A meaningful error message will be great! On Tue, Jun 11, 2019 at 6:15 PM Maryann Xue <maryann...@apache.org> wrote:
> BTW, I've actually just done some work on hint error handling, which might > be helpful to what you mentioned: > > https://github.com/apache/spark/pull/24653 > > On Tue, Jun 11, 2019 at 8:04 PM Maryann Xue <maryann....@databricks.com> > wrote: > >> I believe in the SQL standard, the original name cannot be accessed once >> it’s aliased. >> >> On Tue, Jun 11, 2019 at 7:54 PM John Zhuge <jzh...@apache.org> wrote: >> >>> Yeah, it is a touch scenario. >>> >>> I actually have much simpler cases: >>> >>> 1) select /*+ broadcast(t1) */ * from db.t1 join db.t2 on t1.id = t2.id; >>> 2) select /*+ broadcast(t1) */ * from db.t1 a1 join db.t2 a2 on a1.id = >>> a2.id; >>> >>> 2) is the same as 1) but with aliases. Many users were surprised that 2) >>> stopped working. >>> >>> Thanks, >>> John >>> >>> >>> On Tue, Jun 11, 2019 at 4:38 PM Maryann Xue <maryann...@apache.org> >>> wrote: >>> >>>> Yes, and for a good reason: the hint relation has exactly the same >>>> scope with other elements of queries/sub-queries. >>>> >>>> Suppose there's a query like: >>>> >>>> select /*+ broadcast(s) */ from (select a, b from s) t join (select a, >>>> b from t) s on t1.a = t2.b >>>> >>>> If we allowed the hint resolving to "cross" the scopes, we'd end up >>>> with a really confusing spec. >>>> >>>> >>>> Thanks, >>>> Maryann >>>> >>>> On Tue, Jun 11, 2019 at 5:26 PM John Zhuge <jzh...@apache.org> wrote: >>>> >>>>> Hi Reynold and Maryann, >>>>> >>>>> ResolveHints javadoc indicates the traversal does not go past subquery >>>>> alias. Is there any specific reason? >>>>> >>>>> Thanks, >>>>> John Zhuge >>>>> >>>> >>> >>> -- >>> John Zhuge >>> >> -- John Zhuge
