lmdb-index

• object indexing for LMDB,
• fulltext indexing and queries,
• index and vector based queries using literals, functions, and regular expressions,
• over 50 pre-built functions for use in queries, e.g. $lte, $echoes (soundex), $includes,
• approximate matching using $ior,
• Levenshtein, Euclidean, Manhattan, Cosine, Color distance queries using $distance,
• automatic id generation,
• instantiation of returned objects as instances of their original classes,
• copy, move, and patch operations,
• ACID transactions.

This is a mid-level API. For the highest level LMDB object query API see lmdb-oql.

Installation

npm install lmdb-index --save

You need to install lmdb separately, it is not a dependency of lmdb-index

Usage

import {open} from "lmdb";
import {withExtensions,operators} from "lmdb-index";

const {$gte} = operators;

class Person {
    constructor(config={}) {
        Object.assign(this,config);
    }
}
const db = withExtensions(open("test",{indexOptions:{fulltext:true}}));
db.defineSchema(Object);
db.defineSchema(Person);
const id = await db.put(null,new Person({name:"bill",age:21,address:{city:"New York",state:"NY"}}));
if(id) {
    const person = await db.get(id);
    // the below all log the same thing, i.e. Person {name:"bill",age:21,"#":"Person@<some uuid>"}
    console.log(person);
    [...db.getRangeFromIndex({name:"bill"})].forEach((person) => {
        console.log(person)
    });
    [...db.getRangeFromIndex({name:/bil.*/})].forEach((person) => {
        console.log(person)
    });
    [...db.getRangeFromIndex({age:$gte(21)})].forEach((person) => {
        console.log(person)
    });
    [...db.getRangeFromIndex({age:(value) => value===21 ? value : undefined})].forEach((person) => {
        console.log(person)
    });
    [...db.getRangeFromIndex({
        [/name/]:"bill",
            address:{city:"York"} // fulltext indexing search turned on, partial matches returned so long as `valueMatch` is relaxed
        },
        (value)=>value) // relaxed value match
    ].forEach((person) => {
        console.log(person)
    },null,{fulltext:true});
}

Inclusive Or Search

Vector Based Search

API

async db.copy

async db.copy(key:LMDBKey,?destKey:LMDBKey,?overwrite:boolean,?version:number,?ifVersion:number) - returns LMDBKey

Works similar to lmdb-copy but provides automatic key assignment and indexing.

If key refers to an object for which there is a schema:

• If destKey is nullish, the destKey is given a value using the schema's key generator or a UUIDv4 value if no key generator is defined.
• The copy is inserted and indexed inside a transaction
• The destKey is returned if the transaction succeeds, otherwise undefined is returned.

If key points to a primitive or an object without a schema:

• If destKey is nullish, it given a UUIDv4 value
• The copy is inserted at the destKey (No indexes are updated because primitives are not indexed.)
• The destKey is returned if the insertion succeeds, otherwise undefined is returned.

async db.defineSchema

async db.defineSchema(classConstructor:function|class,?options={}) - returns boolean

• The key names in the array options.indexKeys will be indexed. Nested keys use dot notation. If no value is provided, all keys will be indexed. If options.indexKeys is an empty array, no keys will be indexed.
• If the property options.idKey is provided, its value will be used for unique ids. If options.idKey is not provided, the property # on instances will be used for unique ids.
• If the property options.keyGenerator is provided, its value should be a function that returns a unique id. If options.keyGenerator is not provided, a v4 UUID will be used.
• Schema are not currently used for data validation. They are just used for index and id creation. Validation will be added in a subsequent release.

The options properties and values are inherited by child schema, i.e. if you define them for Object, then you do not need to provide them for other classes.

To index all keys on all objects using UUIDs as ids and # as the id key, call await db.defineSchema(Object).

Note: All operations returning an object attempt to create an instance of the object's original class if a schema is defined.

async db.index

async db.index(value:object,?cname:string|undefined) - returns LMDBKey|undefined

Inserts (puts) object and indexes it using the id property of the object or a generated id if no id property exists. Coerces value to class cname if provided. See db.put(bull,value) for full details.

The id property defaults to # but can be set using db.defineSchema(classConstructor,{idKey:"id"}).

db.getSchema

db.getSchema(value:string|object,create:boolean) - returns object representing schema

Returns the schema for the class of the object or undefined.

If create is true and value is an object and no schema exists, a schema is created and returned.

*db.getRangeFromIndex

*db.getRangeFromIndex(indexMatch:object,?valueMatch:function|object,?select:function|object,{?cname:string,?scan:boolean,?sortable:boolean,?fulltext:boolean|number,?minScore:number,?sort:boolean|function,?versions:boolean,?offset:number,limit=||Infinity}=?options={}) - returns IterableIterator of object`

Yields objects of the form {key,value,count,version} where key is the key of the object, value is the object in the database, count is the number of index matches, score is the percentage (as a decimal) of indexMatch that matched, version` is the version number and is only present if the database was opened using versioning.

indexMatch is an object with keys that may be serialized RegExp and values that may be literals, or RegExp or functions that return non-undefined values or DONE. For example:

Literal value matches are by far the most performant. Function and RegExp value matches require an index scan across all values for a particular property. RegExp property matches require an index scan across all properties. Using both a RegExp property match and a function or RegExp value match is very expensive.

Some query optimization is conducted, e.g. literal matches are processed first so that queries can fail early.

Since indexed are sorted in ascending order, some research is being done into operator level optimizations for things like $gt, $lt.

The standard form for functions used in queries is (value) => ... some code that returns a value or undefined or DONE. These can be quite verbose, so over 50 pre-built functions are provided in the section Operators.

valueMatch defaults to the same value as indexMatch because some of the processing required can't actually be done against indexes, underlying values must be retrieved. However, it can also be a function that accepts a candidate match and returns the candidate if is satisfies certain conditions or undefined if not. Or, it can be a different pattern matching object that is perhaps more restrictive.

select is a function or object that is used to select which properties of the object are returned. If select is an object, then the properties of the select are used as keys to select properties from the match. The properties may be serialized RegExp. The values of these properties can be literals, RegExp, or functions.

Functions returning undefined or RegExp and literals that do not match drop properties. Make sure RegExp that are used for values contain a selection group. Functions are called with the object as the first argument and {root,parent,key} as the second. For example:

{
    name: (value,{root}) => { root.firstInitial = value.match(/(.).*/)[1] }
    [/.*Identifier/]: (value) => value!=null ? value : undefined
}

cname is the name of the class to query. It defaults to the constructor name of the indexMatch except if indexMatch is just an Object. If indexMatch is an Object and no cname is provided, then the match is done across multiple classes, e.g.

class Person { constuctor(config={}) { Object.assign(this,config); } }
class Pet { constuctor(config={}) { Object.assign(this,config); } }
await db.defineSchema(Person);
await db.defineSchema(Pet);
await db.put(null,new Person({name:"bill",age:21}));
await db.put(null,new Pet({name:"bill",age:2}));
[...db.getRangeFromIndex({name:"bill"})].forEach((object) => {
    console.log(object); // logs both the Person and the Pet
});
[...db.getRangeFromIndex({name:"bill"},null,null,{cname:"Person"})].forEach((object) => {
    console.log(object); // logs just the Person
});

scan tells the engine to scan the class instances if no indexed fields are present in the indexMatch and test each one.

sortable, 'fulltext, and sort start returning entries almost immediately based on partial index matches.

sortable and fulltext return entries in the order they are indexed.

fulltext (short for full text index), returns entries that have partial matches for string property values.

If fulltext is true, then all partial matches are returned.

If fulltext is a number between 0 and 1, then only matches that exceed the number as a percent match are returned.

To fully utilize fulltext, ensure the database is opened and entries have been indexed with {indexOptions:{fulltext:true}}. For example:

const db = open("test",{indexOptions:{fulltext:true}});
db.put(null,{name:"bill",address:{city:"New York",state:"NY"}});
db.put(null,{name:"joe",address:{city:"York",state:"PA"}});
[...db.getRangeFromIndex({address:{city:"York"}},null,null,{fulltext:true})].forEach((person) => {
    console.log(person); // logs both bill and joe
});

Note: When {indexOptions:{fulltext:true}} is set, passing functions as property values in indexMatch causes an index scan that may be expensive.

If sort is true entries are returned based on how many index matches occurred, with the highest first. If sort is a function, then entries are returned in the order determined by the function. Note, both of these are expensive since they require resolving all matches first.

// this `indexMatch` pattern
{
    name: /bil.*/,
    age: $gte(21), // this is the same as (value) => value>=21 ? value : undefined,
    address: {
        city: "Old York"
    },
    [/.*Identifier/]: (value) => value!=null ? value : undefined
}
// would match this object
{
    name: "Bill", // count++,
    age: 21, // count++
    address: {
        city: "New York" // count++
    },
    someIdentifier: "12345" // count++
}
// with fulltext indexing and search turned on
// count = 4, would have been 5 if looking for Old York (one hit for each word)
// score = .8333, i.e. 5/6 possible /bil.*, $gte, one for New, one for York, /.*Identifier/, (value) =>

*db.getRangeFromVector

BETA

*db.getRangeFromVector(vector:object,?valueMatch:function|object,?select:function|object,{?cname:string,?distance=euclidianDistance,?scan:boolean,?sortable:boolean,?maxDistance:number,?sort:boolean|function,?versions:boolean,?offset:number,limit=||Infinity}=?options={}) - returns IterableIterator of object

Works the same as getRangeFromIndex above, except that the vector is used to select the items and a slightly different object is returned.

vector is NOT an array, it is an object from which a vector array is automatically generated base on a schema definition. The schema can be automatically inferred using db.inferSchmea.

Yields objects of the form {key,value,distance,version} where key is the key of the object, value is the object in the database, distance is the distance from the selection vector, version is the version number and is only present if the database was opened using versioning.

See Vector Search below for more information.

db.inferSchema

BETA

db.inferSchema(dataset:array,{?cname,?vectors,?save,baseURI="",?$schema,?schemata={}}={}) - returns schemata option

dataset is an array of objects to be used to infer the schema.

If cname is provided, then the schema is learned for that class. If cname is not provided, then the classname of each object processed is used.

If vectors is true, then the schema is enhanced for vector search.

If save is true, then the schema is saved to the database.

The baseURI defaults to "", but a URL can be provided in case you are implementing a server that will return schema. The value is used in generating schema $id values.

If $schema is provided, it is used to specify the dialect of JSON schema to use, per the JSON schema standard.

If schemata is provided, then it is used to provide default schema values for named schema. Each key name in schemata is a classname and the corresponding property holds a schema definition. The schemata object is updated as new schema are learned. Pass db.schema, if you wish to use the existing schema.

See Schema below for more information.

async db.move

db.move(key:lmdbKey,destKey:lmdbKey,?overwrite:boolean,?version:number,?ifVersion:number) - returns LMDBKey|undefined

Works similar to lmdb-move but provides automatic key assignment and indexing.

Works the same as copy above, except the entry at the original key is removed inside the transaction.

async db.patch

async db.patch(key:string,patch:object,?version,?ifVersion) - returns boolean

Inside a single transaction, updates the index after the patch.

Also see lmdb-patch

async db.put

async db.put(key:LMDBKey,value,?cname,?version,?ifVersion) - returns LMDBKey|undefined

Works similar to lmdb put

When putting an object for indexing, the key should be null. It is retrieved from the object using the idKey of the schema. If there is a mismatch between the key and the idKey of the object, an Error will be thrown.

If value is an object and key is null, it will be indexed by the keys of the object so long as it is an instance of an object controlled by a schema declared with defineSchema. To index all top level keys on all objects, call db.defineSchema(Object). If key is null, a unique id will be generated and added to the object. See defineSchema for more information.

When an object is indexed db.get and db.getEntry with return the object as an instance of its original class.

The cname argument is used to specify the class name of the object being put. If cname is not provided, the class name is determined by the constructor name of the value argument. This allows the developer to use plain objects. If value is a primitive, cname is ignored.

If there is a mismatch between the key and the idKey of the object, an Error will be thrown.

The key or the object id (in the case of indexed object) is returned if the transaction succeeds, otherwise undefined is returned.

db.putSync

db.putSync(key:LMDBKey,value,?cname,?version,?ifVersion) - returns LMDBKey|undefined

Synchronous version of db.put.

DO NOT USE: An underlying issue with lmdb results in this occasionally returning a Promise instead of a value. Use await db.put instead.

async db.remove

async db.remove(key:LMDBKey,?version:number,?ifVersion:number) - returns LMDBKey|undefined

Same behavior as lmdb except that the index entries are removed inside a transaction

db.removeSync

db.removeSync(key:LMDBKey,?version:number,?ifVersion:number) - returns LMDBKey|undefined

Synchronous version of db.remove.

DO NOT USE: An underlying issue with lmdb results in this occasionally returning a Promise instead of a value. Use await db.put instead.

withExtensions

withExtensions(db:LMDBDatabase,?extensions:object) - returns LMDBDatabase

Extends an LMDB database and any child databases it opens to have the extensions provided. This utility is common to other lmdb extensions like lmdb-patch, lmdb-copy, lmdb-move.

Returns a child database that has the extensions copy, getRangeFromIndex, getRangeFromVector, getVector, index, indexSync, inferSchema, move, patch and modified behavior of clearAsync, clearSync, put, putSync,remove and removeSync.

Schema

With a few minor enhancements, lmdb-index uses JSON Schema to define the structure of objects stored in the database.

The schema is used to automatically generate indexes and vectors for vector search.

It is not currently used for validation, but that is planned for a future release.

Defining Schema

Schema can be defined using db.defineSchema or db.inferSchema.

Schema used by lmdb-index is a superset of JSON Schema. The following additions are supported:

schema.indexKeys

• If schema.indexKeys is missing, then all keys of objects of the class are indexed.
• If schema.indexKeys is an array, then only the keys in the array are indexed. Use an empty list to turn off indexing.

property.type = "date" BETA

• Currently used to support vectorizing dates.

property.vector BETA

• If property.vector is true, then the property is used when a vector representations of the object managed by the schema is generated. If the property is constrained by oneOf, then a position of the value in the oneOf array divided by the number of values is used. If the property is a number constrained by minimumValue and maximumValue, then a normalized value (value - minimumValue)/,maximumValue is used.
• If property.vector is a function, then the function is called with the property value and the result is used. This is useful for things like color.

When learning schema, a partially defined schema can be provided using the schemata options argument.

During the learning process, schema are updated with the following information if it is not already defined when a schema is passed in as part of the schemata option:

• $schema
• $id
• $ref
• property types
• property minimum and maximum values if numbers
• property oneOf specifications if strings
• if a property is present in all samples, it is marked as required

CAUTION: Vector functions are not currently persisted and restored from the database with their schema. Hence, all schema using vector functions must be defined each time the database is opened. With the poduction release of this feature, they will be persisted and restored. As a result, they will present a security risk. Do not provide end-user ability to edit schema execept in a fully trusted environment.

Approximate Match Searches

lmdb-index offers three mechanisms for searching data for approximate or partial matches.

1. Partial matches against full text indexes
2. Partial matches using the $ior operator that increments a count for each portion of the or that is matched
3. Partial matches using vector representations

The first two approaches can be accomplished using db.getRangeFromIndex. The third approach can be accomplished using db.getRangeFromVector

Full Text Indexing

When a database is opened with {indexOptions:{fulltext:boolean|array}}, all string values associated with objects and properties enabled for full text search are tokenized and stop words, e.g. and, or, but, are removed.

All objects are enabled for full text search by providing true as the value for fulltext; otherwise, all classes with constructors identified an array are indexed.

The identification of classes can be done by providing their class/constructor or a string name.

When fulltext indexing is enabled, the number of index entries increases at a minimum to the number of unique tokens. As a result, many more possible objects match each entry.

See db.getRangeFromIndex for more detail

Vector Search

Vector based search is a technique for finding the closest matches to a vector in a multi-dimensional space. It is often used in machine learning and artificial intelligence, but can also be used for simple approximate matching.

See this the Medium article Vector Search Simplified for an explanation.

This feature is in early BETA. It is not recommended for production use.

The current release supports two types of vector comparisons:

1. comparisons of object property values using the operator $distance
2. comparisons of full objects converted to vectors

$distance

The vector comparison methods supported include:

• Cosine Similarity
• Euclidean Distance
• Manhattan Distance
• Jaccard Distance
• Color distance (see below)

The operator $distance has the signature:

$distance([value:string|array,maxDistance:number,?method:function=typeof(value)==="string" ? levenshteinDistance : euclidianDistance]) - returns value the string is being compared to if the similarity is greater than or equal to lowerBound, otherwise undefined.

Note: levenshteinDistance is not a vector comparison method, but is a common means of determining string similarity.

If maxDistance is a fraction between 0 and .99999999999, Euclidean and Manhattan vectors are normalized by dividing each element in the vector by the maximum value in the vector. This can be used to make the distance comparable to the other methods, it will not change the actual similarity. And things that look like percentages are often easier to think about that absolute numbers, even if they are not really a percentage.

The methods are:

• cosineDistance
• euclideanDistance
• manhattanDistance
• jaccardDistance
• colorDistance

For cosineDistance, euclideanDistance and manhattanDistance, if the value argument and property it is being compared to are strings, they will be tokenized using the same dictionary. If the value and target property are arrays of numbers, they will be used as is. Attempts to compare other combinations will return undefined.

For colorDistance, uses euclideanDistance after doing some conversions adn validation. If the value argument and property are color names, hex values, rgb(R,G,B) or rgba(R,G,B,A), they will be converted to RGBA vectors as [R,G,B,A]. A defaults to 1. If they are 3 element numeric arrays an opacity of 1 will be added. If they are 4 element numeric arrays they will be used without change. Attempts to compare other combinations or arrays with invalid color values, i.e. numbers outside the range 0-256 for the first three elements and 0 to 1 for the 4th will return undefined.

The methods can be imported from lmdb-index/src/operators.js.

Although the methods will throw errors if called directly with the wrong types, the $distance operator will return undefined if the types are wrong. This is because JSON object properties are not strongly typed. If you want to ensure that the types are correct, you can use the $type operator to check the type before calling the $distance operator.

For example:

import {open} from "lmdb";
import {withExtensions} from "lmdb-index";
const db = withExtensions(open("test.db"));
db.defineSchema(Object);
await db.put(null,{type:"newspaper",headline:"Elon Musk's Boring Co to build high-speed airport link in Chicago"});
await db.put(null,{type:"newspaper",headline:"Elon Musk's Boring Company to build high-speed Chicago airport link"});
await db.put(null,{type:"newspaper",headline:"The quick brown fox really jumped over the lazy dog in Seattle"});

for(const item of db.getRangeFromIndex({type:"newspaper",headline:$similarity(["Elon Musk's Boring Co to build high-speed O'Hare Airport link",.75,cosineDistance])})) {
    console.log(item); // logs the first two items
}
for(const item of db.getRangeFromIndex({type:"newspaper",headline:$similarity(["Elon Musk's Boring Co to build high-speed airport link in Chicago",.99,cosineDistance])})) {
    console.log(item); // logs only the first item
}

Objects As Vectors

Traditional JavaScript objects can be turned into vectors by:

1. Flattening their structure while turning properties and values into numbers
2. Representing most string values as their 1 based offset into arrays of all possible values / number of possible values
3. Representing most numeric values normalized to the range 0 to 1, (value - min of all possible values) / maximum of all possible values
4. Boolean values as 0 and 1.
5. Dates can be converted into their numeric value in milliseconds, and normalized like other numbers.

The method db.getRangeFromVector turns objects into vectors this automatically based on schema definitions.

The method db.inferSchema, can be used to learn schema definitions from a set of objects.

The method db.getVector can be used to get the vector for an object.

The default distance function for vectorized objects is euclideanDistance.

Operators

The following operators are supported in indexMatch, valueMatch and select.

Logical

• $and(...operatorResult) - logical and
• $or(...operatorResult)) - logical or
• $not(...operatorResult)) - logical not
• $ior(...operatorResult)) - BETA:fuzzy matching inclusive or (more matches increase score)
• $xor(...operatorResult)) - BETA:exclusive or

Comparison

• $lt(boolean|number|string) - less than
• $lte(boolean|number|string) - less than or equal to
• $gt(boolean|number|string) - greater than
• $gte(boolean|number|string) - greater than or equal to
• $eq(boolean|number|string) - equal to
• $eeq(boolean|number|string) - equal to and same type, e.g. 1 is not equal to `'1'
• $neq(boolean|number|string) - not equal to
• $between(boolean|number|string,boolean|number|string) - property value is between the two values (inclusive)
• $outside(boolean|number|string,boolean|number|string) - property value is not between the two values (exclusive)

String

• $startsWith(string) - property value starts with string
• $endsWith(string) - property value ends with string
• $matches(RegExp) - property value matches regular expression
• $echoes(string) - property value sounds like the string
• $distance([value:string|array,upperBound:number,?method:function=levenshteinDistance) - BETA property value within upperBound distance to the value. See Vector Search for more information on upperBound and method.

Arrays and Sets

• $in(array) - property value is in array
• $nin(array) - property values is not in array
• $includes(boolean|number|string|null) - property value is an array and includes value
• $excludes(boolean|number|string|null) - property value is an array and does not include value
• $intersects(array) - property value is an array and intersects with array
• $disjoint(array) - property value is an array and does not intersect with array
• $subset(array) - property value is an array and is a subset of array
• $superset(array) - property value is an array and is a superset of array
• $symmetric(array) - property value is an array and has same elements as array

Basic Types

• $type(typeName:string) - property value is of typeName type
• $isOdd() - property value is odd
• $isEven() - property value is even
• $isPrime() - property value is prime
• $isComposite() - property value is composite
• $isPositive() - property value is positive
• $isNegative() - property value is negative
• $isInteger() - property value is an integer
• $isFloat() - property value is a float
• $isNaN() - property value is not a number
• $isArray() - property value is an array
• $isObject() - property value is an object
• $isPrimitive() - property value is a primitive
• $isUndefined() - property value is undefined
• $isNull() - property value is null
• $isTruthy() - property value is truthy
• $isFalsy() - property value is falsy

Extended Types

• $isCreditCard() - property value is a credit card number
• $isEmail() - property value is an email address
• $isHexColor() - property value is a hex color
• $isIPV4Address() - property value is an IP address
• $isIPV6Address() - property value is an IP address
• $isISBN() - property value is an ISBN
• $isMACAddress() - property value is a MAC address
• $isURL() - property value is a URL
• $isUUID() - property value is a UUID
• $isZIPCode() - property value is a ZIP code

Testing

Unit Testing

Unit testing conducted with jest.

File	% Stmts	% Branch	% Funcs	% Lines	Uncovered Line #s
All files	89.12	79.22	96.66	92.28
lmdb-index	87.2	73.95	92.3	90.65
index.js	87.2	73.95	92.3	90.65	...6,532-538,547,682,694,826,988,993-996,1023-1024,1068-1069
lmdb-index/src	95.55	93.47	100	98.36
operators.js	95.55	93.47	100	98.36	91-93

Performance Testing

Performance testing conducted with benchmark.

At low scale, i.e. 3 million records, lmdb-index handles:

• about 500,000 records a second asynchronous inserting primitives,
• about 1,500 records a second asynchronous inserting and indexing objects without disk write (but available to transactions),
• about 1,000 records a second sequential asynchronous inserting, indexing objects and writing to disk,
• 1,500,000+ records per second for direct key retrieval,
• 80,000 to 120,000 records a second for search and retrieval depending on the complexity of the query.

The maximum batch put size before instability is about 500,000 records.

Since vector search is in BETA, it's performance has not yet been measured.

Release Notes (Reverse Chronological Order)

2023-06-04 v1.1.0 Enhanced documentation. Added full vector query capability (BETA) with db.getRangeFromVector, db.inferSchema, db.getVector and updates to db.put, db.copy, etc.

2023-06-02 v1.0.0 Enhanced documentation. Fixed issue with db.clearAsync not awaiting all clears of child databases. Added key to entries returned by db.getRangeFromIndex to match documentation. Added selector as exported function. Enabled caching on primary and index databases. Added $distance operator.

2023-05-22 v0.11.3 Enhanced documentation. Added unit tests. Resolved the underlying issue related to db.index, it will no longer be deprecated. Final BETA release.

2023-05-21 v0.11.2 Enhanced documentation. Addressed/documented underlying issues with lmdb: https://github.com/kriszyp/lmdb-js/issues/235 and https://github.com/kriszyp/lmdb-js/issues/234.

2023-05-21 v0.11.1 Enhanced documentation. Added unit tests. Added scan option to db.getRangeFromIndex. Fixed issue with putSync where special values were not getting serialized and underlying lmdb library falsely reporting a db.putSync failed. Improved fulltext index matching counts when the same word is repeated. Discovered and documented that db.putSync sometimes returns a Promise. Advise against using.

2023-05-19 v0.11.0 Added an index to better support queries where the value is known but properties might be ambiguous.

2023-05-19 v0.10.2 Fixed issues related to indexOptions not being passed in when database opened.

2023-05-19 v0.10.1 Refined operator functions that are order dependent so they return DONE after upper bound. db.clearAsync and db.clearSync now clear indexes. db.putSync returning Promise. Corrected v0.9.1 and v0.9.0 dates below. Minor modifications to index structure. db.index and db.indexSync will be deprecated prior to v1, use db.put(null,object) or db.putsync(null,object) instead. BREAKING CHANGE: Fulltext indexing must now be enabled with indexOptions:{fulltext:true} when opening a database.

2023-05-17 v0.9.1 Removed un-necessary files from npm package.

2023-06-17 v0.9.0 Added unit tests. Addressed issue with RegExp and select. $echoes now handles numbers. Added some performance benchmarks.

2023-05-16 v0.8.1 Added unit tests. Addressed issue with nested object indexing and matching keys with RegExp.

2023-05-15 v0.8.0 Updated documentation. Corrected issue with indexKeys on schema not being processed, select processing not handling root assignment and removal of unselected properties. Supplied a range of pre-built operator functions for index and value matching.

2023-05-14 v0.7.3 Added unit tests. Corrected issues with: copy not adding id to objects that have no schema, selecting objects of multiple class types at the same time, select support being dropped during an earlier upgrade, sort as function. Updated documentation.

2023-05-14 v0.7.2 Updated documentation.

2023-05-13 v0.7.1 Updated documentation. Not yet documented, built in full-text indexing and search.

2023-05-13 v0.7.0 Reworked indexing to simplify, improve speed and support full text indexing.

2023-05-06 v0.6.6 Removed test db from Git.

2023-05-05 v0.6.5 Fixed issue with not removing un-indexed primitives.

2023-05-05 v0.6.4 Updated lmdb-query to 1.5.4. Fixed issue with put not awaiting.

2023-05-05 v0.6.3 Updated lmdb-query to 1.5.3.

2023-05-04 0.6.2 Updated dependencies.

2023-05-03 v0.6.1 Addressed issue where properties with non-primitve values were not being indexed.

2023-05-02 v0.6.0 Added support for storing plain arrays as top level database entries, e.g. await db.put(null,[1,2,3])

2023-04-29 v0.5.1 Adjusted defineSchema so that it updates rather than overwrites definition if called a second time. Updated dependency versions. Adjusted unit tests. Fixed issues with copy and move not calling underlying code with this context. Unit test coverage over 90%, moving to BETA.

2023-04-28 v0.5.0 Added getSchema to exports. Enhanced unit tests.

2023-04-28 v0.4.3 Adjusted to use withextensions from lmdb-query. Enhanced documentation.

2023-04-27 v0.4.2 Added support for patch. Simplified withExtensions use. Enhanced documentation.

2023-04-24 v0.4.1 Adjustments to copy and move to ensure correct id assignment. Documentation formatting and typo corrections.

2023-04-24 v0.4.0 copy and move now supported.

2023-04-23 v0.3.1 Fix to fully deleting objects from indexes.

2023-04-23 v0.3.0 Using child databases sometimes caused crashes in a clustered environment. Removed child use and instead using a key prefix of @@. Added ability to query across multiple class types.

2023-04-23 v0.2.0 Moved indexes to a separate child database so that they will not conflict with custom indexes and keys developed by application programmers using this library.

2023-04-23 v0.1.1 Documentation updates. Adjusted so that corrupt indexes do not return values for keys that do not exist.

2023-04-22 v0.1.0 Made API more consistent with lmdb-query.

2023-04-22 v0.0.2 Documentation updates. Addition of defineSchema and UUID generation.

2023-04-21 v0.0.1 Initial public release

License

This software is provided as-is under the MIT license.

Copyright (c) 2023, AnyWhichWay, LLC and Simon Y. Blackwell.