fix(config): drop malformed top-level fields instead of crashing the load

Reported by Ben Matthews on Discord against v1.14.45: a `skills:` field
shaped as an array (rather than the schema's `{paths?, urls?}` object)
caused the entire config load to fail with ConfigInvalidError, the
server returned 500, and the desktop app couldn't start.

Per Kit:
> for all of these things that we load from the user's computer, they
> should be kind of tolerant. We can have warnings that we log
> somewhere, but Jesus. It shouldn't break opencode.

This makes `ConfigParse.effectSchema` field-tolerant:

- Unknown top-level keys are stripped (with a warning log) instead of
  throwing `unrecognized_keys`.
- Top-level fields whose value fails decoding are dropped (with a
  warning log naming the field + the issue summary), the rest of the
  config decodes normally.
- Root-level errors (e.g. data is not an object at all) and post-strip
  failures still throw the original `InvalidError` so we don't silently
  swallow truly broken configs.

Three new tests cover the reported shape (skills as array), unknown
keys, and multi-bad-field cases. Two existing tests that asserted the
strict-throw behavior were updated to the new tolerant contract.

Pre-fix: opencode unstartable for users with malformed configs.
Post-fix: opencode starts, the bad field is ignored, the user sees a
warning in the log naming what was dropped.
This commit is contained in:
Kit Langton
2026-05-09 20:37:44 -04:00
parent 7a79f3a5ea
commit 18f3b31f1c
2 changed files with 136 additions and 35 deletions

View File

@@ -2,10 +2,13 @@ export * as ConfigParse from "./parse"
import { type ParseError as JsoncParseError, parse as parseJsoncImpl, printParseErrorCode } from "jsonc-parser"
import { Cause, Exit, Schema as EffectSchema, SchemaIssue } from "effect"
import * as Log from "@opencode-ai/core/util/log"
import z from "zod"
import type { DeepMutable } from "@opencode-ai/core/schema"
import { InvalidError, JsonError } from "./error"
const log = Log.create({ service: "config.parse" })
type ZodSchema<T> = z.ZodType<T>
export function jsonc(text: string, filepath: string): unknown {
@@ -50,34 +53,70 @@ export function effectSchema<S extends EffectSchema.Decoder<unknown, never>>(
data: unknown,
source: string,
): DeepMutable<S["Type"]> {
const extra = topLevelExtraKeys(schema, data)
if (extra.length) {
throw new InvalidError({
path: source,
issues: [
{
code: "unrecognized_keys",
keys: extra,
path: [],
message: `Unrecognized key${extra.length === 1 ? "" : "s"}: ${extra.join(", ")}`,
} as z.core.$ZodIssue,
],
})
}
// The user's config lives on disk and may legitimately be stale, hand-edited,
// or carry leftover keys from older versions. Crashing the whole load on a
// single bad field would make opencode unstartable for those users (see Ben
// Matthews / Discord, v1.14.45). Strip the malformed top-level fields and
// keep going — log every drop so users can see what was ignored and fix it.
const cleaned = stripUnknownTopLevelKeys(schema, data, source)
return decodeWithFieldTolerance(schema, cleaned, source)
}
function stripUnknownTopLevelKeys(schema: EffectSchema.Top, data: unknown, source: string): unknown {
if (typeof data !== "object" || data === null || Array.isArray(data)) return data
const extra = topLevelExtraKeys(schema, data)
if (extra.length === 0) return data
log.warn("ignoring unrecognized config keys", { source, keys: extra })
const obj = data as Record<string, unknown>
return Object.fromEntries(Object.entries(obj).filter(([key]) => !extra.includes(key)))
}
function decodeWithFieldTolerance<S extends EffectSchema.Decoder<unknown, never>>(
schema: S,
data: unknown,
source: string,
): DeepMutable<S["Type"]> {
// Try a clean decode first. If it succeeds we're done — common path.
const decoded = EffectSchema.decodeUnknownExit(schema)(data, { errors: "all", propertyOrder: "original" })
if (Exit.isSuccess(decoded)) return decoded.value as DeepMutable<S["Type"]>
const error = Cause.squash(decoded.cause)
const issues = EffectSchema.isSchemaError(error)
? (SchemaIssue.makeFormatterStandardSchemaV1()(error.issue).issues as z.core.$ZodIssue[])
: ([{ code: "custom", message: String(error), path: [] }] as z.core.$ZodIssue[])
throw new InvalidError(
{
path: source,
issues: EffectSchema.isSchemaError(error)
? (SchemaIssue.makeFormatterStandardSchemaV1()(error.issue).issues as z.core.$ZodIssue[])
: ([{ code: "custom", message: String(error), path: [] }] as z.core.$ZodIssue[]),
},
{ cause: error },
)
// Identify malformed top-level fields. Anything with a non-empty path is a
// field-scoped issue we can drop and retry. Issues with an empty path are
// root-level (e.g. data is not an object at all) and can't be field-recovered.
const badFields = collectTopLevelFieldNames(issues)
if (badFields.size === 0 || typeof data !== "object" || data === null || Array.isArray(data)) {
throw new InvalidError({ path: source, issues }, { cause: error })
}
log.warn("ignoring invalid config fields", {
source,
fields: [...badFields],
summary: issues
.filter((issue) => issue.path && issue.path.length > 0)
.map((issue) => `${issue.path?.join(".")}: ${issue.message}`)
.slice(0, 8),
})
const obj = data as Record<string, unknown>
const cleaned = Object.fromEntries(Object.entries(obj).filter(([key]) => !badFields.has(key)))
// Retry without the bad fields. If THIS fails, we're past field-tolerance —
// fall back to the original strict error so the user sees the real cause.
const retry = EffectSchema.decodeUnknownExit(schema)(cleaned, { errors: "all", propertyOrder: "original" })
if (Exit.isSuccess(retry)) return retry.value as DeepMutable<S["Type"]>
throw new InvalidError({ path: source, issues }, { cause: error })
}
function collectTopLevelFieldNames(issues: z.core.$ZodIssue[]): Set<string> {
const names = new Set<string>()
for (const issue of issues) {
const head = issue.path?.[0]
if (typeof head === "string") names.add(head)
}
return names
}
function topLevelExtraKeys(schema: EffectSchema.Top, data: unknown) {

View File

@@ -558,20 +558,22 @@ test("handles file inclusion with replacement tokens", async () => {
})
})
test("validates config schema and throws on invalid fields", async () => {
test("config loader is tolerant: drops unknown fields, keeps the rest", async () => {
await using tmp = await tmpdir({
init: async (dir) => {
await writeConfig(dir, {
$schema: "https://opencode.ai/config.json",
invalid_field: "should cause error",
username: "kept",
invalid_field: "should be dropped, not crash the app",
})
},
})
await provideTestInstance({
directory: tmp.path,
fn: async () => {
// Strict schema should throw an error for invalid fields
await expect(load()).rejects.toThrow()
const config = await load()
expect(config.username).toBe("kept")
expect((config as Record<string, unknown>).invalid_field).toBeUndefined()
},
})
})
@@ -1681,7 +1683,70 @@ test("permission config preserves user key order", async () => {
})
})
test("Effect config parser preserves permission order while rejecting unknown top-level keys", () => {
// Discord bug report (Ben Matthews, v1.14.45): a malformed `skills:` field
// (array instead of object) made the WHOLE config fail to load, the server
// returned 500, and the desktop app couldn't start. Per Kit:
// "for all of these things that we load from the user's computer, they
// should be kind of tolerant. ... It shouldn't break opencode."
// The contract: drop the malformed top-level field, log a warning, keep
// the rest of the config so the app starts.
test("config parser is tolerant: drops malformed top-level fields, keeps the rest", () => {
const config = ConfigParse.effectSchema(
Config.Info,
{
$schema: "https://opencode.ai/config.json",
username: "ben",
// Wrong shape — schema expects { paths?, urls? }, user has an array
// (looks like the LOADED skills list got pasted into the config).
skills: [
{ name: "scss-layout-accessibility", path: ".opencode/skills/scss-layout-accessibility.md" },
{ name: "testing", path: ".opencode/skills/testing.md" },
],
},
"test",
)
// Pre-fix this throws ConfigInvalidError and the user can't start opencode.
// Post-fix the bad field is dropped and the rest of the config loads.
expect(config.username).toBe("ben")
expect(config.skills).toBeUndefined()
})
test("config parser is tolerant: drops unrecognized top-level keys instead of throwing", () => {
const config = ConfigParse.effectSchema(
Config.Info,
{
$schema: "https://opencode.ai/config.json",
username: "ben",
// Typo or stale key — pre-fix this threw `unrecognized_keys`.
autoshrare: true,
},
"test",
)
expect(config.username).toBe("ben")
expect((config as Record<string, unknown>).autoshrare).toBeUndefined()
})
test("config parser is tolerant: drops multiple bad fields in one pass", () => {
const config = ConfigParse.effectSchema(
Config.Info,
{
$schema: "https://opencode.ai/config.json",
username: "ben",
skills: ["wrong shape"],
autoshare: 42, // wrong type — schema wants string literal | undefined
not_a_real_key: "ignore me",
},
"test",
)
expect(config.username).toBe("ben")
expect(config.skills).toBeUndefined()
expect(config.autoshare).toBeUndefined()
})
test("Effect config parser preserves permission order while dropping unknown top-level keys", () => {
const config = ConfigParse.effectSchema(
Config.Info,
{
@@ -1695,13 +1760,10 @@ test("Effect config parser preserves permission order while rejecting unknown to
)
expect(Object.keys(config.permission!)).toEqual(["bash", "*", "edit"])
try {
ConfigParse.effectSchema(Config.Info, { invalid_field: true }, "test")
throw new Error("expected config parse to fail")
} catch (err) {
const error = err as { data?: { issues?: Array<{ code?: string; keys?: string[]; path?: string[] }> } }
expect(error.data?.issues?.[0]).toMatchObject({ code: "unrecognized_keys", keys: ["invalid_field"], path: [] })
}
// Tolerant parser: unknown keys are stripped (with a warning log) instead
// of failing the entire config load.
const stripped = ConfigParse.effectSchema(Config.Info, { invalid_field: true }, "test")
expect((stripped as Record<string, unknown>).invalid_field).toBeUndefined()
})
// MCP config merging tests