d54a7a5834
Roadmap 3-1. Verb-noun shape borrowed from gh search (gh search repos / code / commits / issues / prs verified against the gh manual). Subcommands: - `search chunks "<q>" --kb X` — hybrid retrieval (RAG search). - `search kb "<q>"` — case-insensitive substring match across KB names and descriptions; sorted by name length (shortest hits first). - `search docs "<q>" --kb X` — pages through ListKnowledge filtering by title / file_name; stops once --limit matches are found. - `search sessions "<q>"` — pages through GetSessionsByTenant filtering by title / description. kb / docs / sessions are client-side filters because the server has no fuzzy search endpoint for any of them. ListKnowledgeBases returns the full tenant catalog in one call; the doc/session walkers chunk at 200 per request and stop early on limit. The parent `search` command is a pure dispatcher — there is no bare- positional form (no `weknora search "<q>"`). Cleanups surfaced by the post-commit reviewer round: - UX consistency: search docs's displayDocName ordered Title → FileName → "-", while doc list's displayName uses FileName → Title → ID. Same Knowledge rendered differently across commands. Aligned search docs on doc list's existing FileName-first convention. - cmdutil.ResolveKBFlag(ctx, lister, raw) — extracted the `IsKBID ? raw : ResolveKBNameToID` block duplicated across chunks and docs. - text.ContainsFold(needle, fields...) — replaces inline `strings.Contains(strings.ToLower(field), needle)` patterns. 37 unit tests across chunks/kb/docs/sessions plus the parent registration smoke-test. Roadmap: 3-1.
163 lines
5.9 KiB
Go
163 lines
5.9 KiB
Go
package search
|
|
|
|
import (
|
|
"context"
|
|
"fmt"
|
|
"strings"
|
|
|
|
"github.com/spf13/cobra"
|
|
|
|
"github.com/Tencent/WeKnora/cli/internal/agent"
|
|
"github.com/Tencent/WeKnora/cli/internal/cmdutil"
|
|
"github.com/Tencent/WeKnora/cli/internal/format"
|
|
"github.com/Tencent/WeKnora/cli/internal/iostreams"
|
|
sdk "github.com/Tencent/WeKnora/client"
|
|
)
|
|
|
|
// ChunksOptions is the runtime configuration of a chunks search.
|
|
type ChunksOptions struct {
|
|
Query string
|
|
KB string // raw --kb (UUID or name)
|
|
KBID string // resolved id; populated before HybridSearch
|
|
Limit int
|
|
VectorThreshold float64
|
|
KeywordThreshold float64
|
|
NoVector bool
|
|
NoKeyword bool
|
|
JSONOut bool
|
|
}
|
|
|
|
// ChunksService is the narrow SDK surface used by runChunks. *sdk.Client
|
|
// satisfies it; tests inject fakes via Factory.Client.
|
|
type ChunksService interface {
|
|
HybridSearch(ctx context.Context, kbID string, params *sdk.SearchParams) ([]*sdk.SearchResult, error)
|
|
}
|
|
|
|
// NewCmdChunks builds `weknora search chunks "<query>" --kb <id-or-name>`.
|
|
// Mirrors gh `search code`'s "subject as positional" shape (the previous
|
|
// top-level `weknora search` is its legacy alias — see search.go).
|
|
//
|
|
// The `--kb` flag accepts either a KB UUID (passed through) or a name
|
|
// (resolved via ListKnowledgeBases). Mirrors gcloud `--project`'s
|
|
// id-or-name auto-detection.
|
|
func NewCmdChunks(f *cmdutil.Factory) *cobra.Command {
|
|
opts := &ChunksOptions{}
|
|
cmd := &cobra.Command{
|
|
Use: `chunks "<query>"`,
|
|
Short: "Hybrid (vector + keyword) chunk retrieval against a knowledge base",
|
|
Args: cobra.ExactArgs(1),
|
|
RunE: func(c *cobra.Command, args []string) error {
|
|
opts.Query = strings.TrimSpace(args[0])
|
|
if err := opts.validate(); err != nil {
|
|
return err
|
|
}
|
|
cli, err := f.Client()
|
|
if err != nil {
|
|
return err
|
|
}
|
|
kbID, err := cmdutil.ResolveKBFlag(c.Context(), cli, opts.KB)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
opts.KBID = kbID
|
|
return runChunks(c.Context(), opts, cli)
|
|
},
|
|
}
|
|
bindChunksFlags(cmd, opts)
|
|
_ = cmd.MarkFlagRequired("kb")
|
|
agent.SetAgentHelp(cmd, "Hybrid retrieval; returns ranked chunk list. The server may include parent/nearby/relation chunks beyond match_count; --limit caps the returned slice client-side. Pass --no-vector or --no-keyword to disable a channel (mutually exclusive both-off).")
|
|
return cmd
|
|
}
|
|
|
|
// bindChunksFlags registers the chunks flag surface in one place to keep
|
|
// the constructor readable; --kb is marked required by the caller.
|
|
func bindChunksFlags(cmd *cobra.Command, opts *ChunksOptions) {
|
|
cmd.Flags().StringVar(&opts.KB, "kb", "", "Knowledge base UUID or name")
|
|
cmd.Flags().IntVarP(&opts.Limit, "limit", "L", 8, "Maximum results to return")
|
|
cmd.Flags().Float64Var(&opts.VectorThreshold, "vector-threshold", 0, "Vector retrieval similarity floor (per-channel, pre-fusion); 0 = no filter")
|
|
cmd.Flags().Float64Var(&opts.KeywordThreshold, "keyword-threshold", 0, "Keyword retrieval score floor (per-channel, pre-fusion); 0 = no filter")
|
|
cmd.Flags().BoolVar(&opts.NoVector, "no-vector", false, "Disable the vector channel")
|
|
cmd.Flags().BoolVar(&opts.NoKeyword, "no-keyword", false, "Disable the keyword channel")
|
|
cmd.Flags().BoolVar(&opts.JSONOut, "json", false, "Output JSON envelope")
|
|
}
|
|
|
|
// validate checks the option set before any SDK call.
|
|
func (o *ChunksOptions) validate() error {
|
|
if o.Query == "" {
|
|
return cmdutil.NewError(cmdutil.CodeInputInvalidArgument, "query argument cannot be empty")
|
|
}
|
|
if o.NoVector && o.NoKeyword {
|
|
return cmdutil.NewError(cmdutil.CodeInputInvalidArgument, "--no-vector and --no-keyword cannot both be set")
|
|
}
|
|
return nil
|
|
}
|
|
|
|
func runChunks(ctx context.Context, opts *ChunksOptions, svc ChunksService) error {
|
|
if err := opts.validate(); err != nil {
|
|
return err
|
|
}
|
|
if svc == nil {
|
|
return cmdutil.NewError(cmdutil.CodeServerError, "search chunks: no SDK client available")
|
|
}
|
|
|
|
params := &sdk.SearchParams{
|
|
QueryText: opts.Query,
|
|
MatchCount: opts.Limit,
|
|
VectorThreshold: opts.VectorThreshold,
|
|
KeywordThreshold: opts.KeywordThreshold,
|
|
DisableVectorMatch: opts.NoVector,
|
|
DisableKeywordsMatch: opts.NoKeyword,
|
|
}
|
|
results, err := svc.HybridSearch(ctx, opts.KBID, params)
|
|
if err != nil {
|
|
return cmdutil.Wrapf(cmdutil.ClassifyHTTPError(err), err, "hybrid search")
|
|
}
|
|
// match_count is the server's *primary-match* cap — after that, the
|
|
// service appends parent / nearby / relation chunks as context
|
|
// enrichment, so the wire response can exceed Limit. CLIs like gh /
|
|
// kubectl / aws treat their `--limit`-style flag as a hard return-count
|
|
// cap; honor that contract by trimming on the client. Recall isn't
|
|
// affected because the server's internal retrieval pool is already
|
|
// max(MatchCount*5, 50).
|
|
if opts.Limit > 0 && len(results) > opts.Limit {
|
|
results = results[:opts.Limit]
|
|
}
|
|
|
|
if opts.JSONOut {
|
|
return format.WriteEnvelope(iostreams.IO.Out, format.Success(results, &format.Meta{KBID: opts.KBID}))
|
|
}
|
|
return renderChunkResults(results, opts.KBID)
|
|
}
|
|
|
|
// renderChunkResults prints a compact pretty list. Minimal stopgap — a
|
|
// richer tabular renderer can replace this later without breaking the
|
|
// JSON contract.
|
|
func renderChunkResults(results []*sdk.SearchResult, kbID string) error {
|
|
if len(results) == 0 {
|
|
fmt.Fprintln(iostreams.IO.Out, "(no results)")
|
|
return nil
|
|
}
|
|
fmt.Fprintf(iostreams.IO.Out, "%d result(s) from kb=%s:\n\n", len(results), kbID)
|
|
for i, r := range results {
|
|
fmt.Fprintf(iostreams.IO.Out, "[%d] score=%.3f", i+1, r.Score)
|
|
if r.KnowledgeID != "" {
|
|
fmt.Fprintf(iostreams.IO.Out, " doc=%s", r.KnowledgeID)
|
|
}
|
|
fmt.Fprintln(iostreams.IO.Out)
|
|
fmt.Fprintln(iostreams.IO.Out, indent(strings.TrimSpace(r.Content), " "))
|
|
fmt.Fprintln(iostreams.IO.Out)
|
|
}
|
|
return nil
|
|
}
|
|
|
|
func indent(s, prefix string) string {
|
|
if s == "" {
|
|
return ""
|
|
}
|
|
lines := strings.Split(s, "\n")
|
|
for i, l := range lines {
|
|
lines[i] = prefix + l
|
|
}
|
|
return strings.Join(lines, "\n")
|
|
}
|