Describe the enhancement requested
#502 was closed with a format change to suggest writers emit INT96 statistics with a specific order (compare lower 4 bytes followed by upper 8 bytes). This is correct, but there are many files in the wild with incorrect stats as some older writer versions produced stats with simple bytewise comparisons. To disambiguate correct stats that can be used for processing from incorrect stats that should be ignored, I propose defining a new sort order specifically for INT96 timestamp values.
Describe the enhancement requested
#502 was closed with a format change to suggest writers emit INT96 statistics with a specific order (compare lower 4 bytes followed by upper 8 bytes). This is correct, but there are many files in the wild with incorrect stats as some older writer versions produced stats with simple bytewise comparisons. To disambiguate correct stats that can be used for processing from incorrect stats that should be ignored, I propose defining a new sort order specifically for INT96 timestamp values.