Persistence layer code generation for golang

Marlow is a code generation tool written in golang designed to create useful constructs that provide an ergonomic API for interacting with a project's data persistence layer while maintaining strong compile time type safety assurance.

Objective & Inspiration

Marlow was created to improve developer velocity on projects written in golang that interact with a data persistence layer, like mysql or postgres. In other web application backend environments, these interfaces are usually provided by an application framework's ORM, like the ActiveRecord library used by Rails.

For web applications leveraging the benefits of using golang, it can be difficult to construct the equivalent abstraction of their database that provides crud operations - especially one that is type-safe and dry. There are several open source projects in the golang ecosystem who's goal is exactly that; gorm, beego, and gorp to name a few. Marlow differs from these other projects in its philosophy; rather than attempt to provide an eloquent orm for your project at runtime, it generates a tailored solution at compile time.

Usage

At its core, marlow simply reads a package's field tags and generates valid golang code. The marlowc executable can be installed & used directly via:

go get -u github.com/marlow/marlowc
marlowc -input=./examples/library/models -stdout=true

For a full list of options supported by the compiler refer to marlowc -help. The command line tool can also be used as the executable target for golang's go generate command using //go:generate comment syntax:

package models

//go:generate marlowc -input=book.go

type Book struct {
  ID       string `marlow="column=id"`
  AuthorID string `marlow="column=author_id"`
}

The generated files will live in the same directory as their source counterparts, with an optional suffix to distinguish them (useful if a project is using make to manage the build pipeline). In general it is encouraged that the generated files are not committed to your project's revision control; the source should always be generated immediately before the rest of the package's source code is compiled.

Generated Code & Field Tag Configuration

The compiler parses the marlow field tag value using the net/url package's parseQuery function. This means that each configuration option supported by marlow would end up in delimited by the ampersand (&) character where the key and value are separated by an equal sign (=). For example, a user record may look like:

package models

type User struct {
	table        string `marlow:"tableName=users"`
	ID           uint   `marlow:"column=id"`
	Name         string `marlow:"column=name"`
	Email        string `marlow:"column=email"`
	SettingsMask uint8  `marlow:"column=settings&bitmask"`
}

In this example, marlow would create the following UserStore interface in the models package:

type UserStore interface {
	UpdateUserID(uint, *UserBlueprint) (int64, error)
	FindUsers(*UserBlueprint) ([]*User, error)
	SelectUserEmails(*UserBlueprint) ([]string, error)
	CreateUsers(...User) (int64, error)
	UpdateUserEmail(string, *UserBlueprint) (int64, error)
	DropUserSettingsMask(uint8, *UserBlueprint) (int64, error)
	UpdateUserName(string, *UserBlueprint) (int64, error)
	SelectUserNames(*UserBlueprint) ([]string, error)
	AddUserSettingsMask(uint8, *UserBlueprint) (int64, error)
	CountUsers(*UserBlueprint) (int, error)
	SelectUserIDs(*UserBlueprint) ([]uint, error)
	UpdateUserSettingsMask(uint8, *UserBlueprint) (int64, error)
	DeleteUsers(*UserBlueprint) (int64, error)
	SelectUserSettingsMasks(*UserBlueprint) ([]uint8, error)
}

For every store that is generated, marlow will create a "blueprint" struct that defines a set of fields to be used for querying against the database. In this example, the UserBlueprint generated for the store above would look like:

type UserBlueprint struct {
	IDRange           []uint
	ID                []uint
	NameLike          []string
	Name              []string
	EmailLike         []string
	Email             []string
	SettingsMaskRange []uint8
	SettingsMask      []uint8
	Inclusive         bool
	Limit             int
	Offset            int
	OrderBy           string
	OrderDirection    string
}

Special table field

If present, marlow will recognize the table field's marlow tag value as a container for developer specified overrides for default marlow assumptions about the table.

All other fields

For every other field found on the source struct, marlow will use the following field tag values:

Generated Coverage & Documentation

While everyone's generated marlow code will likely be unique, the examples/library application includes a comprehensive test suite that demonstrates the features of marlow using 3 models - Author, Book and Genre.

Supported Drivers

The follow is a list of officially support driver.Driver implementations supported by the generated marlow code. To request an additional driver, feel free to open up an issue.