Method vault_embed

atomic-repository/src/repository/vault_embeddings.rs:51–129

Embed a single vault entry using the provided embedding function. Chunks the content by heading/paragraph, computes embeddings via `embed_fn`, and stores them in redb. Returns the number of chunks stored. Staleness detection: skips re-embedding if the content hash matches the existing stored hash.

(
        &self,
        path: &str,
        embed_fn: &F,
        config: &EmbedConfig,
    )

    /// Staleness detection: skips re-embedding if the content hash matches
    /// the existing stored hash.
    pub fn vault_embed<F>(
        &self,
        path: &str,
        embed_fn: &F,
        config: &EmbedConfig,
    ) -> Result<usize, RepositoryError>
    where
        F: Fn(&str) -> Vec<f32>,
    {
        let entry = self
            .vault_retrieve(path)?
            .ok_or_else(|| RepositoryError::FileNotFound {
                path: std::path::PathBuf::from(path),
            })?;

        let content = String::from_utf8_lossy(&entry.content_bytes);
        let content_hash = entry.content_hash;

        // Check staleness — if first chunk has same hash, skip
        {
            let txn = self
                .pristine
                .read_txn()
                .map_err(|e| RepositoryError::Database(e.to_string()))?;
            if let Ok(Some(existing)) = txn.get_embedding(path, 0) {
                if existing.content_hash == content_hash {
                    return Ok(0); // Content unchanged, skip
                }
            }
        }

        // Chunk the content
        let chunks = chunk_by_heading(&content, config.max_chunk_tokens);

        // Delete old embeddings for this path
        {
            let mut txn = self
                .pristine
                .write_txn()
                .map_err(|e| RepositoryError::Database(e.to_string()))?;
            let _ = txn
                .del_embeddings(path)
                .map_err(|e| RepositoryError::Database(e.to_string()))?;
            txn.commit()
                .map_err(|e| RepositoryError::Database(e.to_string()))?;
        }

        // Embed and store each chunk
        let mut txn = self
            .pristine
            .write_txn()
            .map_err(|e| RepositoryError::Database(e.to_string()))?;

        let mut count = 0;
        for (idx, chunk) in chunks.iter().enumerate() {
            if chunk.text.trim().is_empty() {
                continue;
            }