I am a huge +1 this behaviour, ie not failing queries, use a slow full table scan if needed and warn users that they should create an index. This allows people to easily start constructing and improving their query without having to presetup their query with indexes before they know what query they actually want. When they have their query they can be guided as to exactly what indexes are needed to make these queries fast.
On 22 November 2017 at 20:38, Tony Sun <tony.sun...@gmail.com> wrote: > Hello Everyone, > > Will and I were hoping to get some more opinions for > https://github.com/apache/couchdb/pull/962. > > To provide some context, Mango had a long existing bug that allowed users > to incorrectly use an index for a query. This bug was fixed in > https://github.com/apache/couchdb/pull/816. However, this fix can lead to > a > breaking change for users who incorrectly used the index prior to the fix. > Instead of results being returned, the user will see a "no_usable_index" > error. > > PR 962 mitigates this issue by essentially removing "no_usable_index". > We either find another usable index, or perform a full table scan via > _all_docs. > > Should we allow this new behavior or continue to throw errors and force > users to create a new index? > > Happy Holidays! > > Tony >
