shandy-sqlfmt 0.7.0

sqlfmt is an opinionated CLI tool that formats your sql files

sqlfmt

sqlfmt formats your dbt SQL files so you don't have to. It is similar in nature to black, gofmt, and rustfmt (but for SQL). When you use sqlfmt:

1. You never have to mention (or argue about) code style in code reviews again
2. You make it easier to collaborate and solicit contributions from new people
3. You will be able to read your team's code as if you wrote it
4. You can forget about formatting your code, and spend your time on business logic instead

sqlfmt is not configurable, except for line length. It enforces a single style. sqlfmt maintains comments and some extra newlines, but largely ignores all indentation and line breaks in the input file.

sqlfmt is not a linter. It does not parse your code into an AST; it just lexes it and tracks a small subset of tokens that impact formatting. This lets us "do one thing and do it well:" sqlfmt is very fast, and easier to maintain and extend than linters that need a full SQL grammar.

sqlfmt is designed to work with templated SQL files that contain jinja tags and blocks. It formats the code that users look at, and therefore doesn't need to know anything about what happens after the templates are rendered.

For now, sqlfmt only works on select statements (which is all you need if you use sqlfmt with a dbt project). In the future, it will be extended to DDL statements, as well.

Installation

Try it first

Want to test out sqlfmt on a query before you install it? Go to sqlfmt.com to use the interactive, web-based version.

Prerequisites

You will need Python 3.7-3.10 installed. You should use pipx or install into a virtual environment (maybe as a dev-dependency in your project). If you do not know how to install python and use pipx and/or virtual environments, go read about that first.

Install Using pipx (recommended)

pipx install shandy-sqlfmt

To install with the jinjafmt extra (which will also install the Python code formatter, black):

pipx install shandy-sqlfmt[jinjafmt]

Other Installation Options

You should use a virtual environment to isolate sqlfmt's dependencies from others on your system. We recommend poetry (poetry add -D shandy-sqlfmt[jinjafmt] or poetry add -D shandy-sqlfmt), or pipenv (pipenv install -d shandy-sqlfmt[jinjafmt], etc.), but a simple pip install shandy-sqlfmt will also work.

Getting Started

Other prerequisites

sqlfmt is an alpha product and will not always produce the formatted output you might want. It might even break your SQL syntax. It is highly recommended to only run sqlfmt on files in a version control system (like git), so that it is easy for you to revert any changes made by sqlfmt. On your first run, be sure to make a commit before running sqlfmt.

Using sqlfmt

sqlfmt is a command-line tool. It works on any posix terminal and on Windows Powershell. If you have used black, the sqlfmt commands will look familiar.

Note: The $ in the code snippets below indicate that these are commands that can be typed into your terminal. The $ is the prompt symbol in bash; it is equivalent to > on Windows or % on zsh. Do not include the $ in the command you type into your terminal

To list commands and options:

$ sqlfmt --help

If you want to format all .sql and .sql.jinja files in your current working directory (and all nested directories), simply type:

$ sqlfmt .

You can also supply a path to a one or more files or directories as arguments:

$ sqlfmt /path/to/my/dir /path/to/a/file.sql

If you don't want to format the files you have on disk, you can run sqlfmt with the --check option. sqlfmt will exit with code 1 if the files on disk are not properly formatted:

$ sqlfmt --check .
$ sqlfmt --check path/to/my/dir

If you want to print a diff of changes that sqlfmt would make to format a file (but not update the file on disk), you can use the --diff option. --diff also exits with 1 on changes:

$ sqlfmt --diff .

Disabling sqlfmt

If you would like sqlfmt to ignore a file, or part of a file, you can add -- fmt: off and -- fmt: on comments to your code (or # fmt: off on MySQL or BigQuery). sqlfmt will not change any code between those comments; a single -- fmt: off at the top of a file will keep the entire file intact.

The jinjafmt extra

sqlfmt loves properly-formatted jinja, too.

sqlfmt will safely attempt to import the Python code formatter, black. If it is successful (either because sqlfmt was installed with the jinjafmt extra or because black was installed separately in the same environment), it will use black to format the contents of jinja tags. If you do not want sqlfmt to use black to format your jinja, then specify the --no-jinjafmt flag when running sqlfmt.

Installing sqlfmt with the jinjafmt extra will also install black. You can do this with pipx install sqlfmt[jinjafmt] If you want to pin a specific black version, you should specify that separately, as a direct dependency of your project (in your Pipfile, pyproject.toml, etc.).

If sqlfmt was installed without the jinjafmt extra, and black is not otherwise installed, then sqlfmt will not attempt to format the contents of jinja tags, except for enforcing a single space inside each curly.

Configuring sqlfmt using pyproject.toml

Any command-line option for sqlfmt can also be set in a pyproject.toml file, under a [tool.sqlfmt] section header. Options passed at the command line will override the settings in the config file.

sqlfmt will search for the pyproject.toml file using the files passed to it as arguments. It starts in the lowest (most specific) common parent directory to all the files and recurses up to the root directory. It will load settings from the first pyproject.toml file it finds in this search.

Example of a pyproject.toml file to override the default behaviors (run sqlfmt --help for more options):

[tool.sqlfmt]
line_length = 100
check = true

Using sqlfmt with dbt

sqlfmt was built for dbt, so only minimal configuration is required. We recommend excluding your target and dbt_packages directories from formatting. You can do this with the command-line --exclude option, or by setting exclude in your pyproject.toml file

[tool.sqlfmt]
exclude=["target/**/*", "dbt_packages/**/*"]

Using sqlfmt with pre-commit

You can configure pre-commit to run sqlfmt on your repository before you commit changes.

Add the following config to your .pre-commit-config.yaml file:

repos:
  - repo: https://github.com/tconbeer/sqlfmt
    rev: v0.7.0
    hooks:
      - id: sqlfmt
        language_version: python

You should replace rev with the latest available release, but we do suggest pinning to a specific rev, to avoid unexpected formatting changes.

Using sqlfmt with SQLFluff

You can (and should!) use SQLFluff to lint your SQL queries after they are formatted by sqlfmt. However, the two tools do not see eye-to-eye on formatting (by default), so to avoid lint errors, add the following to your .sqlfluff config file:

[sqlfluff]
exclude_rules = L003, L018, L036

[sqlfluff:rules]
max_line_length = 88  # or whatever you set in sqlfmt
capitalisation_policy = lower
extended_capitalisation_policy = lower

[sqlfluff:indentation]
indented_joins = False
indented_using_on = True
template_blocks_indent = False

[sqlfluff:rules:L052]
multiline_newline = True

The sqlfmt style

The only thing you can configure with sqlfmt is the desired line length of the formatted file. You can do this with the --line-length or -l options. The default is 88.

Given the desired line length, sqlfmt has four objectives:

1. Break and indent lines to make the syntactical structure of the code apparent
2. Break lines so they are shorter than the desired line length, if possible
3. Combine lines to use the least possible vertical space, without violating #1 and #2
4. Standardize capitalization (to lowercase) and in-line whitespace

sqlfmt borrows elements from well-accepted styles from other programming languages. It places opening brackets on the same line as preceding function names (like black for python and 1TBS for C). It indents closing brackets to the same depth as the opening bracket (this is extended to statements that must be closed, like case and end).

The sqlfmt style is as simple as possible, with little-to-no special-casing of formatting concerns. While at first blush, this may not create a format that is as "nice" or "expressive" as hand-crafted indentation, over time, as you grow accustomed to the style, formatting becomes transparent and the consistency will allow you to jump between files, projects, and even companies much faster.

Why lowercase?

There are several reasons that sqlfmt insists on lowercase SQL keywords:

1. We believe that SQL is code (this is a surprisingly controversial statement!). Shouting-case keywords perpetuate the myth that SQL isn't "real code", or isn't "modern" and somehow is trapped in the era of low-level imperative languages: BASIC, COBOL, and FORTRAN. The reality is that SQL is an incredibly powerful, declarative, and modern language. It should look like one
2. Syntax highlighting for SQL makes shouting-case keywords redundant; the syntax highlighter in any text editor is going to be more consistent than any manual shout-casing. If you have a SQL query as a string inside of a block of code in another language, you may want to capitalize your keywords; sqlfmt only operates on dedicated SQL (and templated sql) files, so this is not relevant. However, even without syntax highlighting, the hierarchical and consistent indentation provided by sqlfmt provides sufficient visual structure without shout-casing keywords
3. Even among people who like shout-cased keywords, there are disagreements between what gets shout-cased. SELECT, sure, but SUM? AS? OVER? AND? All-lowercase keywords eliminates this potential source of irregularity and disagreement.
4. Research shows that generally, lowercase words are more readable

Why trailing commas?

1. Using trailing commas follows the convention of every other written language and programming language
2. Leading commas require placing the first field name on the same line as select, which can obscure that field
3. SQL query compilation is extremely fast; the "cost" of "last field" errors is very low. Some dialects (e.g., BigQuery) even allow a trailing comma in the final field of a select statement
4. Trailing commas generalize better within select statements (e.g. group by and partition by clauses) and in other kinds of SQL statements (e.g. insert statements)

Examples

sqlfmt will put very short queries on a single line:

SELECT a,
b,
   c
FROM my_table

becomes

select a, b, c from my_table

If a query doesn't fit on a single line, sqlfmt will format the query to make its hierarchy apparent. The main keywords in a select statement are the top nodes in hierarchy. Individual fields are indented a single level; unless all fields fit on the same line as select, they must all be individually split onto their own lines. This is properly formatted code:

with t as (select * from my_schema."my_QUOTED_ table!")
select
    a_long_field_name,
    another_long_field_name,
    (one_field + another_field) as c,
    a_final_field
from t
where one_field < another_field

Note that the main keywords, with, select, from, and where, are indented to the same depth. If their arguments fit on a single line (as in with, from, and where), they stay on that line, with the keyword. However, unless all arguments for a keyword fit on one line, they are all wrapped to their own line, and indented:

with
    a_long_cte_name as (
        select my_field, sum(another_field) from my_schema."my_QUOTED_ table!"
    )
select
    a_long_field_name,
    another_long_field_name,
    (one_field + another_field) as c,
    a_final_field
from a_long_cte_name
where
    one_field < another_field
    and two_field > another_field
    and three_field = another_field

Any expressions wrapped in parentheses are similarly one-lined if possible, and split if they are too long.

This hierarchical indentation scales to arbitrarily complex and nested expressions. Another example of properly formatted code (at line length of 88):

select
    a,
    sum(a) over () as b,
    row_number() over () as c,
    count(case when a is null then 1 end) over (
        partition by user_id, date_trunc('year', performed_at)
    ) as d,
    first_value(
        coalesce(one_field, another_field, yet_another_field) ignore nulls
    ) over (
        partition by user_id
        order by performed_at desc
        rows between unbounded preceding and unbounded following
    ) as e
from my_table

Want more examples? See the tests/data directory, or go to http://sqlfmt.com to see how sqlfmt will format your queries.

Contributing

Setting up Your Dev Environment and Running Tests

1. Install Poetry if you don't have it already. You may also need or want pyenv, make, and gcc. A complete setup from a fresh install of Ubuntu can be found here
2. Clone this repo into a directory (let's call it sqlfmt), then cd sqlfmt
3. Use poetry install -E jinjafmt to install the project (editable) and its dependencies into a new virtual env. To run sqlfmt_primer, you will need to install it (and its dependencies) by specifying it as an extra: poetry install -E jinjafmt -E sqlfmt_primer
4. Use poetry shell to spawn a subshell
5. Type make to run all tests and linters, or run pytest, black, flake8, isort, and mypy individually.

Note: If encountering a JSONDecodeError during poetry install, you will want to clear the poetry cache with poetry cache clear pypi --all, or upgrade to poetry >= 1.12 with poetry self upgrade