Working with SQL Relations in Go - Part 3

Last time, we refactored our code, and implemented the loading of the relationships for the Post entity. The implementation of this however was rather messy, as we did the equivalentof sweeping everything under the rug. In this post however we will look at what we've done, and see how this could be improved upon.

• The Current Implementation
• Refactoring the Relations
• Abstracting the Loading
• Binding Models
• Implementing the Abstractions
• Cleaning up the Implementations
• Conclusion

The Current Implementation

If we look at the Show handler for the Post entity, you'll see lot's of repetition. First we query the tags, then we query the category, and finally the user. Once we have all the data to hand we just cram it into an anonymous struct and encode it to JSON. It works, but it's not ideal.

I would prefer changing the following things about the current implementation,

1. Placing the relations on the Post struct directly
2. Using an abstraction to handle the loading of the relationships
3. Being able to define the relations of the Post entity in code

So let's go about doing those things.

Refactoring the Relations

All we need to do for this is simply add the necessary fields to our existing Post struct,

// post/post.go
package post

import (
...
    "blogger/category"
    "blogger/user"
)

type Post struct {
    ID         int64  `db:"id"`
    UserID     int64  `db:"user_id"`
    CategoryID int64  `db:"category_id"`
    Title      string `db:"title"`
    Body       string `db:"body"`

    User     *user.User         `db:"-"`
    Category *category.Category `db:"-"`
    Tags     []*Tag             `db:"-"`
}

Then in our handler we can remove the anonymous struct entirely, and set the relations directly on the post we have.

// post/handler.go
...
func (h Handler) Show(w http.ResponseWriter, r *http.Request) {
...
    p.User = u
    p.Category = c
    p.Tags = tt

    w.Header().Set("Content-Type", "application/json; charset=utf-8")
    json.NewEncoder(w).Encode(p)
}

Ok, so this is the first issue solve, now let's look at implementing an abstraction for handling the loading of the relations.

Abstracting the Loading

Let's breakdown what we want this abstraction to do when it comes to loading the models,

1. We want to tell the abstraction what column to query against
2. We want to tell the abstraction which values to use when querying
3. We want to tell the abstraction how to bind the loaded relations to the given model

So let's use the Tag relation as an example, currently we do the following,

tt, err := h.TagStore.Get(query.Where("post_id", "=", p.ID))

with our abstraction we would tell it to use the post_id column, and give it p.ID, as the value to use. But how would we go about telling the abstraction how to bind the model to the Post struct we have?

This is where our previously defined model.Model interface comes to play. With this, and a callback function we should be able to achieve what we desire. So, let's start writing out a Loader interface.

// model/model.go
...
type LoaderFunc func(int, Model)

type Loader interface {
    Load(string, []interface{}, LoaderFunc) error
}

The idea behind this abstraction is that we will call Load on the underlying implementation and pass it the column, a slice of values, and a function that will actually do the loading of the relation to the appropriate model. With our Tag example it would look something like this,

load := func(_ int, model.Model) {
    // bind the relation
}

err := h.TagStore.Load("post_id", []interface{}{p.ID}, load)

Before we proceed with implementing this on the TagStore however I should note a few things. First, why does the LoaderFunc accept an int as the first parameter? This will allow flexibility when loading in a relationship to multiple models, let's assume we have a list of posts, and we want to load in all of the tags for each post. Well, with this interface we could do it like so,

pp, _ := h.Store.All()
ids := make([]interface{}, 0, len(pp))

for _, p := range pp {
    ids = append(ids, p.ID)
}

load := func(i int, m model.Model) {
    pp[i].Tags = append(pp[i].Tags, m)
}

h.TagStore.Load("post_id", ids, load)

The next thing I should point out, is that the LoaderFunc also expects model.Model as the second parameter. So, in order to actually use this abstraction we need to make sure that each model being loaded actually implements model.Model.

And now the final thing to mention, is that we don't actually have a nice way of binding models to each other via the LoaderFunc callback. So let's look at how we can achieve this.

Binding Models

All of the models in our blogging application are somewhat similar, and can be reasonably represented via the model.Model interface. Right now however, we don't have a way of binding models to each other. What do I mean by binding, and how does this differ from loading a model?

Loading a model, in this context, is what I mean once we have the models retrieved from the database. Binding however, is what will occur when we set the model as a relation to another model. So, with this in mind let's implement another interface to handle the binding of models.

// model/model.go
type Binder interface {
    Bind(...Model)
}

Let's also embed this interface into the model.Model interface itself too. I'll come back to why this was made a separate interface later on.

// model/model.go
type Model Interface {
    Binder
...

Looking back at our Tag example from earlier, we should be able to do the following, providing that the post.Tag, and post.Post models both implement the model.Model interface.

load := func(_ int, m model.Model) {
    p.Bind(m)
}

h.TagStore.Load("post_id", []interface{}{p.ID}, load)

So, let's go about implementing this for the Tag entity for the time being.

Implementing the Abstractions

Our implementation of the model.Loader for our post.TagStore will look something like this,

// post/tag.go
...
import (
    "blogger/model"
...
)
...
var (
    _ model.Moadl  = (*Tag)(nil)
    _ model.Loader = (*TagStore)(nil)

    tagTable = "post_tags"
)
...
func (s TagStore) Load(key string, vals []interface{}, load model.LoaderFunc) error {
    tt, err := s.All(query.Where(key, "IN", vals...))

    if err != nil {
        return err
    }

    for i := range vals {
        for _, t := range tt {
            load(i, t)
        }
    }
    return nil
}

In order for this to work however we need to make sure that post.Tag implements the model.Model interface so we can actually pass it to the given callback. So let's do that next,

// post/tag.go
...
func (t *Tag) Bind(mm ...model.Model) {}

func (t *Tag) Primary() (string, int64) {
    if t == nil {
        return "id", 0
    }
    return "id", t.ID
}

func (t *Tag) IsZero() bool {
    return t == nil || t.ID == 0 && t.PostID == 0 && t.Name == ""
}

func (t *Tag) Values() map[string]interface{} {
    if t == nil {
        return map[string]interface{}{}
    }
    return map[string]interface{}{
        "post_id": t.PostID,
        "name":    t.Name,
    }
}

Next we need to make sure that our post.Post model implements model.Model too, so we can make use of the bind functionality.

// post/post.go
func (p *Post) Bind(mm ...model.Model) {
    if p == nil {
        return
    }

    for _, m := range mm {
        switch m.(type) {
        case *user.User:
            p.User = m.(*user.User)
        case *category.Category:
            p.Category = m.(*category.Category)
        case *Tag:
            p.Tags = append(p.Tags, m.(*Tag))
        }
    }
}

func (p *Post) Primary() (string, int64) {
    if t == nil {
        return "id", 0
    }
    return "id", p.ID
}

func (p *Post) Values() map[string]interface{} {
    if p == nil {
        return map[string]interface{}{}
    }
    return map[string]interface{}{
        "user_id":     p.UserID,
        "category_id": p.CategoryID,
        "title":       p.Title,
        "body":        p.Body,
    }
}

You'll notice that in the implementation of Bind we do a type assert on the given model to see if its one of the relations a Post could have.

Now in our handler.Show method we should be able to do the following,

// post/handler.go
...
p, err := h.Store.Get(query.Where("id", "=", id))

if err != nil {
    // handle error
}

load := func(_ int, m model.Model) {
    p.Bind(m)
}

h.TagStore.Load("post_id", []interface{}{p.ID}, load))
...

Let's assume we go ahead and implement the model.Loader interface for user.Store, and category.Store, then we could refactor our handler to be something like this,

// post/handler.go
...
p, err := h.Store.Get(query.Where("id", "=", id))

if err != nil {
    // handle error
}

load := func(_ int, m model.Model) {
    p.Bind(m)
}

h.TagStore.Load("post_id", []interface{}{p.ID}, load)
h.UserStore.Load("id", []interface{}{p.UserID}, load)
h.CategoryStore.Load("id", []interface{}{p.CategoryID}, load)
...

This is better than what we had before, but still not very ideal. There's a few things I'd like to change about this,

1. Dynamically extract the values from the given model based on the key we want
2. Specify how we want the binding to happen based on the foreign key and primary key of the models

So let's go about doing both of these things.

Cleaning up the Implementations

It would be nice if we could simply extract the values we want from the model based on the key of that value, instead of having to clumsily write out []interface{}{p.ID}. Thankfully we know that our models implement the model.Model interface, which gives us access to the Primary and Values methods, so let's set about doing this.

// model/model.go
...
func MapKey(key string, mm ...Model) []interface{} {
    vals := make([]interface{}, 0, len(mm))

    for _, m := range mm {
        if col, val := m.Primary(); key == col {
            vals = append(vals, val)
            continue
        }

        if val, ok := m.Values()[key]; ok {
            vals = append(vals, val)
        }
    }
    return vals
}

Note: We check the given key against whatever is returned from Primary first, since the primary key is omitted from the map returned from Values.

With the model.MapKey function in place we can now specify the values we want from the given models, based off the value's name like so,

h.TagStore.Load("post_id", model.MapKey("id", p), load)

Now let's look at specifying how we want the binding to happen based on the foreign key, and the primary key of the models. If we look at what's happening during loading, we can safely assume it's going to be the same across all of our models. Because of this, it would be smart to have a generic function to handle the binding of related models. Let's see how this could be done.

// model/model.go
...
func getInt64(key string, m Model) int64 {
    if col, val := m.Primary(); key == col {
        return val
    }
    i, _ := m.Values()[key].(int64)
    return i
}

func Bind(a, b string, mm ...Model) LoaderFunc {
    return func(i int, r Model) {
        if i > len(mm) || len(mm) == 0 {
            return
        }

        m := mm[i]
        if getInt64(a, m) == getInt64(b, m) {
            m.Bind(r)
        }
    }
}

There's quite a bit going on here so let me explain. First, we implement a getInt64 function that will return the underlying int64 type of the given key. This is working off the assumption that foreign, and primary keys are both 64-bit integers.

Next, we implement the Bind function itself, which returns a LoaderFunc. The Bind function takes two strings, followed by a variadic list of Model interfaces. The two strings a, and b, represent the two keys we want to pull from the model on which binding will occur, and the model which will be bound respectively.

Within the returned LoaderFunc we first check to see if the model exists in the variadic list based on the given index. Then we actually do the comparison of keys. Here we assume that the keys being compared are both int64, and if they match then we bind the model r to the model m.

This seems confusing at first, so let's demonstrate what this would look like with our Tag example.

h.TagStore("post_id", model.MapKey("id", p), model.Bind("id", "post_id", p))

And if we were to expand the model.Bind function call it would look something like this,

h.TagStore("post_id", model.MapKey("id", p), func(i int, r model.Model) {
    if i > len(mm) || len(mm) == 0 {
        return
    }

    m := mm[i]
    if getInt64("id", m) == getInt64("post_id", r) {
        m.Bind(r)
    }
})

So with these new functions in place let's look at our relationship loading now.

h.TagStore.Load("post_id", model.MapKey("id", p), model.Bind("id", "post_id", p))
h.UserStore.Load("id", model.MapKey("user_id", p), model.Bind("user_id", "id", p)))
h.CategoryStore.Load("id", model.MapKey("category_id", p), model.Bind("category_id", "id", p))

We're slowly improving, but I'm still not satisfied with what we have.

Conclusion

We managed to make quite a few improvements here when it comes to loading in the relationships for a model. However I still think some things could be improved upon, specifically the defining the entity relationships themselves in the code.

In the next post we will explore how we can go about utilising more function callbacks to achieve this.

Note: I know in the previous post I said we would look at defining relationships idiomatically in code, however this post is already rather long.

I know that some of the things mentioned here perhaps seem like they were plucked out of thin air. But I do intend on writing a summary post at the end of this series explaining how I came about these implementations, and why I felt the need on writing this series in the first place.