SQL Formatter

How it actually works

Reformatting is done by sql-formatter 13.0.0 from Cloudflare's CDN, the same tokenizer-based library used in editor extensions, so keywords are recognised rather than guessed from a word list, and casing is normalised to upper while identifiers and string contents are left exactly as typed. Indentation is a choice of spaces or tabs at 2, 4 or 8 wide, and stacked statements are separated by a single blank line.

Because the library renamed its configuration options along the way, the tool asks the loaded build which vocabulary it understands once and then uses that, so formatting keeps working whichever release the content delivery network serves. If the script never arrives - offline, or a blocked CDN - the tool says so plainly instead of showing an unhandled error.

Minify was rewritten. Collapsing all whitespace corrupted real queries: spaces inside a string literal were squeezed, so a value of 'a = b' became 'a=b', and deleting a line comment without keeping its newline turned the rest of the script into part of that comment. Literals are now lifted out untouched, optimizer hints are kept, comments are dropped with their line boundary preserved, and only the code between them is compacted.

What that means for your results

Formatting is semantics-preserving by design. The tokenizer knows that text inside single quotes is data, which is why a formatted query can be pasted back into production without auditing every character.

The minified output is still valid SQL with the same string values as the input, verified against cases containing quoted operators, multi-line literals and hints.

Where it falls short — honestly

The dialect is standard SQL: there is no selector for MySQL, PostgreSQL, T-SQL or Oracle variants, so vendor-specific syntax such as backquoted identifiers, LIMIT with OFFSET clauses or window functions may be laid out imperfectly, and a plugin's odd grammar can confuse the tokenizer. Nothing is executed or validated against a schema, so a formatted query is not necessarily a correct one.

Minify removes comments other than optimizer hints, which is the point but makes the output a poor place to keep annotations.

A real end-to-end workflow

Use it when a query has grown: paste the one-line monster from a log, format, and the joins, predicates and grouping become readable enough to review. Then set indentation to match your repository so the formatted query diffs cleanly when it goes back into source.

For a query embedded in a shell command or a script tag, minify it, and check the quoted values first. If the SQL came from an application config in JSON, JSON Validator is the right sibling tool; for regular expressions against query logs, use Regex Tester.