api Mod Controller Compatibility Info and Required Disclosures system

[budak7273]

Backend side of satisfactorymodding/smr-frontend#255

• General fixes to get development working from Windows, updates to mise and documentation
• Prevent reusing semver of deleted mod versions (Prevent deleted mod version semver reuse #134)

---

• Controller compatibility info
  • Can be set back to Unknown unlike EA/EXP compatibility info
• Unused ToggleNetworkUse removed
• Required Disclosures system (this+AI Disclosures #132)
  • Network Activity Transparency is now a dedicated field instead of to be put in mod descriptions
  • New AI usage disclosure dedicated field
  • New Mod flow intentionally does not allow specifying these disclosures, they can be filled out later once the author has actual answers to them
  • New mod version uploads are blocked until disclosures have been provided
  • Existing mods have no disclosure set

[codecov]

Codecov Report

❌ Patch coverage is 80.28169% with 14 lines in your changes missing coverage. Please review.
✅ Project coverage is 42.23%. Comparing base (7b96fac) to head (ce51290).
✅ All tests successful. No failed tests found.

Files with missing lines	Patch %	Lines
gql/resolver_versions.go	0.00%	3 Missing and 3 partials ⚠️
gql/gql_types.go	88.23%	1 Missing and 1 partial ⚠️
gql/gql_utils.go	50.00%	1 Missing and 1 partial ⚠️
gql/resolver_mods.go	86.66%	1 Missing and 1 partial ⚠️
workflows/versionupload/extract_mod_info.go	81.81%	1 Missing and 1 partial ⚠️

Additional details and impacted files

@@             Coverage Diff             @@
##           staging     #138      +/-   ##
===========================================
+ Coverage    41.73%   42.23%   +0.50%     
===========================================
  Files          128      128              
  Lines         6146     6203      +57     
===========================================
+ Hits          2565     2620      +55     
+ Misses        3191     3185       -6     
- Partials       390      398       +8     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[Vilsol]

@claude review

[Vilsol]

(Bot didn't work when I tagged it, so I had Claude Code run it, here is the result of that:)

Did a thorough pass over the hand-written changes (excluding generated/ and mise.lock). Most of the PR looks great — the disclosure plumbing is well-placed and the new gating correctly rides the already-fetched mod row instead of adding a DB round-trip. A few things to address before merging though, plus some cleanup suggestions.

🔴 Ship-blockers

1. Nil-pointer panic in UpdateMod AI disclosure handling

gql/resolver_mods.go:197

if requiresMessage(aiDisclosureUpdate.DisclosureType) && *aiDisclosureUpdate.DisclosureString == "" {

schemas/mod.graphql declares disclosure_string: String (nullable), so aiDisclosureUpdate.DisclosureString is *string. A client sending {disclosure_type: ai_usage} without a disclosure_string will crash the resolver on the deref.

// suggested
if requiresMessage(aiDisclosureUpdate.DisclosureType) &&
    (aiDisclosureUpdate.DisclosureString == nil || *aiDisclosureUpdate.DisclosureString == "") {

2. requiresMessage disagrees with the version-upload gate

gql/resolver_mods.go:162-164 says a message is required only for ai_usage / runtime_ai_usage.
gql/resolver_versions.go:54-60 treats anything != NoAiUsage (i.e. including no_disclosure) as "needs a message" before unblocking uploads.

So no_disclosure is a valid persisted state from UpdateMod's perspective but a blocker from CreateVersion's perspective. Pick one definition of "set" and centralize it — currently each resolver answers a slightly different question.

3. testza.AssertNoError uses token as the failure message

tests/version_test.go:112

token, _, err = makeUser(ctx)
testza.AssertNoError(t, err, token)

token is being passed as msg ...interface{}. If makeUser fails, the test failure will print the (likely empty) token rather than something useful. Drop the third arg.

🟠 Concrete cleanup

4. Controller-default migration does two UPDATEs on the same rows

migrations/sql/20260501041455_add_default_controller_compatibility.up.sql

The first statement writes {"Controller":""}, the second overwrites it with {"Note":"","State":"Untested"}. Every matching row is rewritten twice (2× WAL, 2× heap update, 2× lock window) and the intermediate Controller: "" is semantically an invalid Compatibility. Collapses to one statement:

UPDATE "mods"
  SET "compatibility" = jsonb_set(compatibility, '{Controller}',
                                   '{"Note":"","State":"Untested"}'::jsonb)
  WHERE NOT compatibility ? 'Controller' AND compatibility ? 'EA';

5. util.AIUseDisclosureInfo is stringly-typed

util/db.go:30-33

type AIUseDisclosureInfo struct {
    DisclosureType   string
    DisclosureString string
}

Forces .String() / raw-string comparisons at gql/resolver_versions.go:58 (!= generated.AIUseDisclosureTypeNoAiUsage.String()) and gql/resolver_mods.go:201. The existing util.Stability (top of the same file) is the established pattern for a typed enum-string:

type AIUseDisclosureType string
const ( AIUseNoDisclosure  AIUseDisclosureType = "no_disclosure" /* ... */ )

Drops every .String() dance and lets go vet catch typos.

6. GenControllerCompToDBControllerComp duplicates GenCompToDBComp

gql/gql_types.go:27 — same util.Compatibility{State, Note} build from a near-identical generated input. Either generalize with generics over a small interface{ GetState() string; GetNote() *string }, or share one input type in the schema (controller-specific enum on a separate field).

7. Two Count() queries in the extract activity

workflows/versionupload/extract_mod_info.go:54-77 — the new soft-deleted check sits right next to the existing live check on the same (mod_id, version) key. Two Postgres round-trips back-to-back. Both are presence-only — Exist(ctx) is cheaper than Count(ctx) (SELECT 1 LIMIT 1 vs aggregate), or merge into a single SkipSoftDelete query partitioned in Go.

8. Duplicated default-disclosure test setup

tests/version_test.go:156 and tests/sml_versions_test.go:77 — near-verbatim mutation block setting AiUseDisclosure: NoAiUsage so uploads aren't blocked. Belongs as a helper in tests/utils.go next to makeUser — any future upload test will need it too.

9. network_use_disclosure is the only .Nillable() optional string in db/schema/mod.go

db/schema/mod.go:43 — every other optional string field uses bare .Optional(). The choice looks intentional (the gating needs *string to distinguish "unset" from "explicitly empty"), but it silently introduces a new convention. Worth a comment on the field. It also creates an asymmetry with ai_use_disclosure (a field.JSON(...).Optional(), not Nillable) that happens to behave the same way only because JSON fields default to nil pointer.

🟡 Style / next time

10. Migration filename outlier

20250411023207_AddNetworkUsageDisclosure.up.sql is the only migration in the repo using PascalCase, and its timestamp is from 2025-04 (a year behind the other three migrations in this PR). Looks carried over from an earlier branch. Renaming now is high-friction (atlas.sum re-hash, anyone with the migration already applied needs a manual fix-up), so feel free to leave — but a next time note for the snake_case + current-date convention.

11. AI + Network disclosure update branches have parallel control flow

gql/resolver_mods.go:195-211 — "if set: validate-and-apply; if cleared while previously set: error" is duplicated across both fields. Could be a SetOmittableNoClear[T] helper next to the existing SetINNF family in gql/gql_utils.go.

✅ Verified non-issues

• New disclosure gating in CreateVersion does not add a DB round-trip — reads off the mod row the resolver already loaded.
• No missing indexes — disclosure columns are only read via Mod.Get(id), never filtered/sorted on.
• Windows .ps1 vs Linux .sh mise tasks: duplication is acceptable given genuine shell semantics differences.
• Controller unknown state and Compatibility re-use: correctly threaded through util.Compatibility.

[budak7273]

Addressed 1, 3, 4, 6, 7, 8, 9, 10

Sorta addressed 2

Removed the Unspecified enum option, now covered by the field being unset

Skipped 5

Not sure how to make this work with the gorm:"type:string" json:"DisclosureType" directive present too

Skipped 11

Enough purpose built logic where trying to extract would be more work than it's worth

[Vilsol]

Was something broken why this had to be done?

[Vilsol]

This should not be pinned to "latest"

[Vilsol]

Is this file needed?