PSeitz

> My specific requirement is to write an FFI search function for C++ usage. I meant the "stringent requirements regarding search latency" > So, if I change the indexing method...

Duplicate of https://github.com/quickwit-oss/tantivy/issues/301

Can you tell more about your use case? What happens to existing data if you add a new field to the schema? Would docs indexed before the schema change simply...

Elastic search 7.2: > [7.2] Deprecated in 7.2. `interval` field is deprecated Historically both calendar and fixed intervals were configured in a single interval field, which led to confusing semantics....

Looking at the OpenSearch docs, it's unclear if they handle calendar aware intervals or not. So should we implement the `interval` halfway, only with values higher than 1?

`1m` is date-aware in elastic search, this is incompatible with the fixed interval handling we have. Most users are familiar with elastic search and they would all run into the...

`1m`, `1h`, `1d`, `1w`, `1M`, `1q`, `1y`. They are all date aware. All other values are not, e.g. `24h`, `2w` etc. So we could implement the interval halfway, only for...

The issue appeared locally, this is a flamegraph from the process 

Hi, all the aggregations in tantivy are in the aggregation folder: https://github.com/quickwit-oss/tantivy/tree/main/src/aggregation The aggregation docs (latest release, not main branch. better build that locally with `cargo doc`) https://docs.rs/tantivy/latest/tantivy/aggregation/index.html For the...

This PR https://github.com/quickwit-oss/tantivy/pull/2062 fixes this mostly. Only allocation is now the `BoxTokenStream` per text, which could be avoided with some lifetime hacks (and unsafe probably).