You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Commit e0c7963
Browse filesBrowse the repository at this point in the historyBrowse files
test: run the sketch SQL files instead of excluding them globally
`ignoreList` registers a ScalaTest `ignore` before execution, so listing
`thetasketch.sql` and `tuplesketch.sql` there skipped both files even with
`ENABLE_COMET=false`, dropping 364 `SELECT`s from the Spark baseline.
The recorded rationale was also wrong. The sketch aggregates run on Spark's
`ObjectHashAggregate` and are never converted; what fails is the native scan
reading the `t_string_collation` fixture, which contains invalid UTF-8
(`CAST(X'C1' AS STRING)`, `CAST(X'80' AS STRING)`):
Parquet error: Arrow: Parquet argument error: Parquet error: encountered
non UTF-8 data
That is apache#4121, and it has nothing to do with collation: the uncollated query
at tuplesketch.sql:825 fails identically.
Switch both files to the `--SET spark.comet.enabled = false` header already
used for `hll.sql`, which fails on the same fixture for the same reason. Every
query then runs and is verified against the golden file in both the baseline
and the Comet job, rather than the whole file being skipped.
0 commit comments