Hello,
+1 (binding). Nice work on making the spec reach a satisfactory state!
The only issues left AFAICT are minor wording and doc organization
issues (such as: splitting the detailed encoding description into a
separate page).
Regards
Antoine.
Le 20/07/2026 à 15:45, PRATEEK GAUR a écrit :
Hello everyone,
This is a vote to add ALP (Adaptive Lossless floating-Point) as a new
encoding in Parquet, as discussed in the design document and proposal pull
request.
We've had discussions on this in the design document (
https://docs.google.com/document/d/1xz2cudDpN2Y1ImFcTXh15s-3fPtD_aWt/edit)
and the proposal pull request (
https://github.com/apache/parquet-format/pull/557), as well as the
discussion thread (
https://lists.apache.org/thread/tjtln1mmjqfoql1ls2dw9xpdk91r1909).
Reference implementations are available in
- parquet-java (https://github.com/apache/parquet-java/pull/3397)
- Arrow C++ (https://github.com/apache/arrow/pull/48345)
- Arrow Go (https://github.com/apache/arrow-go/pull/704).
- Test artifacts : https://github.com/apache/parquet-testing/pull/100
Cross-language compatibility is verified: the Arrow C++ decoder reads
Java-written data bit, covering V1 and V2 pages, multiple vector sizes, and
several real datasets — zero mismatches.
This vote will be open for at least 72 hours:
[ ] +1: Add ALP encoding to the Parquet spec
[ ] +0: Not going to block, but prefer . . .
[ ] -1: Do not add ALP to Parquet
Thank you for all the feedback, input, and involvement in developing this
new encoding!
Thanks,
Prateek Gaur
