hrorm

Hrorm is Java's Hand Rolled Object Relational Mapper.

Source Code - Download - Documentation (Prior) - Javadocs (Prior) - Forums - Feedback

Maven:

<dependency>
    <groupId>org.hrorm</groupId>
    <artifactId>hrorm</artifactId>
    <version>0.15.1</version>
</dependency>

Basics


Hrorm is an Object Relational Mapper (ORM) for Java. Hrorm is a concise, declarative, type-checked library for the creation of Data Access Objects (DAOs). Hrorm requires no external configuration or dependencies.

Hrorm reduces the amount of boilerplate code (in both Java and SQL) you have to write to manage persistence of Java objects to a relational (SQL) data store without inflicting you with XMLosis or annotationitis.

The hrorm DaoBuilder class provides methods to describe the relationship between an object and its backing table. Once your DaoBuilder objects are defined and you have a java.sql.Connection, you can create Dao objects for performing CRUD operations.

Hrorm is opinionated about how Java object models and SQL schemas should be designed. It will not play well with all schemas and objects. If you want freedom, look elsewhere.

Quick Start


Suppose you are working on a program for managing recipes that has a domain model that looks like this.

    class Author {
        Long id;
        String name;
    }

    class Recipe {
        Long id;
        String name;
        Author author;
        List<Ingredient> ingredients;
    }

    class Ingredient {
        Long id;
        String name;
        long amount;
    }

You might ask yourself: how will I persist such a rich, complex model? And you might think ... I know, I'll put everything into a List and call serialize() and write that to a file on disk. How many recipes and simultaneous users are we expecting to support anyway?

But instead, you come up with the following relational schema.

sql schema for recipes showing author recipe and ingredients

Because you are the absolute Empress of Overkill.

Some might think to use Hibernate or Mybatis or something, but no, you are a real glutton for punishment, so you decide to use hrorm and write some code like this, which defines in terms of hrorm DaoBuilder objects the relationship between the database schema and the Java object model.

Disclaimer: Yeah, I assume that your schema has some sequences that are not in the picture above. Also, you should have created getters and setters for all of the fields on your objects or used Lombok or something.

    DaoBuilder<Author> authorDaoBuilder = new DaoBuilder<>("AUTHOR", Author::new)
            .withPrimaryKey("ID", "AUTHOR_SEQUENCE", Author::getId, Author::setId)
            .withStringColumn("NAME", Author::getName, Author::setName);

    DaoBuilder<Ingredient> ingredientDaoBuilder = new DaoBuilder<>("INGREDIENT", Ingredient::new)
            .withPrimaryKey("ID", "INGREDIENT_SEQUENCE", Ingredient::getId, Ingredient::setId)
            .withParentColumn("RECIPE_ID")
            .withStringColumn("NAME", Ingredient::getName, Ingredient::setName)
            .withLongColumn("AMOUNT", Ingredient::getAmount, Ingredient::setAmount);

    DaoBuilder<Recipe> recipeDaoBuilder = new DaoBuilder<>("RECIPE", Recipe::new)
            .withPrimaryKey("ID", "RECIPE_SEQUENCE", Recipe::getId, Recipe::setId)
            .withStringColumn("NAME", Recipe::getName, Recipe::setName)
            .withJoinColumn("AUTHOR_ID", Recipe::getAuthor, Recipe::setAuthor, authorDaoBuilder)
            .withChildren(Recipe::getIngredients, Recipe::setIngredients, ingredientDaoBuilder);

That's pretty painful. Thank God for autocompletion. Once that's done, and assuming you have some way to make a java.sql.Connection object now you can actually make a Dao. Then you can write some code like this.

    Connection connection = // somehow this happened

    // Let's make a new author object!
    Author juliaChild = new Author();
    juliaChild.setName("Julia Child");

    // hrorm makes it easy to persist
    Dao<Author> authorDao = authorDaoBuilder.buildDao(connection);
    authorDao.insert(juliaChild);

    // now for Julia's famous Beef bourguignon
    Recipe beefStew = new Recipe():
    beefStew.setName("Beef Stew");
    beefStew.setAuthor(juliaChild);

    // needs some ingredients
    Ingredient carrots = new Ingredient();
    carrots.setName("Carrots");
    carrots.setAmount(4L);
    Ingredient onions = new Ingredient();
    onions.setName("Onions");
    onions.setAmount(2L);
    Ingredient beef = new Ingredient();
    beef.setName("Cow");
    beef.setAmount(1L);
    Ingredient wine = new Ingredient();
    wine.setName("Red Wine");
    beef.setAmount(10L);

    beefStew.setIngredients(Arrays.asList(carrots, onions, beef, wine));

    // with hrorm, we just persist the recipe, it will take care of
    // the ingredients for us
    Dao<Recipe> recipeDao = recipeDaoBuilder.buildDao(connection);
    recipeDao.insert(beefStew);

    // that was perfectly cromulent

Of course, a Dao can do more than just insert. A Dao can also update, delete, and various forms of select.

And really, that's all there is to hrorm.

Designing with hrorm


Hrorm is opinionated about how entities should be built. It will not work with any object model or schema. However, many of the opinions that hrorm has are good practices to follow, regardless of how database records are translated to and from objects.

Here's the skinny about hrorm entities. (Entities means the things being modeled, either in their Java class representation, or in their database table representation.)

  • Entities should have a numeric primary key. (With limited exceptions.)
  • That primary key should be created from values issued from a database sequence.
  • Entities can contain lists of other entities. In this case, the object containing the other entities is referred to as a parent, and the entities in its list, its children. Parent-child relationships in hrorm are owning relationships. The persistence of children is dependent upon the persistence of parents.
  • Entities can contain references to other entities without ownership, but in this case, the an object has a reference to a single other object. Hrorm refers to these cases as sibling relationships.
  • References between entities are accomplished through their primary keys in the database, but through object references in Java.

Entity models that do not follow the practices above will be difficult or impossible to support using hrorm.

Things hrorm Does Not Do


Hrorm is not an all-singing, all-dancing, all-cooking-a-five-course-meal framework. It's a small library for reducing the amount of time you spend writing select FOO, BAR from BAZ and ResultSet rs = statement.execute(). It does not even attempt to go into territory that many other ORM tools do.

This is only a partial list of all the things hrorm will not do.

  • Any kind of caching.
  • Attempt to minimize database chatter by lazy-loading or object change tracking. Hrorm assumes that if you make a call to a Dao it's because you want to read or write something in the database. (But see here).
  • Transaction management. Transactions are for applications, not for hrorm. (Mostly.)
  • Pollute your code with vexing annotations.
  • Involve itself with any XML or other configuration. Or anything else that cannot support the type information your entities deserve. Hrorm thinks Java is a better language than XML. (Yep.)
  • Take over your application with background threads or expect some kind of container to exist. Hrorm is a library, if you do not call the methods, hrorm won't do anything.
  • Connection pooling.
  • Reflection.
  • Invent its own query language.
  • Expect you to write SQL yourself.
  • Manage integrity constraints to your schema or object model. (Mostly.)
  • Declare checked exceptions in its client facing interfaces.
  • Have left pad as a dependency.
  • Mine cryptocurrencies.

ORM Comparison

As mentioned above, many ORM tools are packed with features. Some of them are even useful. Here's a comparison with some well-known ORM tools.

Hibernate Mybatis hrorm
Packed with features Yes Yes No
Wide community support Yes Yes No
Battle tested in major enterprise deployments Yes Yes No
Has a sweet icon and other branding Yes Yes No
Supports your schema and object model how you want it Maybe Somewhat Absolutely not

Falling Off a Cliff


Sometimes when using an ORM tool, development goes great right until you fall off a cliff. The ORM tool takes care of many details about state tracking, caching, SQL generation, transactions, etc, and everything is just fine. But then you need to do something that your ORM tool does not support. Perhaps not a large thing, perhaps just optimize one particular query or other. And you find, you cannot do it. Yes, you can optimize the query, but there's no way to inject that logic into the framework. And if you just have that query outside the framework, then you cannot integrate it with all the caching, state-tracking, transaction managing stuff your framework provides you.

You just fell off the cliff.

It's like the old joke about the traveler asking the farmer how to get to some village in the area and receiving the reply: "You can't get there from here."

Hrorm is a minimal library, not a maximal framework. It should be possible to integrate custom features you need in your application easily. To make Dao objects, all you need is a vanilla java.sql.Connection. You can freely mix calls to hrorm objects with custom logic for managing transactions or querying the database or anything else as you see fit.

Hrorm will not fall off a cliff. It starts at the bottom, and stays there.