Trino / Presto SQL Formatter

An ANSI-strict dialect

Trino follows the SQL standard closely, and its quoting rules are simpler than in most warehouses:

• '...' is always a string, and an embedded quote is doubled: 'it''s'. Backslash has no special meaning, so 'C:\data\' is a complete literal.
• "..." is always an identifier. "order id" or "hive"."sales"."orders" stay exactly as written.
• Backticks are not valid at all, so a query carried over from Spark or Hive with hive.sales fails with a “not valid Trino / Presto syntax” error.
• U&'caf\00E9' Unicode strings and X'65683F' binary literals are read as single tokens.
• ? is the parameter marker for PREPARE ... FROM and EXECUTE ... USING; both statements format normally.

This matters for formatting. If you picked Spark SQL by mistake, "EU" would look like a string and backticks would be accepted, and a query the Trino parser later rejects could pass through looking fine. The reverse catches a real bug: in Trino mode region = "EU" formats cleanly, but Trino will read "EU" as a column and report that it cannot be resolved.

Federated queries, arrays and lambdas

Three-part names like hive.sales.orders joined to postgresql.crm.customers are kept whole, so a federated join reads like any other. Complex types are recognised: ARRAY['sku', 'qty'], ROW(id BIGINT, amount DOUBLE) and the type names inside CAST follow the keyword-case setting, while map(...), element_at and cardinality are treated as ordinary functions. Long nested calls are broken over several lines with their arguments indented.

Lambda expressions keep their arrows: transform(array_agg(amount), x -> round(x * 1.2, 2)) and reduce(amounts, 0, (s, x) -> s + x, s -> s) stay on one line when they fit.

CREATE TABLE ... WITH (partitioning = ARRAY[...], format = 'PARQUET') AS SELECT puts the table properties in an indented block. OFFSET ... FETCH FIRST n ROWS ONLY, WITH RECURSIVE, EXPLAIN ANALYZE, SHOW and DESCRIBE are all supported.

Layout quirks to expect

• CROSS JOIN UNNEST(o.skus) WITH ORDINALITY AS t(sku, line_no) is split after WITH, as if it began a CTE. The result is still valid; join the two lines by hand if you prefer.
• An aggregate filter, sum(amount) FILTER (WHERE status = 'paid'), is expanded into a small WHERE block inside the brackets.
• In MATCH_RECOGNIZE, quantifiers get a space (b+ becomes b +), which Trino’s parser ignores.
• Column names that are also Trino keywords, such as path, get upper-cased along with real keywords.
• :name placeholders are not Trino syntax and get mangled: = :region turns into =: region. Use ? instead.

Like every mode, the formatter checks quotes, comments and brackets precisely, but it is not a full Trino parser; a query can format cleanly and still fail on a missing column or a type mismatch.

Getting readable output for Athena and Presto

Athena engine versions 2 and 3 are built on Presto and Trino, so this mode suits them. Dashboards often store queries as one long line; paste it and format to see the structure. For review-friendly diffs, choose Commas: Leading and AND / OR: Start of line. Use Function case: lower if you want built-ins like approx_percentile and date_trunc to look consistent next to upper-case keywords. More tips are in the SQL style guide.

Examples

select o.order_id, t.sku, t.line_no from hive.sales.orders o cross join unnest(o.skus) with ordinality as t(sku, line_no) where o.order_date >= date '2026-01-01'

SELECT
  o.order_id,
  t.sku,
  t.line_no
FROM
  hive.sales.orders o
  CROSS JOIN UNNEST (o.skus)
WITH
  ORDINALITY AS t (sku, line_no)
WHERE
  o.order_date >= date '2026-01-01'

select c.region, approx_percentile(o.amount, 0.95) as p95_amount, count_if(o.status = 'refunded') as refunds, transform(array_agg(o.amount), x -> round(x * 1.2, 2)) as gross_amounts from hive.sales.orders o join postgresql.crm.customers c on c.customer_id = o.customer_id where o.order_ts >= current_timestamp - interval '30' day group by c.region

SELECT
  c.region,
  approx_percentile(o.amount, 0.95) AS p95_amount,
  count_if(o.status = 'refunded') AS refunds,
  transform(array_agg(o.amount), x -> round(x * 1.2, 2)) AS gross_amounts
FROM
  hive.sales.orders o
  JOIN postgresql.crm.customers c ON c.customer_id = o.customer_id
WHERE
  o.order_ts >= current_timestamp - INTERVAL '30' day
GROUP BY
  c.region

create table iceberg.analytics.daily_revenue with (partitioning = array['day(order_ts)'], format = 'PARQUET') as select date(order_ts) as day, sum(amount) as revenue from hive.sales.orders group by 1

CREATE TABLE iceberg.analytics.daily_revenue
WITH
  (
    partitioning = ARRAY['day(order_ts)'],
    format = 'PARQUET'
  ) AS
SELECT
  date(order_ts) AS day,
  sum(amount) AS revenue
FROM
  hive.sales.orders
GROUP BY
  1

prepare top_customers from select customer_id, sum(amount) as revenue from hive.sales.orders where region = ? group by customer_id order by revenue desc limit 20;
execute top_customers using 'EU';

PREPARE top_customers
FROM
SELECT
  customer_id,
  sum(amount) AS revenue
FROM
  hive.sales.orders
WHERE
  region = ?
GROUP BY
  customer_id
ORDER BY
  revenue DESC
LIMIT
  20;

EXECUTE top_customers USING 'EU';

Common errors and how to fix them

Related tools

• SQL formatter
• sql formatter/spark
• sql formatter/postgresql
• sql formatter/bigquery
• sql validator
• sql minifier
• JSON to SQL
• sql unbalanced parentheses