xo is a command-line tool to generate idiomatic code for different languages code based on a database schema or a custom query.

At the moment, xo only supports Go. Support for other languages will come soon.

In schema mode, xo connects to your database and generates code using Go templates. xo works by using database metadata and SQL introspection queries to discover the types and relationships contained within a schema, and applying a standard set of base (or customized) Go templates against the discovered relationships.

Currently, xo can generate types for tables, enums, stored procedures, and custom SQL queries for PostgreSQL, MySQL, Oracle, Microsoft SQL Server, and SQLite3 databases.

Note: While the code generated by xo is production quality, it is not the goal, nor the intention for xo to be a "silver bullet," nor to completely eliminate the manual authoring of SQL / Go code.

In query mode, xo parses your query to generate code from Go templates. It finds related tables in your database to ensure type safety.

The following is a matrix of the feature support for each database:

Install xo in the usual Go way:

$ go install github.com/xo/xo@latest

Note: Go 1.19+ is needed for building xo from source.

The following is a quick overview of using xo on the command-line:

# Make an output directory for generated code.
$ mkdir -p models

# Generate code from your Postgres schema. (Default output folder is models)
$ xo schema postgres://user:pass@host/dbname

# Generate code from a Microsoft SQL schema using a custom template directory (see notes below)
$ mkdir -p mssqlmodels
$ xo schema mssql://user:pass@host/dbname -o mssqlmodels --src custom/templates

# Generate code from a custom SQL query for Postgres
$ xo query postgres://user:pass@host/dbname -M -B -T -2 AuthorResult << ENDSQL
SELECT
  a.name::varchar AS name,
  b.type::integer AS my_type
FROM authors a
  INNER JOIN authortypes b ON a.id = b.author_id
WHERE
  a.id = %%authorID int%%
LIMIT %%limit int%%
ENDSQL

# Build generated code - verify it compiles
$ go build ./models/
$ go build ./mssqlmodels/

The following are xo's command-line commands, arguments, and options:

$ xo --help-long
usage: xo [<flags>] <command> [<args> ...]

Flags:
      --help     Show context-sensitive help (also try --help-long and
                 --help-man).
  -v, --verbose  enable verbose output
      --version  display version and exit

Commands:
  help [<command>...]
    Show help.


  query [<flags>] <DSN>
    Generate code for a database custom query from a template.

    -s, --schema=<name>            database schema name
    -t, --template=go              template type (createdb, dot, go, json, yaml;
                                   default: go)
    -f, --suffix=<ext>             file extension suffix for generated files
                                   (otherwise set by template type)
    -o, --out=models               out path (default: models)
    -a, --append                   enable append mode
    -S, --single=<file>            enable single file output
    -D, --debug                    debug generated code (writes generated code
                                   to disk without post processing)
    -Q, --query=""                 custom database query (uses stdin if not
                                   provided)
    -T, --type=<name>              type name
        --type-comment=""          type comment
    -F, --func=<name>              func name
        --func-comment=""          func comment
    -M, --trim                     enable trimming whitespace
    -B, --strip                    enable stripping type casts
    -1, --one                      enable returning single (only one) result
    -l, --flat                     enable returning unstructured values
    -X, --exec                     enable exec (no introspection performed)
    -I, --interpolate              enable interpolation of embedded params
    -L, --delimiter=%%             delimiter used for embedded params (default:
                                   %%)
    -Z, --fields=<field>           override field names for results
    -U, --allow-nulls              allow result fields with NULL values
    -d, --src=<path>               template source directory
    -2, --go-not-first             disable package comment (ie, not first
                                   generated file)
        --go-int32=int             int32 type (default: int)
        --go-uint32=uint           uint32 type (default: uint)
        --go-pkg=<name>            package name
        --go-tag="" ...            build tags
        --go-import="" ...         package imports
        --go-uuid=<pkg>            uuid type package
        --go-custom=<name>         package name for custom types
        --go-conflict=Val          name conflict suffix (default: Val)
        --go-initialism=<val> ...  add initialism (i.e ID, API, URI)
        --go-esc=none ...          escape fields (none, schema, table, column,
                                   all; default: none)
    -g, --go-field-tag=<tag>       field tag
        --go-context=only          context mode (disable, both, only; default:
                                   only)
        --go-inject=""             insert code into generated file headers
        --go-inject-file=<file>    insert code into generated file headers from
                                   a file
        --go-legacy                enables legacy v1 template funcs
        --go-enum-table-prefix     enables table name prefix to enums
        --json-indent="  "         indent spacing
        --json-ugly                disable indentation

  schema [<flags>] <DSN>
    Generate code for a database schema from a template.

    -s, --schema=<name>            database schema name
    -t, --template=go              template type (createdb, dot, go, json, yaml;
                                   default: go)
    -f, --suffix=<ext>             file extension suffix for generated files
                                   (otherwise set by template type)
    -o, --out=models               out path (default: models)
    -a, --append                   enable append mode
    -S, --single=<file>            enable single file output
    -D, --debug                    debug generated code (writes generated code
                                   to disk without post processing)
    -k, --fk-mode=smart            foreign key resolution mode (smart, parent,
                                   field, key; default: smart)
    -i, --include=<glob> ...       include types (<type>)
    -e, --exclude=<glob> ...       exclude types/fields (<type>[.<field>])
    -j, --use-index-names          use index names as defined in schema for
                                   generated code
    -d, --src=<path>               template source directory
        --createdb-fmt=<path>      fmt command (default:
                                   /home/ken/.npm-global/bin/sql-formatter)
        --createdb-fmt-opts=<opts> ...
                                   fmt options (default: -u, -l={{ . }}, -i=2,
                                   --lines-between-queries=2)
        --createdb-constraint      enable constraint name in output (postgres,
                                   mysql, sqlite3)
        --createdb-escape=none     escape mode (none, types, all; default: none)
        --createdb-engine=""       mysql table engine (default: InnoDB)
        --createdb-trim-comment    trim leading comment from views and procs
                                   (--no-createdb-trim-comment)
        --dot-defaults="" ...      default statements (default: node
                                   [shape=none, margin=0])
        --dot-bold                 bold header row
        --dot-color=""             header color (default: lightblue)
        --dot-row=""               row value template (default: {{ .Name }}: {{
                                   .Type.Type }})
        --dot-direction            enable edge directions
    -2, --go-not-first             disable package comment (ie, not first
                                   generated file)
        --go-int32=int             int32 type (default: int)
        --go-uint32=uint           uint32 type (default: uint)
        --go-pkg=<name>            package name
        --go-tag="" ...            build tags
        --go-import="" ...         package imports
        --go-uuid=<pkg>            uuid type package
        --go-custom=<name>         package name for custom types
        --go-conflict=Val          name conflict suffix (default: Val)
        --go-initialism=<val> ...  add initialism (i.e ID, API, URI)
        --go-esc=none ...          escape fields (none, schema, table, column,
                                   all; default: none)
    -g, --go-field-tag=<tag>       field tag
        --go-context=only          context mode (disable, both, only; default:
                                   only)
        --go-inject=""             insert code into generated file headers
        --go-inject-file=<file>    insert code into generated file headers from
                                   a file
        --go-legacy                enables legacy v1 template funcs
        --go-enum-table-prefix     enables table name prefix to enums
        --json-indent="  "         indent spacing
        --json-ugly                disable indentation
        --postgres-oids            enable postgres OIDs

  dump [<flags>] <out>
    Dump internal templates to path.

    -t, --template=go   template type (createdb, dot, go, json, yaml; default:
                        go)
    -f, --suffix=<ext>  file extension suffix for generated files (otherwise set
                        by template type)

xo provides a set of generic "base" templates for each of the supported databases, but it is understood these templates are not suitable for every organization or every schema out there. As such, you can author your own custom templates, or modify the base templates available in the xo source tree, and use those with xo by a passing a directory path via the --src flag.

For non-trivial schemas, custom templates are the most practical, common, and best way to use xo (see below quickstart and related example).

The following is a quick overview of copying the base templates contained in the xo project's templates/ directory, editing to suit, and using with xo:

# Create a working directory
$ mkdir -p my-tpl

# Dump an embedded template to disk
$ xo dump -t createdb my-tpl

# edit base template files
$ vi my-tpl/*.go.tpl

# see command line options for the template
$ xo schema --src my-tpl --help

# generate a schema using the custom template
$ xo schema --src my-tpl -o models postgres://user:pass@host/db

See the Custom Template example below for more information on adapting the base templates in the xo source tree for use within your own project.

Ideally, custom templates for your project/schema should be stored alongside your project. and generated as part of an automated build pipeline using go generate:

# Add to custom xo command to go generate:
$ tee -a gen.go << END
package mypackage

//go:generate xo postgres://user:pass@host/db -o models --src templates
END

# Run go generate
$ go generate

# Add custom templates and gen.go to project
$ git add templates gen.go && git commit -m 'Adding custom xo templates for models'

Note: via the --template/-t parameter of xo dump you can generate other templates with xo. The default template is the go template.

xo templates are standard Go text templates. Please see the documentation for Go's standard text/template package for information concerning the syntax, logic, and variable use within Go templates.

The contexts (ie, the . identifier in templates) made available to custom templates can be found in templates/types.go (see below table for more information on which file uses which type).

Each language, has its own set of templates for $TYPE and are available in the templates/.

For example, Go has templates/gotpl/schema/foreignkey.xo.go.tpl which defines the template used by xo for generating a function to get the foreign key type in Go. The templates are designed to be Database agnostic, so they are used for both PostgreSQL and Microsoft SQL the same, and all other supported database types. The template is passed a different instance of templates.ForeignKey instance (for each foreign key in a table). To get the Name field in from ForeignKey, the template can use {{ .Data.Name }}, or any other field similarly.

Please see the booktest example for a full end-to-end example for each supported database, showcasing how to use a database schema with xo, and the resulting code generated by xo.

Additionally, please see the northwind and django for a demonstration of running xo against larger schema and against databases from other frameworks. Please note that these examples are works in progress, and may not work properly in all scenarios.

Sometimes you may wish to have the database manage the values of columns instead of having them managed by code generated by xo. As such, when you need xo to ignore fields for a database schema, you can use the -e or --exclude flag. For example, a common use case is to define a table with created_at and/or modified_at timestamps fields, where the database is responsible for setting column values on INSERT and UPDATE, respectively.

Consider the following PostgreSQL schema where a users table has a created_at and modified_at field, where created_at has a default value of now() and where modified_at is updated by a trigger on UPDATE:

CREATE TABLE users (
  id          SERIAL PRIMARY KEY,
  name        text NOT NULL DEFAULT '' UNIQUE,
  created_at  timestamptz   default now(),
  modified_at timestamptz   default now()
);

CREATEOR REPLACE FUNCTION update_modified_column() RETURNS TRIGGER AS $$
BEGIN
    NEW.modified_at= now();
RETURN NEW;
END;
$$language 'plpgsql';

CREATE TRIGGER update_users_modtime BEFORE UPDATE ON users
  FOR EACH ROW EXECUTE PROCEDURE update_modified_column();

We can ensure that these columns are managed by PostgreSQL and not by the application logic but by xo by passing the --exclude or -e flag:

# Ignore special fields
$ xo schema postgres://user:pass@host/db -e users.created_at -e users.modified_at
# or, To ignore these fields in all tables
$ xo schema postgres://user:pass@host/db -e *.created_at -e *.modified_at

Often, a schema has a common layout/pattern, such as every table having a created_at and modified_at field (as in the PostgreSQL schema in the previous example). It is then a common use-case to have a GetMostRecent lookup for each table type, retrieving the most recently modified rows for each table (up to some limit, N).

To accomplish this with xo, we will need to create our own set of custom templates, and then add a GetMostRecent lookup to the .type.go.tpl template.

First, we dump the base xo Go template:

$ mkdir -p my-tpl

$ xo dump my-tpl

We can now modify the templates to suit our specific schema, adding lookups, helpers, or anything else necessary for our schema.

To add a GetMostRecent lookup, we edit our copy of the typedef.xo.go.tpl template:

$ vi templates/gotpl/schema/typedef.xo.go.tpl

And add the following templated GetMostRecent func at the end of the file:

// GetMostRecent{{ $type.Name }} returns n most recent rows from '{{ $table }}',
// ordered by "created_at" in descending order.
func GetMostRecent{{ $type.Name }}(ctx context.Context, db DB, n int) ([]*{{ $type.Name }}, error) {
    const sqlstr = `SELECT ` +
        `{{ $type.Fields "created_at" "modified_at" }}` +
        `FROM {{ $table }} ` +
        `ORDER BY created_at DESC LIMIT $1`

    rows, err := db.QueryContext(ctx, sqlstr, n)
    if err != nil {
        return nil, logerror(err)
    }
    defer rows.Close()

    // load results
    var res []*{{ $type.Name }}
    for rows.Next() {
        {{ $short }} := {{ $type.Name }}{
        {{- if $type.PrimaryKey }}
            _exists: true,
        {{ end -}}
        }
        // scan
        if err := rows.Scan({{ fieldnames $type.Fields (print "&" $short) }}); err != nil {
            return nil, logerror(err)
        }
        res = append(res, &{{ $short }})
    }
    return res, nil
}

We can then use the templates in conjunction with xo to generate our "model" code:

$ xo schema postgres://user:pass@localhost/dbname --src templates/

There will now be a GetMostRecentUsers func defined in models/user.xo.go, which can be used as follows:

db, err := dburl.Open("postgres://user:pass@localhost/dbname")
if err != nil { /* ... */ }

// retrieve 15 most recent items
mostRecentUsers, err := models.GetMostRecentUsers(context.Background(), db, 15)
if err != nil { /* ... */ }
for _, user := range users {
    log.Printf("got user: %+v", user)
}

Please note that the base xo templates do not import any SQL drivers. It is left for the user of xo's generated code to import the actual drivers. For reference, these are the expected drivers to use with the code generated by xo:

Additionally, please see below for usage notes on specific SQL database drivers.

If your schema or custom query contains table or column names that need to be escaped using any of the --escape-* options, you must pass the sql_mode=ansi option to the MySQL driver:

$ xo --escape-all 'mysql://user:pass@host/?parseTime=true&sql_mode=ansi' -o models

And when opening a database connection:

db, err := dburl.Open("mysql://user:pass@host/?parseTime=true&sql_mode=ansi")

Additionally, when working with date/time column types in MySQL, one should pass the parseTime=true option to the MySQL driver:

$ xo schema 'mysql://user:pass@host/dbname?parseTime=true' -o models

And when opening a database connection:

db, err := dburl.Open("mysql://user:pass@host/dbname?parseTime=true")

While not required, one should specify the loc=auto option when using xo with a SQLite3 database:

$ xo schema 'file:mydatabase.sqlite3?loc=auto' -o models

And when opening a database connection:

db, err := dburl.Open("file:mydatabase.sqlite3?loc=auto")

For row inserts xo determines whether the primary key is automatically generated by the DB or must be provided by the application for the table row being inserted. For example a table that has a primary key that is also a foreign key to another table, or a table that has multiple primary keys in a many-to-many link table, it is desired that the application provide the primary key(s) for the insert rather than the DB.

xo will query the schema to determine if the database provides an automatic primary key and if the table does not provide one then it will require that the application provide the primary key for the object passed to the Insert method. Below is information on how the logic works for each database type to determine if the DB automatically provides the PK.

• Checks for a sequence that is owned by the table in question.

• Checks for an autoincrement row in the information_schema for the table in question.

• Checks the SQL that is used to generate the table contains the AUTOINCREMENT keyword.
• Checks that the table was created with the primary key type of INTEGER.

If either of the above conditions are satisfied then the PK is determined to be automatically provided by the DB. For the case of integer PK's when you want to override that the PK be manually provided then you can define the key type as INT instead of INTEGER, for example as in the following many-to-many link table:

  CREATE TABLE site_contacts (
  contact_id	INT NOT NULL,
  site_id	INT NOT NULL,
  PRIMARY KEY(contact_id,siteid),
  FOREIGN KEY(contact_id) REFERENCES contacts (contact_id),
  FOREIGN KEY(site_id) REFERENCES sites (site_id)
)

• Checks for an identity associated with one of the columns for the table in question.

ALWAYS GENERATED types will be parsed as Auto PK types for Oracle.

xo can likely get you 99% "of the way there" on medium or large database schemas and 100% of the way there for small or trivial database schemas. In short, xo is a great launching point for developing standardized packages for standard database abstractions/relationships, and xo's most common use-case is indeed in a code generation pipeline, ala stringer.

xo is NOT designed to be an ORM or to generate an ORM. Instead, xo is designed to vastly reduce the overhead/redundancy of (re-)writing types and funcs for common database queries/relationships -- it is not meant to be a "silver bullet".

xo was originally developed while migrating a large application written in PHP to Go. The schema in use in the original app, while well-designed, had become inconsistent over multiple iterations/generations, mainly due to different naming styles adopted by various developers/database admins over the preceding years. Additionally, some components had been written in different languages (Ruby, Java) and had also accumulated significant drift from the original application and accompanying schema. Simultaneously, a large amount of growth meant that the PHP/Ruby code could no longer efficiently serve the traffic volumes.

In late 2014/early 2015, a decision was made to unify and strip out certain backend services and to fully isolate the API from the original application, allowing the various components to instead speak to a common API layer instead of directly to the database, and to build that service layer in Go.

However, unraveling the old PHP/Ruby/Java code became a large headache, as the code, the database, and the API, all had significant drift -- thus, underlying function names, fields, and API methods no longer coincided with the actual database schema, and were named differently in each language. As such, after a round of standardizing names, dropping cruft, and adding a few relationship changes to the schema, the various codebases were fixed to match the schema changes. After that was determined to be a success, the next target was to rewrite the backend services in Go.

In order to keep a similar and consistent workflow for the developers, the previous code generator (written in PHP and Twig templates) was modified to generate Go code. Additionally, at this time, but tangential to the story, the API definitions were ported from JSON to Protobuf to make use of its code generation abilities as well.

xo is the open source version of that code generation tool, and is the fruits of those development efforts. It is hoped that others will be able to use and expand xo to support other databases -- SQL or otherwise -- and that xo can become a common tool in any Go developer's toolbox.

Part of xo's goals is to avoid writing an ORM, or an ORM-like in Go, and to instead generate static, type-safe, fast, and idiomatic Go code across languages and databases. Additionally, the xo developers are of the opinion that relational databases should have proper, well-designed relationships and all the related definitions should reside within the database schema itself: ie, a "self-documenting" schema. xo is an end to that pursuit.

• dburl - a Go package providing a standard, URL style mechanism for parsing and opening database connection URLs
• usql - a universal command-line interface for SQL databases

The following projects work with similar concepts as xo:

• ModelQ
• sqlgen
• squirrel
• scaneo
• acorn and rootx (read overview here)

• sqlc