All notable changes to this project will be documented in this file.

• the contents of jinja blocks are now indented if the block wraps onto multiple rows (#403). This is now the proper sqlfmt style:

  select
      some_field,
      {% for some_item in some_sequence %}
          some_function({{ some_item }}){% if not loop.last %}, {% endif %}
      {% endfor %}

  While in this simple example the new style makes it less clear that some_field and some_function are at the same SQL depth, the formatting of complex files with nested jinja blocks is much improved. For example:

  {%- for col in cols -%}
      {%- if col.column.lower() not in remove | map(
          "lower"
      ) and col.column.lower() not in exclude | map("lower") -%}
          {% do include_cols.append(col) %}
      {%- endif %}
  {%- endfor %}

  See also this discussion. Thank you @dave-connors-3 and @alrocar!

• sqlfmt now supports all Postgres frame clauses, not just those that start with rows between. (#404)

• fixed a bug where format-off tokens could cause sqlfmt to raise a bracket mismatch error (#395 - thank you, @AndrewLaneAtPowerSchool!).

• sqlfmt now defaults to reading and writing files using the utf-8 encoding. Previously, we used Python's default behavior of using the encoding from the host machine's locale. However, as utf-8 becomes a de-facto standard, this was causing issues for some Windows users, whose locale was set to use older encodings. You can use the --encoding option to specify a different encoding. Setting encoding to inherit, e.g., sqlfmt --encoding inherit foo.sql will revert to the old behavior of using the host's locale. sqlfmt will detect and preserve a UTF BOM if it is present. If you specify --encoding utf-8-sig, sqlfmt will always write a UTF-8 BOM in the formatted file. (#350, #381, #383 - thank you @profesia-company, @cmcnicoll, @aersam, and @ryanmeekins!)

• sqlfmt no longer merges lines that contain comments, unless the position of those comments can be preserved (#348 - thank you, @rileyschack and @IanEdington!). Accordingly, comments that are inline will stay inline, even if they are too long to fit.
• sqlfmt no longer merges together lines containing multiline jinja blocks unless those lines start with an operator or comma (#365 - thank you, @gavlt!).
• fixed a bug where adding a jinja end tag (e.g., {% endif %}) to a line could cause bad formatting of everything on that line

• adds support for ARM-based platforms using Docker.

• added a Dockerfile for running sqlfmt in a container. New versions of sqlfmt will include Docker builds pushed to the GitHub Container Registry (thank you @ysmilda!).

• sqlfmt now removes extra blank lines (#249 - thank you, @nfcampos!). Basically, no more than 1 blank line inside queries or blocks; no more than 2 between queries or blocks.
• sqlfmt now supports create <object> ... clone statements (#313).
• sqlfmt will now format all files that end with *.sql and *.sql.jinja, even those with other dots in their filenames (#354 - thank you @ysmilda!).
• fixed a bug where {% call %} blocks with arguments like {% call(foo) bar(baz) %} would cause a parsing error (#353 - thank you @IgnorantWalking!).
• sqlfmt now supports bun placeholders (#356 - thank you @ysmilda!)

• by default, sqlfmt now runs an additional safety check that parses the formatted output to ensure it contains all of the same content as the raw input. This incurs a slight (~20%) performance penalty. To bypass this safety check, you can use the command line option --fast, the corresponding TOML or environment variable config, or pass Mode(fast=True) to any API method. The safety check is automatically bypassed if sqlfmt is run with the --check or --diff options. If the safety check fails, the CLI will include an error in the report, and the format_string API will raise a SqlfmtEquivalenceError, which is a subclass of SqlfmtError.

• fixed a bug where very long lines could raise RecursionError (#343 - thank you @kcem-flyr!).

• fixed a bug where nested {% set %} and {% call %} blocks would cause a parsing error (#338 - thank you @AndrewLane!).

• sqlfmt now supports is [not] distinct from as a word operator (#327 - thank you @IgnorantWalking, @kadekillary!).
• fixed a bug where jinja {% call %} blocks that called a macro that wasn't statement caused a parsing error (#335 - thank you @AndrewLane!).

• sqlfmt runs finish in 20% less time due to algorithmic improvements

• sqlfmt now supports {% materialization ... %} and {% call statement(...) %} blocks (#309).
• sqlfmt now resets the SQL depth of a query after encountering an {% endmacro %}, {% endtest %}, {% endcall %}, or {% endmaterialization %} tag.
• sqlfmt now supports create warehouse and alter warehouse statements (#312, #299).
• sqlfmt now supports alter function and drop function statements (#310, #311), and Snowflake's create external function statements (#322).
• sqlfmt better supports numeric constants (number literals), including those using scientific notation (e.g., 1.5e-9) and the unary + or - operators (e.g., +3), and is now smarter about when the - symbol is the unary negative or binary subtraction operator. (#321 - thank you @liaopeiyuan!).
• fixed a bug where we added extra whitespace to the end of empty comment lines (#319 - thank you @eherde!).
• fixed an bug where wrapping unsupported DDL in jinja would cause a parsing error (#326 - thank you @ETG-msimons!). Also improved parsing of unsupported DDL and made false positives less likely.
• fixed a bug where we could have unsafely run black against jinja that contained Python keywords and their safe alternatives (e.g., return(return_())).
• fixed a bug where we deleted some extra whitespace lines (and in very rare cases, nonblank lines)
• fixed a bug where Python recursion limits could cause incorrect formatting in rare cases

• sqlfmt now supports delete statements and the associated keywords using and returning (#281).
• sqlfmt now supports grant and revoke statements and all associated keywords (#283).
• sqlfmt now supports create function statements and all associated keywords (#282).
• sqlfmt now supports the explain keyword (#280).
• sqlfmt now supports BigQuery typed table and struct definitions and literals, like table<a int64, b bytes(5), c string>.
• sqlfmt now supports variables like $foo as ordinary identifiers.

• sqlfmt is now tested against Python 3.11 (#242). Previous versions of sqlfmt are also compatible. with Python 3.11. When installed in 3.11, sqlfmt no longer requires the tomli dependency.

• DDL and DML statements (create, insert, grant, etc.) will no longer be formatted (#243). These statements were never supported by sqlfmt, and the existing algorithm produced bad formatting. Support for DDL and DML statements will be gradually added back in in future versions. For more information, see the tracking issue for DDL support.
• BigQuery typed array literals like array<float64>[1, 2] are now supported, and spaces will no longer be inserted around < and > (#212).
• SparkSQL-specific keywords tablesample, cluster by, distribute by, sort by, and lateral view are now supported by the polyglot dialect (#264).
• pivot and unpivot are now supported as word operators, and will have a space between the keyword and the following parentheses.
• values is now supported as an unterminated keyword; tuples of values will be indented from the values keyword if they span more than one line (#263).

• Any CLI option can now be configured using environment variables. Variable names are prefixed by SQLFMT and are the SHOUTING_CASE spelling of the options. For example, sqlfmt . --line-length 100 is equivalent to SQLFMT_LINE_LENGTH=100 sqlfmt . (#251).

• The README has been shortened and now links to docs.sqlfmt.com.

• The files argument of api.run is now a Collection[pathlib.Path] that represents an exact collection of files to be formatted, instead of a list of paths to search for files. Use api.get_matching_paths(paths, mode) to return the set of exact paths expected by api.run.

• sqlfmt will now display a progress bar for long runs (#231). You can disable this with the --no-progressbar option.
• api.run now accepts an optional callback argument, which must be a Callable[[Awaitable[SqlFormatResult]], None]. Unless the --single-process option is used, the callback is executed after each file is formatted.
• sqlfmt can now be called as a python module, with python -m sqlfmt.

• adds more granularity to operator precedence and will merge lines more aggressively that start with high-precedence operators (#200).
• improves the formatting of between ... and ..., especially in situations where the source includes a line break (#207).
• improves the consistency of formatting long chains of operators that include parentheses (#214).
• fixes a bug that caused unnecessary copying of the cache when using multiprocessing. Large projects should see dramatically faster (near-instant) runs once the cache is warm.
• fixes a bug that could cause lines with long jinja tags to be one character over the line length limit, and could result in unstable formatting (#237 - thank you @nfcampos!).
• fixes a bug that formatted array literals like they were indexing operations (#235 - thank you @nfcampos!).

• sqlfmt now supports the psycopg placeholders %s and %(name)s (#198 - thank you @snorkysnark!).

• sqlfmt now standardizes whitespace inside word tokens (#201).
• using is now treated as a word operator. It gets a space before its brackets and merging with surrounding lines is now much improved (#218 - thank you @nfcampos!).
• within group and filter are now treated like over, and the formatting of those aggregate clauses is improved (#205).

• sqlfmt now supports ClickHouse. When run with the --dialect clickhouse option, sqlfmt will not lowercase names that could be case-sensitive in ClickHouse, like function names, aliases, etc. (#193 - thank you @Shlomixg!).

• formatting for chained boolean operators with complex expressions is now significantly improved (#189 - thank you @Rainymood!).
• formatting for array indexing is now significantly improved (#209) and sqlfmt no longer inserts spaces between the offset() function and its brackets.
• set operators (like union) are now formatted differently. They must be on their own line, and will not cause subsequent blocks to be indented (#188 - thank you @Rainymood!).
• select * except (...) syntax is now explicitly supported, and formatting is improved. Support added for BigQuery and DuckDB star options: except, exclude, replace.
• sqlfmt no longer inserts spaces between nested or repeated brackets, like (()) or ()[].
• a bug causing unstable formatting with long/multiline jinja tags has been fixed (#175).

• jinjafmt is now able to format jinja that contains functions and variables that are reserved python words (e.g., return, except, from) (#177, #155), and ~, the jinja string concatenation operator (#182)
• adds a new command-line option to reset the sqlfmt cache (#184)

• fixes issue where jinjafmt would insert a trailing comma into multiline macro definitions, causing dbt compiling errors (#156)
• fixes issue causing unstable formatting of multiline jinja tags when black is unable to parse the tag (#176)
• fixes issue for developers where pre-commit hooks would not install

• sqlfmt_primer now runs against forked (formatted) repos to make changes easier to detect

• sqlfmt is now more conservative about preserving whitespace around jinja expressions when we remove newlines (#162, #165 - thank you @rcaddell and @rjay98!)
• jinja blocks are now dedented before line merging, instead of after. This results in small changes to formatted output in some cases where jinja blocks are used
• fixes an issue where jinja else and elif statements could cause unstable formatting. May impact whitespace for the tokens following {% else %} and {% elif %} statements

• api.run now accepts files as a List[pathlib.Path] instead of a List[str]

• any command line option can now be set in a pyproject.toml file. See README for more information (#90)
• sqlfmt now accepts an --exclude option to specify a glob of files to exclude from formatting (#131)

• adds support for snapshot blocks, so the contents of those blocks are now properly formatted (#137)
• fixes issue causing unstable formatting of multiline jinja tags when black is not installed (#138)
• fixes formatting of semicolons and semicolon-delimited queries (#132)

• adds support for numbered field references (e.g., $1) and snowflake stages (@my_stage) as identifiers
• do not split lines before the between operator's and keyword (#124 - thank you @WestComputing!)

• formats the contents of jinja tags (the code between the curlies) using black, the Python formatter. If black is not already installed, you can use this feature by re-installing sqlfmt with the jinjafmt extra (pipx install sqlfmt[jinjafmt]). If black is installed, but you do not want to use this feature, you can disable it with the command-line option --no-jinjafmt
• no longer inserts spaces around colons (#103 - thank you @noel!)

• adds "cross join" to list of supported join types. No longer merges the "cross" keyword with the previous statement (#110 - thank you @rdeese!)
• adds support for every valid operator in postgresql, even the weird ones, like @>, ||/, ?-| (#105)

• removes an unnecessary dependency on black that broke installation (#98 - thank you @ljhopkins2!)

• adds an option, --single-process, to force single-processing, even when formatting many files

• when formatting multiple files, uses multiprocessing for ~3x faster throughput

• preserves leading and trailing blank lines when merging lines with content
• no longer prints whitespace on blank lines

• adds special support for jinja, with new formatting rules for jinja statements, expressions, and blocks
• safely standardizes whitespace around jinja statements and expressions
• merges lines within and across jinja blocks while balancing start/end statements
• jinja block tags can no longr be indented farther than any of their contents

• developers can now easily profile sqlfmt performance (after installing the sqlfmt_primer extra) with make profiling

• no longer fails with a parsing error if "end" is used as a name in the query

• refactors lexing using typical callable architecture for more flexibility
• adds new token types for jinja statements and blocks
• refactors calculation of node and line depth to include jinja blocks
• adds some caching to line properties for performance enhancements

• refactors comment parsing for improved formatting of comments and merging around comments
• standardizes comments and splits long comments onto multiple lines if necessary
• improves splitting and merging of lines with operators (like "+", "as", "on", etc.)
• improves formatting of queries that use leading commas
• improves merging of statements with chained brackets ("( something ) + ( something_else )" )

• adds a simple cache and skips formatting files that have not changed since last successful run
• improves the welcome message displayed when running sqlfmt with no arguments

• supports all postgres and bigquery string literals and quoted identifiers, including triple quotes, escaped quotes, dollar-delimited, etc.
• no longer fails with a parsing error when encountering a semicolon
• properly delineates between "*" as "all fields" and as the multiplication operator

• refactors line splitting algorithm and creating a line from nodes; provides 3x speedup of sqlfmt (now formats roughly 100 files/sec)

• refactored lexer for better parsing of tokens on multiple lines

• can format text through stdin by passing - as the files argument
• supports --quiet option
• supports -- fmt: off and -- fmt: on comments in sql files
• supports more select keywords, like window and qualify
• supports back-ticks for quoting relation names
• supports MySQL-style comments (# comment)
• adds a new cli tool, sqlfmt_primer, to run sqlfmt against known OSS projects to gauge changes

• fixes parsing of jinja tags (use lazy regex so we don't match multiple tags at once)
• fixes issue with whitespace around jinja tags
• fixes capitalization of word operators (on, and, etc.)
• fixes parsing error caused by comments without leading spaces

• initial release
• discovers .sql and .sql.jinja files
• formats the files it discovers
• supports --check and --diff options
• supports --no-color