Documentation
¶
Overview ¶
Package indexcore holds the per-file indexing logic shared by DocGraph's two indexing pipelines: indexStore (indexer.go, the CLI index/sync + serve --path path) and indexProjectOpts (internal/workspace/indexing.go, the serve --workspace + watcher path). It lives in its own package because the shared code needs both parser.ParseResult and store.Store, and parser already imports store (so the helper cannot live in store without a cycle) while its two callers sit in different packages (package main and package workspace).
Scope is deliberately narrow: only the byte-identical *dependent tail* — the per-file writes that must run AFTER a file's nodes + section chunks already exist (so foreign keys resolve). The divergent orchestration stays caller-side: node / section-chunk inserts (batched in indexStore, per-file in the workspace path), the stale-row delete block (guarded by force vs baseEmpty, and placed differently), the FTS bulk-rebuild trigger gate, and the nNew counter. See plans/index-pipeline-parity.md for why the pipelines are mirrored, not merged.
Index ¶
Constants ¶
This section is empty.
Variables ¶
This section is empty.
Functions ¶
func WriteDependents ¶
func WriteDependents( st DependentWriter, res *parser.ParseResult, fh git.FileHistory, gitEnabled bool, changedDocIDs *[]string, logPrefix string, ) error
WriteDependents persists the per-file dependent rows that must be written after a file's nodes and section chunks already exist: document / governance / research metadata, the entity graph, edges, unresolved references, the file row, and (when gitEnabled) the git history — then records the document ID in changedDocIDs for the downstream resolver + similarity pass.
It is the verbatim tail shared by both pipelines (indexer.go's batched Phase-B loop and workspace indexing.go's per-file writeOne). The ONLY difference between the two original call sites was the error / stderr message prefix, threaded here as logPrefix: "" for the CLI / serve --path pipeline and "[<project>] " for the workspace pipeline. Both call sites' messages are reproduced byte-for-byte.
It does NOT insert nodes or section chunks, run the stale-row delete block, or own the nNew counter — those stay caller-side because they differ between the pipelines (batched vs per-file, force vs baseEmpty). Metadata, edge, file, and unresolved-ref failures are fatal (returned); entity-graph and git-history failures are non-fatal (logged to stderr) so a bad entity pass or history fork never aborts indexing a document's nodes/edges — matching the pre-extraction behavior of both pipelines.
Types ¶
type DependentWriter ¶ added in v0.3.2
type DependentWriter interface {
InsertDocumentMetadata(nodeID string, tuples []store.MetadataTuple) error
UpsertGovernanceMetadata(nodeID string, tuples []store.MetadataTuple) error
UpsertResearchMetadata(nodeID string, tuples []store.MetadataTuple) error
InsertEdges(edges []store.Edge) error
InsertUnresolvedRefs(refs []store.UnresolvedRef) error
UpsertFile(f store.FileInfo) error
UpsertFileHistory(h store.FileHistory) error
// Required so st can be passed to entitygraph.IndexFile(EntityWriter);
// keep these identical to entitygraph.EntityWriter's method set.
InsertEntities([]store.Entity) error
InsertEntityMentions([]store.Mention) error
}
DependentWriter is the narrow persistence surface WriteDependents needs. It lists every store method WriteDependents calls directly, PLUS the two entity-graph writes that entitygraph.IndexFile needs: because WriteDependents forwards st to entitygraph.IndexFile (which takes an EntityWriter), DependentWriter's method set must be a superset of entitygraph.EntityWriter, so the two entity methods are required and must match it exactly. *store.Store satisfies the whole set (the entity methods via internal/store/forwarders.go).
Source Files