PolicyPlugin: required to-one relations read back null via include/select, but ModelResult types them non-null
Maintainers usually reply within 1 day
Nobody has claimed this yet.
Assessment
- Difficulty
- 3/5
- Estimated time
- 1-2 days
- Newbie friendliness
- 48/100
- Issue type
- Bug
- Clarity
- Clearly specified
- Activity status
- Quiet
- Tech stack
- typescript
- Domain
- backend
Research direction
Read packages/orm/src/client/crud-types.ts around ModelResult's include branch and ModelSelectResult, plus WrapType and ModelFieldIsOptional, to see how the Optional flag reaches the relation type. Run tests/e2e/orm/policy/migrated/nested-to-one.test.ts ('read rejection for non-optional relation') to confirm the runtime null, then add a type-level assertion that m2.m1 must be nullable. Decide the gating mechanism (ExtResult already threads through ModelResult) and verify non-policy results stay unchanged. Done means required to-one relations are nullable only where the policy plugin can null them.
Written by the indexing model from the issue text.
Description
Summary
With the policy plugin, a required to-one relation loaded through include or select can be null at runtime when the related record isn't readable. This is intentional and there's a test covering it, but the generated types don't know about it. ModelResult types the relation as non-nullable, so you can write m2.m1.value and it compiles fine and then crashes.
Optional to-one relations don't have this problem, they're already T | null.
Root cause
In packages/orm/src/client/crud-types.ts, both ModelSelectResult and the include branch of ModelResult use the schema optionality as the Optional flag:
ModelResult<
Schema,
RelationFieldType<Schema, Model, Key>,
/* Select[Key] | I[Key] */,
Options,
ModelFieldIsOptional<Schema, Model, Key>,
FieldIsArray<Schema, Model, Key>,
ExtResult
>
WrapType only adds | null if that flag is true:
type WrapType<T, Optional = false, Array = false> =
Array extends true
? Optional extends true ? T[] | null : T[]
: Optional extends true ? T | null : T;
Nothing here knows that policies can null out a required relation.
Repro
tests/e2e/orm/policy/migrated/nested-to-one.test.ts ("read rejection for non-optional relation") already hits this exact case:
model M1 {
id String @id @default(uuid())
value Int
@@allow('create', true)
@@allow('read', value > 0)
}
model M2 {
id String @id @default(uuid())
m1 M1 @relation(fields: [m1Id], references: [id])
m1Id String @unique
@@allow('all', true)
}
await db.$unuseAll().m1.create({
data: {
id: '1',
value: 0,
m2: { create: { id: '1' } },
},
});
const m2 = await db.m2.findUnique({
where: { id: '1' },
include: { m1: true },
});
m2.m1.value; // compiles
TS infers m1: { id: string; value: number }. Actual value is m1: null. Runtime:
Cannot read properties of null (reading 'value')
The test only checks the runtime value so the type side never gets caught.
Fix idea
For to-one relations, force the Optional flag to true:
FieldIsArray<Schema, Model, Key> extends true
? ModelFieldIsOptional<Schema, Model, Key>
: true
To-many is fine as is, unreadable records just get filtered out of the array.
Ideally this would only apply when the policy plugin is actually in use. Not sure what the best mechanism is. ExtResult is already threaded through ModelResult so maybe a type flag could ride along with that, or a client option.
We've been running this as a patch-package patch (no gating). After applying it TS found ~160 places in our frontend where we were dereferencing these relations without a null check. Some of those were reachable in production.
Can put together a minimal repro repo if needed but the schema/query above is basically it.
Environment
@zenstackhq/orm/@zenstackhq/plugin-policy3.8.3, also reproduced ondev- PostgreSQL
- Node 25.2.1, npm 11.6.2
- Dominant language
- TypeScript
- Stars
- 2.9k
- Forks
- 157
- Avg merge
- 11h 42m
- Merged PRs (30d)
- 20
Getting set up
This project ships no dev container, Dockerfile or contributing guide, so setting up is up to you: start from its README, and see our first-contribution guide for the general steps.
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
More from zenstackhq/zenstack
-
Difficulty 2/5 1-3 hours Newbie friendliness 68/100
zenstackhq/zenstack#2873 ·
Maintainers usually reply within 1 day
-
runtime
Difficulty 2/5 1-3 hours Newbie friendliness 78/100
zenstackhq/zenstack#2868 ·
Maintainers usually reply within 1 day
-
Difficulty 2/5 1-3 hours Newbie friendliness 72/100
zenstackhq/zenstack#2694 · 3 comments ·
Maintainers usually reply within 1 day
-
Difficulty 1/5 Under an hour Newbie friendliness 68/100
zenstackhq/zenstack#2659 · 2 comments ·
Maintainers usually reply within 1 day
-
Difficulty 2/5 1-3 hours Newbie friendliness 65/100
zenstackhq/zenstack#2542 · 1 comment ·
Maintainers usually reply within 1 day
All issues in zenstackhq/zenstack
Similar issues
-
submodule-pointer-regression
Difficulty 1/5 Under an hour Newbie friendliness 72/100
smith-horn/skillsmith#3061 ·
Maintainers usually reply within 1 day
-
area: ops type: test
Difficulty 2/5 1-3 hours Newbie friendliness 79/100
accensa/x402-facilitator-stellar#559 ·
Maintainers usually reply within 1 day
-
documentation
Difficulty 2/5 1-3 hours Newbie friendliness 74/100
cosimochellini/one-piece-zero-spoiler#551 ·
Maintainers usually reply within 1 day
-
getWatched() omits __proto__ directories when cwd is setPossibly taken @maxazure claimed this today. Open
Difficulty 2/5 1-3 hours Newbie friendliness 79/100
-
area:web enhancement
Difficulty 2/5 1-3 hours Newbie friendliness 84/100
Maintainers usually reply within 1 day