@@ -0,0 +1,225 @@ |
| 1 | +// SPDX-License-Identifier: AGPL-3.0-or-later |
| 2 | + |
| 3 | +package jobs |
| 4 | + |
| 5 | +import ( |
| 6 | + "bytes" |
| 7 | + "context" |
| 8 | + "encoding/json" |
| 9 | + "errors" |
| 10 | + "fmt" |
| 11 | + "log/slog" |
| 12 | + "strings" |
| 13 | + |
| 14 | + "github.com/jackc/pgx/v5" |
| 15 | + "github.com/jackc/pgx/v5/pgtype" |
| 16 | + "github.com/jackc/pgx/v5/pgxpool" |
| 17 | + |
| 18 | + "github.com/tenseleyFlow/shithub/internal/infra/storage" |
| 19 | + repogit "github.com/tenseleyFlow/shithub/internal/repos/git" |
| 20 | + reposdb "github.com/tenseleyFlow/shithub/internal/repos/sqlc" |
| 21 | + usersdb "github.com/tenseleyFlow/shithub/internal/users/sqlc" |
| 22 | + "github.com/tenseleyFlow/shithub/internal/worker" |
| 23 | +) |
| 24 | + |
| 25 | +// IndexCodeDeps wires the job. |
| 26 | +type IndexCodeDeps struct { |
| 27 | + Pool *pgxpool.Pool |
| 28 | + RepoFS *storage.RepoFS |
| 29 | + Logger *slog.Logger |
| 30 | +} |
| 31 | + |
| 32 | +// IndexCodePayload is enqueued by the push-process job (or the |
| 33 | +// reconciler) when the default branch advances. |
| 34 | +type IndexCodePayload struct { |
| 35 | + RepoID int64 `json:"repo_id"` |
| 36 | +} |
| 37 | + |
| 38 | +// Code-search size limits, per the spec: |
| 39 | +// - Files > maxFileBytes are skipped from content indexing (path |
| 40 | +// stays indexed regardless). |
| 41 | +// - Per-file indexed content is truncated to maxIndexBytes so the |
| 42 | +// trigram column doesn't bloat for huge text files. |
| 43 | +const ( |
| 44 | + maxFileBytes = 256 * 1024 // 256 KiB |
| 45 | + maxIndexBytes = 64 * 1024 // 64 KiB |
| 46 | +) |
| 47 | + |
| 48 | +// RepoIndexCode walks the repo's default branch tree, indexes paths |
| 49 | +// for every file and content for files that fit the size + textness |
| 50 | +// gates, then atomically swaps the new index in via delete-then- |
| 51 | +// insert in a single tx. Readers never see a partial index. |
| 52 | +// |
| 53 | +// Tracks `repos.last_indexed_oid` so the reconciler can detect drift. |
| 54 | +func RepoIndexCode(deps IndexCodeDeps) worker.Handler { |
| 55 | + return func(ctx context.Context, raw json.RawMessage) error { |
| 56 | + var p IndexCodePayload |
| 57 | + if err := json.Unmarshal(raw, &p); err != nil { |
| 58 | + return worker.PoisonError(fmt.Errorf("bad payload: %w", err)) |
| 59 | + } |
| 60 | + if p.RepoID == 0 { |
| 61 | + return worker.PoisonError(errors.New("missing repo_id")) |
| 62 | + } |
| 63 | + |
| 64 | + rq := reposdb.New() |
| 65 | + uq := usersdb.New() |
| 66 | + repo, err := rq.GetRepoByID(ctx, deps.Pool, p.RepoID) |
| 67 | + if err != nil { |
| 68 | + if errors.Is(err, pgx.ErrNoRows) { |
| 69 | + return worker.PoisonError(fmt.Errorf("repo %d not found", p.RepoID)) |
| 70 | + } |
| 71 | + return err |
| 72 | + } |
| 73 | + if repo.DeletedAt.Valid { |
| 74 | + // Repo went away between enqueue and now. Nothing to do. |
| 75 | + return nil |
| 76 | + } |
| 77 | + if !repo.OwnerUserID.Valid { |
| 78 | + return worker.PoisonError(fmt.Errorf("repo %d has no user owner (org-owned arrives in S31)", repo.ID)) |
| 79 | + } |
| 80 | + owner, err := uq.GetUserByID(ctx, deps.Pool, repo.OwnerUserID.Int64) |
| 81 | + if err != nil { |
| 82 | + return err |
| 83 | + } |
| 84 | + gitDir, err := deps.RepoFS.RepoPath(owner.Username, repo.Name) |
| 85 | + if err != nil { |
| 86 | + return err |
| 87 | + } |
| 88 | + |
| 89 | + ref := repo.DefaultBranch |
| 90 | + // Resolve the current OID for the default branch. If empty |
| 91 | + // (no commits yet) clear any prior index and bail. |
| 92 | + oid, err := repogit.ResolveRefOID(ctx, gitDir, ref) |
| 93 | + if err != nil { |
| 94 | + if errors.Is(err, repogit.ErrRefNotFound) { |
| 95 | + if err := clearRepoIndex(ctx, deps.Pool, repo.ID); err != nil { |
| 96 | + return err |
| 97 | + } |
| 98 | + _ = rq.SetLastIndexedOID(ctx, deps.Pool, reposdb.SetLastIndexedOIDParams{ |
| 99 | + ID: repo.ID, LastIndexedOid: pgtype.Text{Valid: false}, |
| 100 | + }) |
| 101 | + return nil |
| 102 | + } |
| 103 | + return fmt.Errorf("resolve %s: %w", ref, err) |
| 104 | + } |
| 105 | + |
| 106 | + // Walk the tree. |
| 107 | + paths, err := repogit.ListAllPaths(ctx, gitDir, ref) |
| 108 | + if err != nil { |
| 109 | + return fmt.Errorf("ls-tree: %w", err) |
| 110 | + } |
| 111 | + |
| 112 | + // Read each blob; classify size + text. Skipped paths still |
| 113 | + // land in the path index (only content is gated by size). |
| 114 | + type indexed struct { |
| 115 | + path string |
| 116 | + content []byte // empty if skipped from content index |
| 117 | + } |
| 118 | + entries := make([]indexed, 0, len(paths)) |
| 119 | + for _, path := range paths { |
| 120 | + if shouldSkipPath(path) { |
| 121 | + continue |
| 122 | + } |
| 123 | + ent := indexed{path: path} |
| 124 | + blob, err := repogit.ReadBlobBytes(ctx, gitDir, ref, path, maxFileBytes+1) |
| 125 | + if err == nil && len(blob) <= maxFileBytes && isText(blob) { |
| 126 | + if len(blob) > maxIndexBytes { |
| 127 | + blob = blob[:maxIndexBytes] |
| 128 | + } |
| 129 | + ent.content = blob |
| 130 | + } |
| 131 | + entries = append(entries, ent) |
| 132 | + } |
| 133 | + |
| 134 | + // Atomic swap: delete + insert in one tx. |
| 135 | + tx, err := deps.Pool.Begin(ctx) |
| 136 | + if err != nil { |
| 137 | + return err |
| 138 | + } |
| 139 | + committed := false |
| 140 | + defer func() { |
| 141 | + if !committed { |
| 142 | + _ = tx.Rollback(ctx) |
| 143 | + } |
| 144 | + }() |
| 145 | + if _, err := tx.Exec(ctx, |
| 146 | + `DELETE FROM code_search_paths WHERE repo_id = $1`, repo.ID); err != nil { |
| 147 | + return err |
| 148 | + } |
| 149 | + if _, err := tx.Exec(ctx, |
| 150 | + `DELETE FROM code_search_content WHERE repo_id = $1`, repo.ID); err != nil { |
| 151 | + return err |
| 152 | + } |
| 153 | + for _, e := range entries { |
| 154 | + if _, err := tx.Exec(ctx, ` |
| 155 | + INSERT INTO code_search_paths (repo_id, ref_name, path, tsv) |
| 156 | + VALUES ($1, $2, $3, to_tsvector('shithub_search', $3)) |
| 157 | + ON CONFLICT DO NOTHING |
| 158 | + `, repo.ID, ref, e.path); err != nil { |
| 159 | + return fmt.Errorf("insert path %s: %w", e.path, err) |
| 160 | + } |
| 161 | + if e.content != nil { |
| 162 | + content := string(e.content) |
| 163 | + if _, err := tx.Exec(ctx, ` |
| 164 | + INSERT INTO code_search_content |
| 165 | + (repo_id, ref_name, path, content_tsv, content_trgm) |
| 166 | + VALUES ($1, $2, $3, to_tsvector('shithub_search', $4), $4) |
| 167 | + ON CONFLICT DO NOTHING |
| 168 | + `, repo.ID, ref, e.path, content); err != nil { |
| 169 | + return fmt.Errorf("insert content %s: %w", e.path, err) |
| 170 | + } |
| 171 | + } |
| 172 | + } |
| 173 | + if err := rq.SetLastIndexedOID(ctx, tx, reposdb.SetLastIndexedOIDParams{ |
| 174 | + ID: repo.ID, LastIndexedOid: pgtype.Text{String: oid, Valid: true}, |
| 175 | + }); err != nil { |
| 176 | + return err |
| 177 | + } |
| 178 | + if err := tx.Commit(ctx); err != nil { |
| 179 | + return err |
| 180 | + } |
| 181 | + committed = true |
| 182 | + return nil |
| 183 | + } |
| 184 | +} |
| 185 | + |
| 186 | +// clearRepoIndex drops every code-search row for a repo. Used when |
| 187 | +// the default branch is gone (deleted, or never created) so we |
| 188 | +// don't leave stale rows behind. |
| 189 | +func clearRepoIndex(ctx context.Context, pool *pgxpool.Pool, repoID int64) error { |
| 190 | + if _, err := pool.Exec(ctx, |
| 191 | + `DELETE FROM code_search_paths WHERE repo_id = $1`, repoID); err != nil { |
| 192 | + return err |
| 193 | + } |
| 194 | + if _, err := pool.Exec(ctx, |
| 195 | + `DELETE FROM code_search_content WHERE repo_id = $1`, repoID); err != nil { |
| 196 | + return err |
| 197 | + } |
| 198 | + return nil |
| 199 | +} |
| 200 | + |
| 201 | +// shouldSkipPath filters out paths the spec calls out as |
| 202 | +// "skipped by default": vendor/, node_modules/, dist/, .git*. The |
| 203 | +// `path:` operator (post-MVP) lets users opt into them. |
| 204 | +func shouldSkipPath(p string) bool { |
| 205 | + if strings.HasPrefix(p, ".git") { |
| 206 | + return true |
| 207 | + } |
| 208 | + for _, prefix := range []string{"vendor/", "node_modules/", "dist/"} { |
| 209 | + if strings.HasPrefix(p, prefix) || strings.Contains(p, "/"+prefix) { |
| 210 | + return true |
| 211 | + } |
| 212 | + } |
| 213 | + return false |
| 214 | +} |
| 215 | + |
| 216 | +// isText is the same NUL-byte heuristic S17 uses for the blob view: |
| 217 | +// any NUL in the first 8 KiB → binary; otherwise text. Cheap, good |
| 218 | +// enough for code-search content gating. |
| 219 | +func isText(b []byte) bool { |
| 220 | + limit := len(b) |
| 221 | + if limit > 8192 { |
| 222 | + limit = 8192 |
| 223 | + } |
| 224 | + return bytes.IndexByte(b[:limit], 0) < 0 |
| 225 | +} |