Pedro Pedreira
Pedro Pedreira
I recently tried to generate split amalgamated files (`python3 scripts/amalgamation.py --extended --splits=8`) and found some issues (addressed in #4011 and #4012). This suggest this workflow (generate and build amalgamated files)...
Moving protobuf to use resolve_dependency() so it doesn't need to be compiled and installed manually in the user's root system. It also allows us to have more control about the...
Just caught the error in circleCI running on main branch: > https://app.circleci.com/pipelines/github/facebookincubator/velox/10898/workflows/7c8a19a8-7fee-4fdb-9631-aa7bc7627091/jobs/63441 The test is flaky, but you can usually reproduce it by running it 10+ times. The failed assert...
Folly is now being recompiled by "make fuzzertest", causing the fuzzer binary to also recompile and the whole job to timeout. This is an example of the failure: > https://app.circleci.com/pipelines/github/facebookincubator/velox/14386/workflows/0ac4c994-82a8-4961-aaae-f03ebf41c49b/jobs/87226...
VectorFuzzer can generate dictionaries with multiple levels, but today the indirection and base vectors always have the same size. To increase entropy, let's extend it to allow the generation of...
### Bug description Depending on the output buffer alignment, the existing merge join code tries to create an invalid dictionary (more rows than indices), if left join and filters are...
### Description Merge joins are useful to join two sorted streams of data. For queries that require joining two large tables, it is a common pattern to store them as...
Summary: Changing the merge join algorithm to avoid the copy to a new output buffer and rather wrapping the input left and right using dictionaries. Left buffers are always wrapped,...