filter anyOf and oneOf alternative by animeshsahoo1 · Pull Request #180 · hyperjump-io/json-schema-errors

animeshsahoo1 · 2026-03-12T13:58:08Z

The anyof and oneof error handlers now filter alternatives based on the instance's properties. For object instances, an alternative is excluded from the error report if none of its declared properties exist on the instance (Rule 1), or if none of the instance's properties pass validation in that alternative (Rule 2). For non-object instances, the original type-keyword filtering is done.

fixes #175

animeshsahoo1 · 2026-03-12T14:00:09Z

hi @jdesrosiers I have implemented what we discussed about, I tried to not add anything unnecessary but if there is need of changes let me know, I made sure to run the linting and tests as well.

jdesrosiers

Please add tests. They need to cover all the edge cases we discussed and more if you can think of them.

animeshsahoo1 · 2026-03-13T03:05:41Z

I did run it on tests we talk about but I was unsure if I had to add the code for that, I will add it soon.

animeshsahoo1 · 2026-03-14T16:28:21Z

I have added the test case for matchcount=0 for oneOf and anyOf test cases that were discussed, I tried to think of a failing case but didn't really find any let me know if I could add any other test case

animeshsahoo1 · 2026-03-16T14:26:56Z

@jdesrosiers tagging you in case you didn't get notification since I didn't tag last time

jdesrosiers · 2026-03-17T21:19:43Z

You don't need to tag me. I'll get the notification. In fact, bumping the issue will likely make it take longer before I get to it. I generally work on the most stale issue on my plate first. So, if you bump the issue, it effectively moves your issue to the back of the queue.

jdesrosiers

Please try to match the code style used in the rest of the project. Please turn off Prettier or at least give it a much more reasonable line length limit.

I think this works! But, getting it to work is just the first step. This needs some refactoring.

Every time you do a filter operation, you're looping over everything again and creating a new array. Creating a bunch of arrays and objects gets expensive at scale. See if you can refactor so that you can loop over the alternatives only one time.

Also try to avoid things like [...instanceProps]. Again, we're unnecessarily creating an array. Have a look at @hyperjump/pact. It allows you to use filter, map and bunch of other things that work on any iterator (not just arrays) and don't create unnecessary intermediate representations. It looks like it could come in handy in a few places.

src/error-handlers/anyOf.js

jdesrosiers · 2026-03-18T03:28:50Z

src/error-handlers/anyOf.js


+/** @type (alternative: NormalizedOutput, propLocation: string) => boolean */
+const propertyPasses = (alternative, propLocation) => {
+  const propOutput = alternative[propLocation];


alernatative and propLocation aren't used individually. Should we just pass propOutput directly instead?

src/error-handlers/anyOf.js

jdesrosiers · 2026-03-18T03:58:09Z

src/error-handlers/anyOf.js

+      filtered = filtered.filter((alternative) => {
+        const typeResults
+          = alternative[instanceLocation]?.[
+            "https://json-schema.org/keyword/type"
+          ];
+        return (
+          !typeResults || Object.values(typeResults).every((isValid) => isValid)
+        );
+      });

-    if (alternatives.length === 0) {
-      for (const alternative of anyOf) {
-        alternatives.push(await getErrors(alternative, instance, localization));
+      if (filtered.length === 0) {
+        filtered = anyOf;
      }


Is it right for this to be in an else? It seems to me that we should be doing this check always and doing it first. It's the most generic check and the best chance to filter out potential cases early.

yeah that's true I focused too much on object cases this should be done at start to filter out many cases

animeshsahoo1 · 2026-03-20T22:54:03Z

hi I have done the changes and used hyperjump pact to avoid creating intermediate representations, let me know if there is still somewhere I should change

animeshsahoo1 · 2026-03-22T22:48:26Z

Hi I have fixed the failing npm run lint command with new oxlint linting, please check if its ok.

jdesrosiers · 2026-03-22T22:41:11Z

src/error-handlers/anyOf.js

+      filtered = [];
      for (const alternative of anyOf) {
-        alternatives.push(await getErrors(alternative, instance, localization));
+        const typeResults = alternative[instanceLocation]["https://json-schema.org/keyword/type"];
+        if (!typeResults || Object.values(typeResults).every((isValid) => isValid)) {
+          filtered.push(alternative);
+        }
      }


This is now duplicated code. You don't need the if/else. Remove the if wrapper and the else block entirely.

So basically remove if else wrapper put the type check at top for each alternative then if its an object we check rule1 and rule2 right

jdesrosiers · 2026-03-22T22:53:10Z

src/error-handlers/anyOf.js

+        const declaredProps = Object.keys(alternative)
+          .filter((loc) => loc.startsWith(prefix))
+          .map((loc) => /** @type {string} */ (Pact.head(JsonPointer.pointerSegments(loc.slice(prefix.length - 1)))));


Use Pact here to avoid creating intermediate arrays.

jdesrosiers · 2026-03-22T23:51:56Z

src/error-handlers/anyOf.js


+/** @type (propOutput: NormalizedOutput[string] | undefined) => boolean */
+const propertyPasses = (propOutput) => {
+  if (!propOutput || Object.keys(propOutput).length === 0) {


I don't understand the purpose of the check for an empty object and the tests don't cover it. Either remove it or add a test showing it's needed.

jdesrosiers

I found a bug. Here's a test that exposes it.

{
  "$schema": "https://json-schema.org/draft/2020-12/schema",

  "anyOf": [
    {
      "type": "object",
      "properties": {
        "a": { "type": "string" }
      },
      "required": ["a"]
    },
    {
      "type": "object",
      "properties": {
        "b": { "type": "string" }
      },
      "required": ["b"]
    }
  ]
}

{ "a": 42 }

The first alternative is getting filtered by the discriminator check (rule 2), but "a" is not a discriminator because it doesn't have a passing value in any other alternatives.

animeshsahoo1 · 2026-03-25T15:26:27Z

yeah this is definitely an edge case I was thinking we should add to rule 2: only apply this filter if the failing properties all have passing constraints in at least one other alternative
this works on the previous example as well as the current example:

{
  "anyOf": [
    {
      "type": "object",
      "properties": {
        "type": { "const": "a" },
        "x": { "type": "string" }
      },
      "required": ["type", "x"]
    },
    {
      "type": "object",
      "properties": {
        "type": {
          "allOf": [
            { "type": "string" },
            { "minLength": 2 }
          ]
        },
        "x": { "type": "number" }
      },
      "required": ["type", "x"]
    }
  ]
}

Instance: { "type": "a", "x": 42 }
here both don't get filtered as suggested earlier and

{
  "$schema": "https://json-schema.org/draft/2020-12/schema",

  "anyOf": [
    {
      "type": "object",
      "properties": {
        "a": { "type": "string" }
      },
      "required": ["a"]
    },
    {
      "type": "object",
      "properties": {
        "b": { "type": "string" }
      },
      "required": ["b"]
    }
  ]
}

instance: { "a": 42 } this also works

so the new rule2 is:
Filter out an alternative if no instance property passes in it, AND all of its failing properties have a passing constraint in at least one other alternative. I wanna look at its edge case a bit more would love your views too.

jdesrosiers requested changes Mar 12, 2026

View reviewed changes

jdesrosiers requested changes Mar 18, 2026

View reviewed changes

animeshsahoo1 added 4 commits March 23, 2026 04:09

filter anyOf and oneOf alternative

bfb6925

added tests for new behavior

80697b2

used hyperjump pact for better performance

72d179f

fixed linting issues with oxlint

803ece5

animeshsahoo1 force-pushed the filter-anyOf-oneOf-alternatives branch from ba77800 to 803ece5 Compare March 22, 2026 22:46

jdesrosiers requested changes Mar 22, 2026

View reviewed changes

jdesrosiers reviewed Mar 24, 2026

View reviewed changes

Uh oh!

Conversation

animeshsahoo1 commented Mar 12, 2026

Uh oh!

animeshsahoo1 commented Mar 12, 2026

Uh oh!

jdesrosiers left a comment

Choose a reason for hiding this comment

Uh oh!

animeshsahoo1 commented Mar 13, 2026

Uh oh!

animeshsahoo1 commented Mar 14, 2026

Uh oh!

animeshsahoo1 commented Mar 16, 2026

Uh oh!

jdesrosiers commented Mar 17, 2026

Uh oh!

jdesrosiers left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

animeshsahoo1 commented Mar 20, 2026

Uh oh!

animeshsahoo1 commented Mar 22, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

jdesrosiers left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

animeshsahoo1 commented Mar 25, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

jdesrosiers left a comment •

edited

Loading

animeshsahoo1 commented Mar 25, 2026 •

edited

Loading