feat: support truncate table api

[zhangyue19921010]

Semantics

• Introduces truncate table to create a new snapshot version whose manifest has no fragments while preserving the existing schema and table config.
• Acts as an overwrite-style commit for the latest version; does not physically delete historical data files.
• After truncation, new data can be appended or the table can be restored to previous versions.

Differences vs Delete Rows

• Delete Rows:
  • Soft-deletes rows by writing deletion files/vectors; existing fragments remain
• Truncate Table:
  • Replaces the current version with an “empty fragments” version; no deletion files are produced and current-version fragments are not carried forward.

Rust:
ds.truncate_table().await.unwrap();

Python:
ds.truncate_table()

Java:
ds1.truncateTable()

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[wjones127]

• Soft-deletes rows by writing deletion files/vectors; existing fragments remain

This isn't true; if you delete an entire fragment it is removed from the list. If you delete all rows, all fragments are removed. So this is just equivalent to dataset.delete("true"). We even have a fast path for that ¹. Is there any other reason to make a special operation for this?

Footnotes

1. https://github.com/lance-format/lance/blob/0a8d618b5ec6ce6ac978aadced88a0061487f9ec/rust/lance/src/dataset/write/delete.rs#L182-L192 ↩

[zhangyue19921010]

Hi, thanks a lot for your attention @wjones127 ! The addition of support for Truncate table here is mainly intended to supplement the TRUNCATE TABLE related to standard SQL (SQL:2008) at the syntax level, making it more user-friendly for those analysts accustomed to using TRUNCATE TABLE command.

Of course, As u mentioned, Lance users can also use delete("true") and complete it quickly based on the fast path in the code.

[wjones127]

Okay, I get that it's a more familiar API. Could you change the implementation to just be a simple wrapper around delete("true") then?

[zhangyue19921010]

Okay, I get that it's a more familiar API. Could you change the implementation to just be a simple wrapper around delete("true") then?

Sure thing. Changed. PTAL ~

[zhangyue19921010]

Hi @wjones127 would u mind to take another look? Thanks :) I have finished that a simple wrapper around delete("true")

[wjones127]

Looks good!