loader, which in this repo would be both an unowned writer and the trust inversion ADR 0006 exists to prevent - a game bundle overwriting operator config. So registration becomes a second layer exactly like the mode registry: asobi_game_config writes the declared value to `script_registration`, never to the operator's `registration` key, and asobi_registration composes the two with the operator layer on top. A bundle can pick a posture for a deployment that states none, and can never widen one that does. Omitting the global leaves both keys alone, so a self-hoster's sys.config still wins; an unrecognised value is logged and dropped by the loader rather than downgrading the posture to `open`.
Taure
added a commit
that referenced
this pull request
Aug 4, 2026
v0.51.4 predated widgrensit/asobi#353, which ported the four commits that landed in this repo after the merge snapshot was taken. Publishing the alias against v0.51.4 would have shipped an image missing asobi_lua#131, #132, #133 and #134 - exactly the regression this retirement was held back to avoid. Verified in the assembled prod release: asobi-0.53.0 is the only asobi lib, each shared module ships exactly once, and the three code fixes are present in the fetched dependency source.
Taure
added a commit
that referenced
this pull request
Aug 4, 2026
* chore: retire asobi_lua to a Docker alias for asobi The Lua runtime moved into asobi (widgrensit/asobi#339). asobi main now defines 16 of this repo's 17 modules, so any asobi pin at or past that merge makes the release unbuildable: ===> Error generating release: Duplicated modules: asobi_lua_game_error specified in asobi and asobi_lua asobi_lua_dev_errors specified in asobi and asobi_lua Urgent part: .github/workflows/bump-asobi-pin.yml ran daily and moved the pin to asobi main unattended. The next run would have hit exactly that error and left ghcr.io/widgrensit/asobi_lua unpublishable. It is deleted, not paused. The published image keeps building. It is now an alias: same name, same tags, same bin/asobi_lua entrypoint, same port, same environment, but the release is assembled from asobi alone, whose own supervision tree starts the Lua runtime. Nothing changes for anyone running it. - Delete src/, include/ and test/ - all duplicated by asobi. Only asobi_lua_app.erl was unique, and asobi_sup already does its job. - Pin asobi to the v0.51.4 tag rather than a branch, since nothing bumps it automatically any more. - Replace the erlang-ci delegation with a release build plus an assertion that no Lua module ships twice. Delete the nightly property soak; its suites moved. - Close the tracker to new issues via .github/ISSUE_TEMPLATE/config.yml redirecting to widgrensit/asobi. Existing issues stay open and readable. - Deprecation notice at the top of README, plus AGENTS.md and SECURITY.md. * chore(deps): pin asobi v0.53.0, the release carrying the ported fixes v0.51.4 predated widgrensit/asobi#353, which ported the four commits that landed in this repo after the merge snapshot was taken. Publishing the alias against v0.51.4 would have shipped an image missing asobi_lua#131, #132, #133 and #134 - exactly the regression this retirement was held back to avoid. Verified in the assembled prod release: asobi-0.53.0 is the only asobi lib, each shared module ships exactly once, and the three code fixes are present in the fetched dependency source.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Closes #122
Registration posture (
open/oauth_only/closed) was only settable through the release'ssys.config, which no engine-hosted game can edit - so every hosted(game, env)silently ran theopendefault. This follows the ADR 0004 precedent used forguest_auth: make it a game-declared Lua global read throughasobi_lua_config, so self-hosters and engine-hosted games get the same control with zeroasobi_enginediff.What changed
config.luain multi-mode, elsematch.lua):registration = "open" | "oauth_only" | "closed"asobi_lua_config:apply_registration_mode/1, called frommaybe_load_game_config/0alongsideapply_guest_auth/1and exported soasobi_engine's bundle loader can reuse it.apply_guest_auth/1and the new function now share aconfig_script_state/1helper instead of each open-coding the script lookup.Semantics
sys.config(or asobi'sopendefault) still wins - the Lua global is additive, not an override that resets on every boot.open.reload_game_modes/0(the config watcher path) does not touch it, matching the guest-auth rule that a live bundle write cannot flip auth posture.Tests
Four new cases in
asobi_lua_config_tests: single-mode sets the mode, multi-mode reads it fromconfig.lua, absent global keeps the operator's app env, invalid value keeps the configured mode.rebar3 fmt --check,xref,dialyzer,eunitgreen. The oneasobi_lua_world_integration_testsfailure (badkey solidin the spawn-templates test) reproduces unchanged onorigin/mainand is unrelated to this branch.