Change it to match the following process:

1. generation of single testdata
2. add referenceKeys to map
   Format: key => values [IGeneratedFieldValue, ...]
3. add referenceTo to second map; paths is the path where one random element of values gets put in
   Format: toKey => paths Set["object>prop>prop"]
4. transform single testdata to real json
5. save bulk of data (~10000 generated fields, depending on field size) to single sqlite row

repeat for all

1. resolve keys of single bulk of testdata

repeat for all

1. transform single bulk of data
2. save bulk to file if specified and delete bulk from db

repeat for all

1. output bulk to cmd if specified and delete bulk from db

repeat for all

This should increase the performance and reduce the RAM/Disk usage due to way smaller data.

This is the old format in that it is saved:

export interface IGeneratedCollection {
    dbName: string,
    collectionName: string,
    documents: IGeneratedDocument[]
}

export interface IGeneratedDocument {
    documentFields: IGeneratedField []
}

export interface IGeneratedField {
    fieldName: string,
    fieldValue: any,
    fieldNeedsQuotations?: boolean,
    fieldIsObject?: boolean,
    fieldIsJsonObject?: boolean,
    fieldIsArray?: boolean,
    unboxElements?: boolean,
    referenceKey?: number,
    referenceTo?: number
}

Which would result in a example code of this - and there aren't even any sub objects, arrays, references, ...:

{
    "dbName": "db",
    "collectionName": "users",
    "documents": [
        {
            "documentFields": [
                {
                    "fieldName": "Name",
                    "fieldValue": "Hans Peter",
                    "fieldNeedsQuotations": true
                },
                {
                    "fieldName": "Age",
                    "fieldValue": 23,
                    "fieldNeedsQuotations": false
                }
            ]
        }
    ]
}

By saving it with real JSON format it looks like this:

{
    "dbName": "db",
    "collectionName": "users",
    "documents": [
        {"Name": "Hans Peter", "Age": 23}
    ]
}

Obviously some information gets lost there, but the information is only needed for reference resolving and parsing to JSON. So if we extract the reference Information to extra HashMaps, we can directly transform it into real JSON and save about 2/3 of space.

When a dynamic array has 0 elements, it writes null in the mongodb output