I seem to recall delta encodings might use these routines as well, in which case IIRC you get zero bit width if all values in a block are equal.
On Wednesday, December 18, 2024, Gang Wu <ust...@gmail.com> wrote: > IIUC, the bit-width could be 0 when the dictionary contains a single entry > and then all entry ids are zeros. I haven't tried to create such a file and > I suspect that RLE (rather than bit-packing) is in use in this case. > > Best, > Gang > > On Thu, Dec 19, 2024 at 9:43 AM Marko Divjak <mdivja...@gmail.com> wrote: > > > Hi, > > > > I was exploring the apache arrow code and saw this this > > < > > https://github.com/apache/arrow/blob/b655852b260d3b8c3fe457795df0f4 > 2a2ff9c98c/cpp/src/arrow/util/bpacking64_default.h#L35C1-L40C2 > > > > > in > > the repo. > > What does it mean when bit width is zero in dictionary encoded files? > > What is the scenario where data will be encoded with bit width zero and > how > > could such a file be generated? > > > > Regards, > > Marko > > >
