Cannot validate recursive schema - json

Question

I'm stuck on writing a correct json schema for my data.
Condition:
directory is an recursive object with optional property "$meta"
file can be string or object
if file is object it can have optional property "$meta" with oneOf "$data" or "$stringData" not both.
JSON Data:
{
"dir1": {
"dir1A": {},
"dir1B": {
"dir1B01": {
"dir1B0101": {},
"dir1B0102": {},
"dir1B0103": {},
"file1B.txt": {
"$meta": {},
"$data": "ooo",
"$stringData": "Netus et malesuada" // here should failed
}
}
}
},
"dir2": {
"file2.txt": "dolor sit amet" // here should OK but failed
},
"file1.txt": "Lorem Ipsum"
}
Schema:
{
"$schema": "https://json-schema.org/draft/2020-12/schema",
"$id": "https://example.com/dir.schema.json",
"type": "object",
"patternProperties": {
"^[^\\/?%*:|\"<>]+$": { // failed if name consist of invalid chars
"$ref": "#/$defs/directory"
}
},
"additionalProperties": false,
"$defs": {
"file": {
"oneOf": [
{
"type": "string"
},
{
"type": "object",
"properties": {
"$data": {
"type": "string",
"description": "base64 encoded data"
},
"$stringData": {
"type": "string"
}
},
"oneOf": [
{
"not": {
"anyOf": [
{
"required": [
"$data",
"$stringData"
]
}
]
}
},
{
"allOf": [
{
"not": {
"required": [
"$data"
]
}
},
{
"not": {
"required": [
"$stringData"
]
}
}
]
}
]
}
]
},
"directory": {
"anyOf": [
{
"$ref": "#/$defs/file"
},
{
"type": "object",
"properties": {
"$meta": {
"type": "object",
"additionalProperties": true,
"properties": {
"createdAt": {
"type": "string"
},
"size": {
"type": "integer"
},
}
}
},
"patternProperties": {
"^[^\\/?%*:|\"<>]+$": {
"$ref": "#"
}
}
}
]
}
}
}
Live Demo: https://www.jsonschemavalidator.net/s/HNFtNfRw
Thank you for help

Answer 1

There are a few issues, so let's walk through deriving this schema one step at a time.
Let's start with the object representation of a file. Here's the easy part.
{
"type": "object",
"properties": {
"$meta": { "$ref": "#/$defs/meta" },
"$data": { "type": "string" },
"$stringData": { "type": "string" }
}
}
Then we need to add the assertion that one and only one of $data or $stringData is required. The oneOf keyword can express this cleanly.
"oneOf": [
{ "required": ["$data"] },
{ "required": ["$stringData"] }
]
Only one of the schemas can validate to true. If both properties are present, the schema will fail.
Next, let's define a file can be represented as a string or an object.
{
"anyOf": [
{ "type": "string" },
{ "$ref": "#/$defs/object-file" }
]
}
Notice that we are using anyOf instead of oneOf in this case. anyOf passes validation as soon as one of the schemas is valid. If the first one passes validation, it doesn't need to check the second one. If oneOf is used, the validator needs to check both schemas because oneOf requires that only one of the schemas passes validation and the rest fail. Because one of these schemas is a string and the other is an object, it's impossible for any JSON data to be both a string an object. Therefore, using oneOf only makes the validation less efficient.
That covers files, let's do directories next.
{
"type": "object",
"properties": {
"$meta": { "$ref": "#/$defs/meta" }
},
"patternProperties": {
"^[^\\/?%*:|\"<>]+$": { "$ref": "#/$defs/file-or-directory" }
},
"additionalProperties": false
}
The recursive reference part is pretty simple, but we have a problem with the $meta property. $meta matches the regular expression in patternProperties, so $meta will need to validate against #/$defs/meta and #/$defs/file-or-directory, which is not what you want. What you need is something like an additionalPatternProperties keyword, but since that doesn't exist, you need to change your patternProperties regex to exclude $meta. Something like this, "^[^\\/?%*:|\"<>]+(?<!\\$meta)$".
Now all that's left is a schema to express that something is a file or a directory.
"anyOf": [
{ "$ref": "#/$defs/file" },
{ "$ref": "#/$defs/directory" }
]
Again, we want to use anyOf, but there's a bit more to this than you might notice at first glance. The problem is that something that matches the file schema will also match the directory schema. So, we can't use oneOf and the file schema needs to come first. If the directory schema was first, it would incorrectly validate files as directories. When the file schema comes first, it checks if it's a file first and the false positive that we would get from the directory schema becomes irrelevant.
I'll leave it to you to put the pieces together into a full schema, but leave a comment if you have any issues.