Skip to content

BrainlessLabs/bun

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

bun

Description

Bun is a simple to use C++ Object Relational Mapper (ORM) library and much more (key/value store, JSON conversion and Message Pack conversion). Following is the features of bun:

  • Easy to use
  • Object persistence - You can persist C++ objects directly.
  • Non-intrusive - You do not need to modify your classes to make them persistent
  • Constraint Specification in plain C++
  • Persist Nested Objects
  • C++ EDSL Object Query Language (No SQL Query needed)
  • Compile time EDSL syntax check for type safety - Catch bugs before the execution starts
  • Multiple database support - SQLite, Postgres, MySQL
  • Easy to use embedded key/value store
  • Convert C++ objects to JSON and create C++ objects from JSON.
  • Convert C++ objects to Message Pack and create C++ objects from Message Pack.
  • Lazy iteration over the objects in the database using range based for loop in C++

Bun Page

OpenHub

stable

Whats new

  • Bun 1.4.0 Has support for converting objects to JSON and create objects from JSON. It has the capability to convert objects to Message Pack and construct objects from Message Pack.
  • Bun 1.3 Has support for lazy iteration and ranges based for loop support of objects. The same is supported for the key/value store too.
  • Bun 1.2 has support for embedded key/value stores. But default, the key/value store is based on Unqlite.

Articles

Introduction to bun

CodeProject Article

Usage

Object Persistence

With Bun you can persist C++ objects (POD) directly. The magic happens by registring the C++ structure/class with Bun and you are good to go.

#include "blib/bun/bun.hpp"

namespace test {
  // Class that needs to be persisted
  struct Person {
    std::string name;
    std::string uname;
    int age;
    float height;
  };
}

/// @class Child 
struct Child {
    int cf1;
    Child(const int cf = -1) : cf1(cf) {}
    Child& operator=(const int i) {
        cf1 = i;
        return *this;
    }
};

/// @class Parent
struct Parent {
    int f1;
    std::string f2;
    // Nested object
    Child f3;
    Parent() :f1(-1), f2("-1"), f3(-1) {}
};

// Both should be persistable
SPECIALIZE_BUN_HELPER((Child, cf1));
SPECIALIZE_BUN_HELPER((Parent, f1, f2, f3));

/////////////////////////////////////////////////
/// Generate the database bindings at compile time.
/////////////////////////////////////////////////
SPECIALIZE_BUN_HELPER( (test::Person, name, uname, age, height) );

int main() {
  namespace bun = blib::bun;
  namespace query = blib::bun::query;

  // Connect the db. If the db is not there it will be created.
  // It should include the whole path
  // For SQLite
  //bun::connect( "objects.db" );
  // For PostGres
  bun::connect("postgresql://localhost/postgres?user=postgres&password=postgres");
  // Get the fields of the Person. This will be useful in specifying constraints and also
  // querying the object.
  using PersonFields = query::F<test::Person>;

  // Generate the configuration. By default it does nothing.
  bun::Configuration<test::Person> person_config;
  // This is a unique key constraint that is applied.
  // Constraint are applied globally. They need to be set before the
  // execution of the create schema statement
  // The syntax is Field name = Constraint
  // We can club multiple constraints in the same statement as shown below.
  // There is no need for multiple set's to be called. This is how
  // We can chain different constraints in the same statement.
  person_config.set(PersonFields::name = bun::unique_constraint)
                   (PersonFields::uname = bun::unique_constraint);
  
  // Create the schema. We can create the schema multiple times.
  // If it already exists it will be safely ignored.
  // The constraints are applied to the table.
  // Adding constraints doesn't have an effect if the table already exists.
  bun::createSchema<test::Person>();
  
  // Start transaction
  bun::Transaction t;

  // Create some entries in the database
  for (int i = 1; i < 1000; ++i) {
    // PRef is a reference to the persistent object.
    // PRef keeps the ownership of the memory and releases the memory when it is destroyed.
    // Internally it holds the object in a unique_ptr
    // PRef also has an oid associated with the object
    bun::PRef<test::Person> p = new test::Person{};

    // Assign the members values
    p->age = i + 10;
    p->height = 5.6;
    p->name = fmt::format( "Brainless_{}", i );

    // Persist the object and get an oid for the persisted object.
    const bun::SimpleOID oid = p.persist();

    //Getting the object from db using oid.
    bun::PRef<test::Person> p1( oid );
  }

  // Commit the transaction
  t.commit();

  // To get all the object oids of a particular object.
  // person_oids is a vector of type std::vector<bun::SimpleOID<test::Person>>
  const auto person_oids = bun::getAllOids<test::Person>();

  // To get the objects of a particular type
  // std::vector<bun::Pref<test::Person>>
  const auto person_objs = bun::getAllObjects<test::Person>();

  // EDSL QUERY LANGUAGE ----------------------
  // Powerful EDSL object query syntax that is checked for syntax at compile time.
  // The compilation fails at the compile time with a message "Syntax error in Bun Query"
  using FromPerson = query::From<test::Person>;
  FromPerson fromPerson;
  // Grammar is checked for validity of syntax directly at compile time.
  // Currently only &&, ||, <, <=, >, >=, ==, != are supported.
  // They have their respective meaning.

  // Below is a valid query grammar:
  auto valid_query = PersonFields::age > 10 && PersonFields::name != "Brainless_0";
  std::cout << "Valid Grammar?: " << query::IsValidQuery<decltype(valid_query)>::value << std::endl;

  // Oops + is not a valid grammar
  auto invalid_query = PersonFields::age + 10 && 
  PersonFields::name != "Brainless_0";
  std::cout << "Valid Grammar?: " << 
  query::IsValidQuery<decltype(invalid_query)>::value << std::endl;

  // Now let us execute the query.
  // The where function also checks for the validity of the query, and fails at compile time
  const auto objs = fromPerson.where( valid_query ).where( valid_query ).objects();
  // Can even use following way of query
  // As you see we can join queries 
  const auto q = PersonFields::age > 21 && PersonFields::name == "test";
  const auto objs_again = FromPerson().where( q ).objects();
  const auto objs_again_q = FromPerson().where(
    PersonFields::age > 21 && PersonFields::name == "test"
  ).objects()

  // Not going to compile if you enable the below line. 
  // Will get the "Syntax error in Bun Query" compile time message.
  // const auto objs1 = FromPerson.where( invalid_query ).objects();

  // Check the query generated. It does not give the sql query.
  std::cout << fromPerson.query() << std::endl;

  // Support for Nested object persistence and retrieval
  bun::createSchema<Child>();
  bun::createSchema<Parent>();

  std::cout << "How many objects to insert?" << std::endl;
  int count = 0;
  std::cin >> count;

  for (int i = 0; i < count; ++i) {
    bun::l().info("===============Start===================");
    bun::PRef<Parent> p = new Parent{};

    p->f1 = i;
    p->f2 = i % 2 ? "Delete Me" : "Do not Delete Me";
    p->f3 = 10 * i;

    // Persists the Parent and the Nested Child
    p.persist();
    std::cout << "Added to db: \n" << p.toJson() << std::endl;
    bun::l().info("===============End===================\n");
  }

  std::cout << "Get all objects and show" << std::endl;
  auto parents = bun::getAllObjects<Parent>();

  // Iterate and delete the Parent and the nested Child
  // Here p is a PRef type. We can modify the object and persist 
  // the changes if needed.
  for (auto p : parents) {
    std::cout << p.toJson() << std::endl;
    p.del();
  }

  return 0;
}

Range based iteration and Lazy iteration

Its easy to retrieve values from database and iterate over them. The iteration is lazy, which means that the values are not retrieved all at once but paginated.

    // Iterate the parent with range based for loop
    using FromParents = query::From<Parent>;
    using ParentFields = query::F<Parent>;
    FromParents from_parents;

    // Select the query which you want to execute
    auto parents_where = from_parents.where(ParentFields::f2 == "Delete Me");

    // Fetch all the objects satisfying the query. This is a lazy fetch. It will be fetched
    // only when it is called. And not all the objects are fetched.
    // Here v is a PRef so it can be used to modify and persist the object.
    for(auto v : parents_where) {
        std::cout << v.toJson() << std::endl;
    }

Object conversion to JSON & from JSON and Object conversion to Message Pack & from Message Pack

With Bun you can do the following too:

  • Convert an C++ object to JSON
  • Create an C++ object from JSON
  • Convert and C++ object to Message Pack
  • Create an C++ object from Message Pack

The following code demonstrates that:

#include "blib/bun/bun.hpp"

namespace dbg {
    struct C1 {
        int c1;
        C1() : c1(2) {}
    };

    struct C {
        int c;
        C1 c1;
        C(const int i = 1) : c(i) {}
    };

    struct P {
        std::string p;
        C c;
        P() : p("s1"), c(1) {}
    };
}

SPECIALIZE_BUN_HELPER((dbg::C1, c1));
SPECIALIZE_BUN_HELPER((dbg::C, c, c1));
SPECIALIZE_BUN_HELPER((dbg::P, p, c));

int jsonTest() {
    namespace bun = blib::bun;

    bun::PRef<dbg::P> p = new dbg::P{};
    p->p = "s11";
    p->c.c = 10;

    p->c.c1.c1 = 12;

    bun::PRef<dbg::C> c = new dbg::C;
    c->c = 666;

    // Convert the object to JSON
    const std::string json_string = p.toJson();

    // Construct the new object out of JSON
    bun::PRef<dbg::P> p1;
    p1.fromJson(json_string);
    const auto msgpack = p1.toMesssagepack();

    // Construct another object out of messagepack
    bun::PRef<dbg::P> p2;
    p2.fromMessagepack(p1.toMesssagepack());

    // messagepack to string
    std::string msgpack_string;
    for (auto c : msgpack) {
        msgpack_string.push_back(c);
    }
    std::cout << "1. Original object Object:" << json_string << std::endl;
    std::cout << "2. Object from JSON      :" << p1.toJson() << std::endl;
    std::cout << "3. Object to Messagepack :" << msgpack_string << std::endl;
    std::cout << "4. Object from Messagepck:" << p2.toJson() << std::endl;
    return 1;
}

Key/value store

Bun has a key/value store. Currently the key/value store supported is Unqlite. Following example explains how to use it:

#include "blib/bun/bun.hpp"

/// @fn kvTest
/// @brief A test program for using the key/value store
int kvTest() {
    /// @var db
    /// @brief Create the database.
    /// If the database already exists it opens the database.
    bun::KVDb<> db("kv.db");

    /// @brief put a value in database.
    db.put("test", "test");
    std::string val;

    /// @brief get the value. We need to pass a variable by reference to get the value.
    db.get("test", val);
    std::cout << val << std::endl;
    
    const int size = 10000;
    for (int i = 0; i < size; ++i) {
        const std::string s = fmt::format("Value: {}", i);
        db.put(i, s);
    }

    for (int i = 0; i < size; ++i) {
        std::string val;
        db.get(i, val);
        std::cout << val << std::endl;
    }
    
    return 1;
}

Help Wanted

Considering the work needed to take this library further I will be needing any help available. Help is needed in the following areas.

  1. Enhancements
  2. Fix bugs.
  3. Restructure and cleanup code.
  4. Enhance documentation.
  5. Constructive criticism and feature suggestions.
  6. Write tests.
  7. Use Bun

History

Alpha 1 (16th May 2016)

  • Initial version of the library

Alpha 2 (2nd July 2016)

Implementing the Bun EDSL

Alpha 3 (14th March 2018):

  • Integrated SOCI as the database abstraction layer. This makes the library able to use any SQL database like SQLite, Postgres, MySQL. It mostly supports other databases that SOCI also supports but that's not tested yet.
  • Use of Boost Fusion. The code is much cleaner, fewer preprocessor macros. The code is more debuggable.
  • Support for transaction handling using the Transaction class
  • Better error handling and error logging
  • Added a lot of comments to help users

Alpha 4 (5th March 2018)

  • Support for nested objects
  • SimpleOID now uses boost UUID to generate a unique identifier
  • Additional comments
  • Small performance enhancements

Alpha 5 (19th May 2018)

  • Support for constraint before table creation

Alpha 6 (18th July 2018):

  • Adding key/value functionality to bun

Alpha 7 (11 August 2018):

  • Added range based for loop support for object iteration.
  • Added range based for loop support for key-value store iteration.
  • Both the iterations are lazy iterations.

Alpha 8 (19 October 2018)

  • Added support to create C++ object from JSON string
  • Added support to create Message Pack from C++ object
  • Added support to create C++ object from Message Pack