+1 (binding) On Fri, Oct 23, 2026 at 5:32 PM Eduard Tudenhöfner <[email protected]> wrote:
> +1 (binding) > > On Fri, Oct 23, 2026 at 12:06 AM Amogh Jahagirdar <[email protected]> > wrote: > >> +1 (binding) >> >> On Thu, Oct 22, 2026 at 2:27 PM Fokko Driesprong <[email protected]> >> wrote: >> >>> +1 (binding) >>> >>> On 2026/10/22 19:00:04 Russell Spitzer wrote: >>> > +1 (bind) >>> > >>> > On Thu, Oct 22, 2026 at 11:41 AM Andrei Tserakhau via dev < >>> > [email protected]> wrote: >>> > >>> > > +1 (non-binding) >>> > > >>> > > On Thu, Oct 22, 2026 at 8:14 PM Anurag Mantripragada < >>> > > [email protected]> wrote: >>> > > >>> > >> +1 (non-binding) >>> > >> >>> > >> This applies nicely for column updates too. Thanks Anoop. >>> > >> >>> > >> ~ Anurag >>> > >> >>> > >> On Thu, Oct 22, 2026 at 11:11 AM Szehon Ho <[email protected] >>> > >>> > >> wrote: >>> > >> >>> > >>> +1 (binding) >>> > >>> >>> > >>> On Thu, Oct 22, 2026 at 11:03 AM Hongyue Zhang < >>> [email protected]> >>> > >>> wrote: >>> > >>> >>> > >>>> +1 (non-binding) >>> > >>>> >>> > >>>> On Thu, Oct 22, 2026 at 10:49 AM Ryan Blue <[email protected]> >>> wrote: >>> > >>>> >>> > >>>>> +1 (binding) >>> > >>>>> >>> > >>>>> On Thu, Oct 22, 2026 at 10:46 AM Steven Wu <[email protected] >>> > >>> > >>>>> wrote: >>> > >>>>> >>> > >>>>>> +1 (binding) >>> > >>>>>> >>> > >>>>>> On Thu, Oct 22, 2026 at 9:56 AM Anoop Johnson <[email protected] >>> > >>> > >>>>>> wrote: >>> > >>>>>> >>> > >>>>>>> Hi, everyone - >>> > >>>>>>> >>> > >>>>>>> Following the discussion on v4 stats tightness [1], I'd like >>> to call >>> > >>>>>>> a vote on the spec change that replaces the per-column >>> `tight_bounds` field >>> > >>>>>>> with a single file-level `tight_stats` flag. >>> > >>>>>>> >>> > >>>>>>> Summary of the change: >>> > >>>>>>> >>> > >>>>>>> - Stats tightness is tracked once per file with a >>> `tight_stats` >>> > >>>>>>> flag on the tracked file instead of a `tight_bounds` field >>> on >>> > >>>>>>> every column's stats struct. >>> > >>>>>>> - The tightness flag covers all statistics (bounds and null, >>> > >>>>>>> NaN, and value counts), not just bounds, so metadata-only >>> > >>>>>>> queries can trust the counts as well. >>> > >>>>>>> - When the `tight_stats` flag is absent, readers must assume >>> > >>>>>>> stats are not tight. >>> > >>>>>>> - A deletion vector no longer implicitly makes stats wide; >>> > >>>>>>> `tight_stats` is the authoritative indicator, and engines >>> are >>> > >>>>>>> allowed to retighten stats without rewriting data files. >>> > >>>>>>> >>> > >>>>>>> References: >>> > >>>>>>> >>> > >>>>>>> 1. Discussion thread: >>> > >>>>>>> >>> https://lists.apache.org/thread.html/tv7bvqctdfh5qng1sx5rp8r4r00nsv60 >>> > >>>>>>> >>> > >>>>>>> 2. Design options doc: >>> > >>>>>>> >>> https://docs.google.com/document/d/13DbwnwKQXjvY3vRcFJo-qVAQWk-GB4zLzzYmFdPntZg/ >>> > >>>>>>> >>> > >>>>>>> 3. Spec PR: https://github.com/apache/iceberg/pull/18585 >>> > >>>>>>> >>> > >>>>>>> Please vote within the next 72 hours. >>> > >>>>>>> >>> > >>>>>>> [ ] +1 Adopt the tight_stats spec change >>> > >>>>>>> [ ] +0 >>> > >>>>>>> [ ] -1 Do not adopt, with reason >>> > >>>>>>> >>> > >>>>>>> Thanks, >>> > >>>>>>> Anoop >>> > >>>>>>> >>> > >>>>>> >>> > >>> >>
