* Andrew Dunstan (and...@dunslane.net) wrote: > On 04/24/2016 06:02 PM, Sehrope Sarkuni wrote: > >AFAIK, there's also no guarantee on the specific order of the > >resulting properties in the text representation either. I would > >suppose it's fixed for a given jsonb value within a database major > >version but across major versions it could change (if the > >underlying representation changes). > > The order is fixed and very unlikely to change, as it was chosen > quite deliberately to help ensure efficient processing. Any change > in on-disk representation of data types is something we work very > hard to avoid, as it makes it impossible to run pg_upgrade.
We do, from time-to-time, change on-disk formats in a backwards-compatible way though. In any case, it's my understanding that we don't *guarantee* any ordering currently and therefore we should discourage users from depending on it. If we *are* going to guarantee ordering, then we should document what that ordering is. Thanks! Stephen
signature.asc
Description: Digital signature
