Documentation
¶
Overview ¶
Package indexer walks a repository, parses files, and stores the result in the NeuroFS SQLite index.
Index ¶
- Constants
- func BuildChunks(filePath, relPath string, lang models.Lang, content string, ...) []models.Chunk
- func BuildRelations(files []models.FileRecord) []models.FileRelation
- func RequiresEmbeddingReindex(db *storage.DB, provider, model string) (bool, error)
- func RequiresReindex(db *storage.DB) (bool, error)
- func RequiresSourceReindex(cfg *config.Config, db *storage.DB) (bool, error)
- type Options
- type Stats
- type Watcher
Constants ¶
const ( // ProjectMetaKey is the metadata-table key under which indexer persists // the decoded project.Info. Exposed so other packages can read it. ProjectMetaKey = "project_info" )
Variables ¶
This section is empty.
Functions ¶
func BuildChunks ¶
func BuildChunks(filePath, relPath string, lang models.Lang, content string, indexedAt time.Time) []models.Chunk
BuildChunks splits a file into stable, hashable code chunks. Go uses the standard AST for top-level declarations; other languages currently fall back to a whole-file chunk until their AST chunkers land.
func BuildRelations ¶
func BuildRelations(files []models.FileRecord) []models.FileRelation
BuildRelations walks all indexed file records, resolves their imports to target file records, and returns the set of file relationships.
func RequiresEmbeddingReindex ¶
RequiresEmbeddingReindex reports whether db's persisted vectors belong to a different provider/model space than the currently configured client. Comparing vectors across spaces silently corrupts semantic ranking even when their dimensions happen to match.
func RequiresReindex ¶
RequiresReindex reports whether db was built with different parser/chunker semantics. Callers that consume an existing non-empty index (notably retrieval.NewSession) use it to avoid serving stale chunks after a NeuroFS binary upgrade.
func RequiresSourceReindex ¶
RequiresSourceReindex reports whether the indexable working tree differs from the file generations persisted in db. It applies the same traversal, language, size, line-count, and checksum rules as Run, but never mutates the database. Retrieval uses this check before loading a new session so a normal working-tree edit is indexed before its chunks are searched.
Types ¶
type Options ¶
type Options struct {
// Verbose enables per-file logging via the provided function.
// If nil, no per-file output is produced.
Logf func(format string, args ...any)
}
Options configures an indexing run.
type Stats ¶
type Stats struct {
Discovered int // all files visited
Skipped int // unsupported or ignored
Indexed int // successfully written to the DB
Cached int // files skipped because they are already indexed and unmodified
Updated int // existing records refreshed
Removed int // stale records deleted
Symbols int // total symbols extracted
Imports int // total unique imports extracted
Chunks int // total chunks extracted
Errors int // files that produced errors (skipped)
Duration time.Duration
}
Stats summarises what happened during an indexing run.
type Watcher ¶
type Watcher struct {
// contains filtered or unexported fields
}
Watcher monitors a repository filesystem for changes and updates the index incrementally.
Source Files