Snowflake SQL Formatter

Snowflake syntax the tokenizer knows about

Before any layout happens, PasteKit splits the text using Snowflake’s own quoting rules. That step is what keeps a formatter from wrecking a query, and Snowflake has a few rules of its own:

• Single-quoted strings accept both escapes: 'it''s' and 'it\'s' are the same value.
• Double quotes wrap case-sensitive identifiers, so "Order ID" and "SALES"."PUBLIC"."ORDERS" stay exactly as typed.
• $$ ... $$ marks a dollar-quoted string. Stored procedure, UDF and EXECUTE IMMEDIATE bodies inside it are kept verbatim: line breaks, casing and inner comments are not touched.
• // starts a line comment, in addition to -- and /* */.
• $name session variables, IDENTIFIER($table_name) and :name bind variables are recognised as single tokens.

Semi-structured data, FLATTEN and QUALIFY

Queries over VARIANT columns are where unformatted Snowflake SQL gets hardest to read. Colon paths and casts like e.payload:customer.id::number are left as one unit, and the cast type follows your keyword-case setting (::NUMBER). LATERAL FLATTEN(input => e.payload:items) i lands on its own line in the FROM list, with the => named argument intact.

QUALIFY is placed as a separate clause after WHERE, which suits the common “keep one row per key” pattern. The window spec inside OVER (...) is split into PARTITION BY and ORDER BY lines.

Not every word is upper-cased. Keyword case only applies to words that sql-formatter lists as Snowflake keywords, so over, desc, within and clone keep the case you typed. If you need uniform casing, type those few in the case you want.

Where Snowflake support stops

Some Snowflake-only syntax is outside what sql-formatter can parse, and it is better to know in advance:

• Stage references such as @raw.orders_stage or @~/staged in FROM or COPY INTO are rejected. In COPY INTO you can put the location in single quotes ('@raw.orders_stage/2026/'), which Snowflake also accepts. For SELECT $1, $2 FROM @stage, swap in a placeholder table name while formatting.
• ? placeholders are not accepted by the Snowflake grammar here; use :name binds instead.
• Before a bind variable, the formatter drops the space: status = :status becomes status =:status.
• CREATE OR REPLACE PROCEDURE headers come out on an awkward line break after CREATE, although the body is preserved.
• Snowflake Scripting outside $$ (bare DECLARE / BEGIN blocks) is flattened, and INTO :total loses its space. Wrapping the block in EXECUTE IMMEDIATE $$ ... $$ keeps it exactly as written.

Choosing options for Snowflake worksheets

Worksheets often hold a dozen statements in a row. Blank lines between statements controls the gap (0 to 5) after each semicolon. Pick Tabular, left-aligned under Indent style if you like clause keywords padded into a column; Standard is better for deep CTE nesting. Leading commas and AND / OR at the start of the line suit dbt models that change often in review.

Minify is useful for pasting a query into a Python or JavaScript string: whitespace and comments go, while $$ bodies and quoted strings stay unchanged. The SQL minifier page covers that mode in more detail.

Examples

select e.event_id, e.payload:customer.id::number as customer_id, i.value:sku::string as sku, i.value:qty::int as qty, i.value:price::number(10,2) as unit_price from raw.order_events e, lateral flatten(input => e.payload:items) i where e.event_type = 'order_placed' and e.loaded_at >= dateadd(day, -1, current_timestamp())

SELECT
  e.event_id,
  e.payload:customer.id::NUMBER AS customer_id,
  i.value:sku::STRING AS sku,
  i.value:qty::INT AS qty,
  i.value:price::NUMBER(10, 2) AS unit_price
FROM
  raw.order_events e,
  LATERAL flatten(input => e.payload:items) i
WHERE
  e.event_type = 'order_placed'
  AND e.loaded_at >= dateadd(day, -1, current_timestamp())

with sessions as (select session_id, user_id, min(event_ts) as started_at, max(event_ts) as ended_at, count(*) as events from analytics.web_events where event_date = current_date() - 1 group by session_id, user_id) select user_id, session_id, datediff('second', started_at, ended_at) as duration_s, events from sessions where events > 1 qualify row_number() over (partition by user_id order by events desc) = 1

WITH
  sessions AS (
    SELECT
      session_id
    , user_id
    , min(event_ts) AS started_at
    , max(event_ts) AS ended_at
    , count(*) AS events
    FROM
      analytics.web_events
    WHERE
      event_date = current_date() - 1
    GROUP BY
      session_id
    , user_id
  )
SELECT
  user_id
, session_id
, datediff('second', started_at, ended_at) AS duration_s
, events
FROM
  sessions
WHERE
  events > 1
QUALIFY
  row_number() over (
    PARTITION BY
      user_id
    ORDER BY
      events desc
  ) = 1

copy into raw.orders from '@raw.orders_stage/2026/' file_format = (type = 'CSV' skip_header = 1 field_optionally_enclosed_by = '"') on_error = 'CONTINUE';
select count(*) from identifier($target_table) where region = $region // session variables

COPY INTO raw.orders
FROM
  '@raw.orders_stage/2026/' file_format = (
    type = 'CSV' skip_header = 1 field_optionally_enclosed_by = '"'
  ) on_error = 'CONTINUE';


SELECT
  count(*)
FROM
  identifier($target_table)
WHERE
  region = $region // session variables

execute immediate $$
declare
  stale_rows number default 0;
begin
  delete from inventory.stock_snapshots where snapshot_date < dateadd(day, -90, current_date());
  stale_rows := sqlrowcount;
  return 'deleted ' || stale_rows;
end;
$$;

EXECUTE IMMEDIATE $$
declare
  stale_rows number default 0;
begin
  delete from inventory.stock_snapshots where snapshot_date < dateadd(day, -90, current_date());
  stale_rows := sqlrowcount;
  return 'deleted ' || stale_rows;
end;
$$;

Common errors and how to fix them

Related tools

• SQL formatter
• sql formatter/bigquery
• sql formatter/redshift
• sql formatter/postgresql
• sql validator
• sql minifier
• JSON to SQL
• sql style guide