cetra3 commented on PR #11511:
URL: https://github.com/apache/arrow-rs/pull/11511#issuecomment-6968276561

   @alamb @sdf-jkl I have split this into a stack of smaller PRs, as you asked:
   
   1. #12308 refactors the decoding helpers. It is ready for review.
   2. #12309 adds the views and builds the existing types on them. Draft.
   3. #12310 moves `variant_get` onto the views. Draft.
   4. #12311 makes shredding copy leftover fields by field id. Draft.
   
   Each one is stacked on the one before, so 2–4 are drafts until their parent 
merges. This PR stays open as the end state. Once the four land, it rebases 
down to the removal of `borrowed_field_id`.
   
   One thing to know before you review: the first two PRs are slower on some 
benchmarks, and the gains only arrive once the whole shape lands. I benchmarked 
each PR against its parent, locally on x86_64:
   
   - #12308 makes `variant_get_utf8_from_unshredded_string` and 
`variant_get_binary_from_string` 14–15% slower. I do not yet know why a 
refactor does this.
   - #12309 makes four unshredded `variant_get` benchmarks 5–8% slower.
   - #12310 then makes unshredded `variant_get` 18–47% faster, and #12311 makes 
shredding 14–21% faster.
   
   Each PR's description has its own table. If you would rather not merge PRs 
that are slower on their own, I can fold 1 into 2, or 2 into 3.
   


-- 
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]

Reply via email to