Documentation
¶
Overview ¶
Browse artifacts for the client-side reader. Alongside the per-language search indexes (search.go), the build emits a single global doc-id space (one entry per Work, in catalog order) as two RoaringRange sidecars the Hugo WASM reader opens directly:
- browse-facets.rrsf -- an RRSF facet sidecar (language/format/subject/ tag/classification/contributor -> doc-id postings), for client-side facet counts and filtering without pre-rendered facet pages.
- browse-records.bin/.idx -- an RRSR record store mapping a doc id to its compact result-card JSON, for rendering result rows without a page load.
browse-docs.json maps each doc id to its Work id, so the per-language search hits (which carry Work ids via their own docs maps) bridge into this global space, and a card links to the static /works/<id> detail page. browse-subjects.json maps each subject id to its labels + vocabulary scheme, so the fallback facet panel renders localized, scheme-grouped subjects .
Package search builds the catalog's lexical search index from the projected catalog (ARCHITECTURE §8). Per corpus language it emits, by script: a roaringrange term index (.rrt, boolean whole-word presence) paired with a BM25 impact sidecar (.rrb) for segmented scripts, or a trigram index (.rrs, RRSI) for unsegmented scripts (CJK/Thai/...) where word-level tokenization fails. A manifest routes each language to its index and kind -- the data the browser's reader queries. Building is done in Go; term-index queries run in the Rust/WASM reader the Hugo module ships, while the trigram (RRSI) index has a Go reader too.
Index ¶
Constants ¶
const ( BrowseIndexName = "browse-index.rrs" BrowseRecordsBin = "browse-records.bin" BrowseRecordsIdx = "browse-records.idx" BrowseFacetsName = "browse-facets.rrsf" BrowseDocsName = "browse-docs.json" BrowseSubjectsName = "browse-subjects.json" )
Browse artifact filenames.
const ( FacetLanguage = "language" FacetFormat = "format" FacetSubject = "subject" FacetTag = "tag" FacetClassification = "classification" FacetContributor = "contributor" )
Facet field names in the RRSF sidecar; the reader's filter pairs ([field, category]) and the Hugo facet UI key on these.
const SchemaVersion = 3
SchemaVersion is the search-manifest schema version, checked by the reader. v2 adds the per-language BM25 impact sidecar (IndexInfo.Impacts); v3 adds the trigram (RRSI) index kind for unsegmented scripts (IndexInfo.Kind/GramSize).
Variables ¶
This section is empty.
Functions ¶
Types ¶
type IndexInfo ¶
type IndexInfo struct {
Language string `json:"language"` // ISO 639-2 code, or "und" when undeclared
Kind string `json:"kind"` // "terms" (RRTI) or "trigram" (RRSI)
TermLanguage uint8 `json:"termLanguage"` // terms: roaringrange stemmer-language byte
Stemmed bool `json:"stemmed"` // terms only
Stopwords bool `json:"stopwords"` // terms only
GramSize int `json:"gramSize,omitempty"` // trigram: n-gram size
Index string `json:"index"` // .rrt or .rrs filename
Impacts string `json:"impacts,omitempty"` // terms: .rrb BM25 impact sidecar
Docs string `json:"docs"` // JSON array: doc id (index) -> Work id
DocCount int `json:"docCount"`
}
IndexInfo describes one per-language index and how it was tokenized, so the reader tokenizes queries identically. Kind selects the query path: "terms" is a word-level RRTI index (.rrt) with optional stemming + BM25 sidecar; "trigram" is an RRSI n-gram index (.rrs) for unsegmented scripts, queried by NgramKeys.
type Manifest ¶
Manifest is the language->index routing map the browser reader loads: one entry per language present in the corpus (§8).
func BuildIndexes ¶
BuildIndexes writes one index per corpus language into sink -- a word-level term index (+ BM25 sidecar) for segmented scripts, or a trigram (RRSI) index for unsegmented ones -- plus a doc-id->Work-id list per index and a search-manifest.json routing map. A Work is indexed once per language it declares; a Work with none goes to the undetermined index. Doc ids are dense from 0 in the projected (sorted) order.
Source Files