Rich-T-kid opened a new pull request, #27441:
URL: https://github.com/apache/datafusion/pull/27441

   ## Which issue does this PR close?
   
   - Part of #20935.
   
   ## Rationale for this change
   
   `decode` currently coerces dictionary-encoded binary input to a flat binary 
array before decoding. DataFusion's performance contribution workflow requires 
canonical benchmark coverage before changing that behavior.
   
   ## What changes are included in this PR?
   
   This benchmark-only PR extends the existing `dictionary_encoding` Criterion 
target with valid hex-encoded dictionary input. It builds physical casts from 
the UDF's coercion result, so current `main` measures dictionary 
materialization and a later preservation change can use the same benchmark 
unchanged.
   
   The cases cover cold dictionaries, batches sharing dictionary values, flat 
inputs, and a filtered dictionary retaining four times as many backing values 
as rows.
   
   ## What is the testing strategy for this PR?
   
   ```text
   cargo fmt --all -- --check
   cargo bench -p datafusion-physical-expr --bench dictionary_encoding --no-run
   cargo bench -p datafusion-physical-expr --bench dictionary_encoding -- 
'dictionary_encoding/expression/decode' --sample-size 10 --warm-up-time 1 
--measurement-time 1
   git diff --check
   ```
   
   Current `main` baseline (Criterion middle estimates):
   
   | Case | Time |
   |---|---:|
   | cold dictionaries, NDV 10-8,192 | 162.43-180.56 us |
   | shared dictionary, NDV 10-8,192 | 169.63-180.58 us |
   | flat, NDV 10 | 145.57 us |
   | flat, NDV 8,192 | 163.04 us |
   | retained backing 4x | 182.35 us |
   
   ## Are there any user-facing changes?
   
   No. This PR only adds benchmark coverage.
   


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to