feat: add 'date.to_start_of_interval' function for truncating values to interval boundaries

[sunyeongchoi]

Closes #5734

Adds a to_start_of_interval function to provide native truncating values to interval boundaries capabilities in PRQL.

from events | select (event_time | date.to_start_of_interval 15 minute)

[max-sixty]

how about using some form of round or floor here?

either literally or at least using the term as part of the name?

[sunyeongchoi]

@max-sixty
Thanks for the feedback.
I want to make sure I understood correctly. Are you proposing we use a term like 'floor' in the name (e.g., date.floor_interval) to make the floor logic more explicit?

[max-sixty]

I want to make sure I understood correctly. Are you proposing we use a term like 'floor' in the name (e.g., date.floor_interval) to make the floor logic more explicit?

yes, something like this!

@prql-bot maybe you could do a survey of what is supported by other languages, around "rounding" to regular intervals

[prql-bot]

Here's a survey of how other systems name their datetime truncation/rounding-to-interval functions:

SQL Databases

Database	Standard Trunc	Arbitrary Interval (e.g. 15 min)
PostgreSQL	date_trunc('hour', ts)	date_bin('15 min', ts, origin)
ClickHouse	date_trunc('hour', ts)	toStartOfInterval(ts, INTERVAL 15 MINUTE)
DuckDB	date_trunc('hour', ts)	time_bucket(INTERVAL '15 min', ts) / date_bin(...)
BigQuery	TIMESTAMP_TRUNC(ts, HOUR)	(manual arithmetic)
Snowflake	DATE_TRUNC('hour', ts)	TIME_SLICE(ts, 15, 'MINUTE')
Presto/Trino	date_trunc('hour', ts)	date_bin(INTERVAL '15' MINUTE, ts, origin)
SQL Server	DATETRUNC(hour, ts)	(manual DATEADD/DATEDIFF)
TimescaleDB	date_trunc(...)	time_bucket('15 minutes', ts)

Programming Languages / Data Tools

System	Function	Notes
pandas	dt.floor('15min')	Also has dt.round(), dt.ceil()
Polars	dt.truncate('15m')	Also has dt.round()
R (lubridate)	floor_date(ts, '15 minutes')	Also round_date(), ceiling_date()
Julia	Dates.floor(ts, Minute(15))	Also round(), ceil()
Spark	date_trunc('hour', ts)	window(ts, '15 minutes') for arbitrary

Key takeaways

1. All systems use floor semantics — truncation/floor is the universal default, not round-to-nearest.
2. date_trunc is the dominant SQL name but only handles standard calendar units, not arbitrary intervals.
3. For arbitrary intervals, naming is fragmented: date_bin (PG/Trino), time_bucket (TimescaleDB/DuckDB), toStartOfInterval (ClickHouse), TIME_SLICE (Snowflake).
4. Programming languages uniformly use floor — and offer round/ceil as companions.
5. toStartOfInterval is ClickHouse-specific — no other major system uses this naming convention.

Possible PRQL names

Candidate	Pros	Cons
date.floor	Matches pandas/R/Julia; short; natural round/ceil family	Less familiar to SQL-only users
date.truncate	Matches Polars; close to SQL date_trunc	Slightly verbose
date.bin	Matches PostgreSQL/Trino date_bin	Jargon-y; less intuitive
date.bucket	Matches TimescaleDB time_bucket	Narrow adoption
date.to_start_of_interval	Very descriptive	Verbose; ClickHouse-specific

date.floor seems strongest — it's short, widely used in the data ecosystem, and naturally extends to date.round / date.ceil later if needed.

[sunyeongchoi]

@max-sixty

What do you think of date.floor suggested by prql-bot? if you're okay with it, I'll go ahead and rename it.

from events | select (event_time | date.floor 15 minute)

[max-sixty]

sorry I missed this! I think the .floor is good, if you happen to see this!

[prql-bot]

The function was renamed to date.floor during review (following the naming survey), but the PR title still says date.to_start_of_interval. Worth updating it before merge — PRs here land with the PR title as the squash commit message, so it would otherwise ship as a stale name in the history.

Implementation looks good — tests cover the four supported dialects and the error path for the others, and the per-dialect SQL output matches what each engine expects.

[prql-bot]

CI is failing on cargo fmt --check — the date_floor_unsupported_dialects and date_floor_operator blocks in prqlc/prqlc/tests/integration/sql.rs have a few off-by-four-spaces indent issues and a wrapping that rustfmt wants on one line. Running cargo fmt --all locally and committing the result will fix it.

Happy to push the fix if useful — say the word.

[sunyeongchoi]

@max-sixty Thanks! As you suggested, I changed the name to date.floor