Extending the word model

The word-analysis example can find neighbouring pairs and group occurrences using array positions. Highlighting a word in the original document also needs its source location. This page follows that requirement into a richer data model. These are possible adapter choices; the runnable word example does not implement the extensions.

Retain occurrences

To identify a word in the original text, retain its text and source range. The host can expose those as record fields or supply a native handle into the source. With native handles, two occurrences of apple are distinct even though their spellings are equal.

Choose the identity rule before exposing unique. Deduplicating occurrences and deduplicating spellings answer different questions. An explicit text projection gives the query author control over that choice.

Add properties

A native word host can supply text and byte_length projections. The shared projection construct can then build a record for each subject:

project(w, {"word": one(w | text), "bytes": one(w | byte_length)})

The host must register both stages and their result types. Each returns one scalar per word; one checks that cardinality. UTF-8 byte length is a specific choice. Character count, grapheme count and tokenization policy need their own definitions if the application exposes them.

Connect neighbouring words

An occurrence model can expose a next_word edge and source positions. Native matching can retain a word and its successor in the same capture, while a predicate compares their spellings. This supports questions such as finding adjacent repeated words without losing either source location.

Zirql checks the scope of capture names and enumerates candidates. The adapter defines which occurrences are adjacent and how their spellings compare. A bounded path over next_word can also retain intermediate occurrences; the host should expose it only if a concrete query needs that path.

Use the same boundaries for another domain

A syntax tree replaces word occurrences with source nodes. An operation graph replaces adjacency with domain edges. In each case, native values retain their owner and explicit projections construct report data.

The adapter examples explain those decisions in Zincite and Zirium. Their own guides describe application usage. The host integration reference maps these adapter choices to Rust traits.