@@ -13,20 +13,25 @@ import ( |
| 13 | 13 | ) |
| 14 | 14 | |
| 15 | 15 | // RemapAllForPR re-anchors every non-draft review comment on a PR |
| 16 | | -// against the new (base, head) snapshot. For each comment: |
| 16 | +// against the new (base, head) snapshot. For each comment we do a |
| 17 | +// content-aware check: |
| 17 | 18 | // |
| 18 | | -// 1. Re-walk the diff for its file at (newBase..newHead). |
| 19 | | -// 2. If the line at original_line still appears on the chosen side |
| 20 | | -// (left = base, right = head), set current_position to its new |
| 21 | | -// position index. Otherwise NULL → outdated. |
| 19 | +// 1. Read the line text at (original_commit_sha, file_path, |
| 20 | +// original_line). This is what the comment was written against. |
| 21 | +// 2. Read the same line number in the new blob (newHeadOID for |
| 22 | +// side=right; newBaseOID for side=left). |
| 23 | +// 3. If the bytes are identical → keep current_position = original_line. |
| 24 | +// If not → current_position = NULL (outdated). The comment still |
| 25 | +// renders in the conversation timeline; the Files tab hides it |
| 26 | +// until the user clicks "Show outdated." |
| 22 | 27 | // |
| 23 | | -// This is the conservative v1 implementation. The spec calls out |
| 28 | +// This is the conservative v1 mapper. Lines that have been re-indented, |
| 29 | +// shifted by an insertion above, or merely had a comma added all |
| 30 | +// outdate — that's the right default. The spec calls out |
| 24 | 31 | // `git blame --porcelain` as a richer mapper for rebase-heavy PRs; |
| 25 | | -// add that when the simple line-presence check proves insufficient. |
| 32 | +// add that when the simple presence check proves too aggressive. |
| 26 | 33 | // |
| 27 | | -// The function reads `pr_review_comments` then issues per-row |
| 28 | | -// SetPRReviewCommentCurrentPosition updates. Idempotent: re-running |
| 29 | | -// converges on the same answer. |
| 34 | +// Idempotent: re-running converges on the same answer. |
| 30 | 35 | func RemapAllForPR(ctx context.Context, deps Deps, gitDir string, prID int64, newBaseOID, newHeadOID string) error { |
| 31 | 36 | if newBaseOID == "" || newHeadOID == "" { |
| 32 | 37 | return nil |
@@ -40,47 +45,41 @@ func RemapAllForPR(ctx context.Context, deps Deps, gitDir string, prID int64, ne |
| 40 | 45 | return nil |
| 41 | 46 | } |
| 42 | 47 | |
| 43 | | - // Build per-file caches keyed on (file_path, side). Each entry |
| 44 | | - // stores the line-text → position map for that side of the new |
| 45 | | - // diff. v1 reads file contents at the relevant snapshot OID and |
| 46 | | - // computes positions on demand. |
| 47 | | - type sideKey struct { |
| 48 | + // Per-(ref, path) blob cache so we read each blob at most once per |
| 49 | + // PR, regardless of how many comments anchor into it. |
| 50 | + type blobKey struct { |
| 51 | + ref string |
| 48 | 52 | path string |
| 49 | | - side string |
| 50 | 53 | } |
| 51 | | - cache := map[sideKey]map[int]int{} // line# (from original) → new position |
| 52 | | - |
| 53 | | - getMap := func(path, side string) (map[int]int, error) { |
| 54 | | - k := sideKey{path, side} |
| 55 | | - if m, ok := cache[k]; ok { |
| 56 | | - return m, nil |
| 57 | | - } |
| 58 | | - // For the v1 mapper we just compute position-by-line via the |
| 59 | | - // snapshot file content. side=right uses newHeadOID; side=left |
| 60 | | - // uses newBaseOID. |
| 61 | | - ref := newHeadOID |
| 62 | | - if side == "left" { |
| 63 | | - ref = newBaseOID |
| 54 | + cache := map[blobKey][]byte{} |
| 55 | + loadBlob := func(ref, path string) []byte { |
| 56 | + k := blobKey{ref, path} |
| 57 | + if b, ok := cache[k]; ok { |
| 58 | + return b |
| 64 | 59 | } |
| 65 | | - // Inline cap on file size — we don't bother mapping files > |
| 66 | | - // 1 MiB; those comments outdate. |
| 60 | + // We don't bother mapping files > 1 MiB; those comments outdate. |
| 67 | 61 | const maxBytes = 1 << 20 |
| 68 | 62 | blob, err := repogit.ReadBlobBytes(ctx, gitDir, ref, path, maxBytes) |
| 69 | 63 | if err != nil { |
| 70 | | - cache[k] = nil // miss → outdate everything in this file |
| 71 | | - return nil, nil |
| 64 | + cache[k] = nil |
| 65 | + return nil |
| 72 | 66 | } |
| 73 | | - m := buildLineMap(blob) |
| 74 | | - cache[k] = m |
| 75 | | - return m, nil |
| 67 | + cache[k] = blob |
| 68 | + return blob |
| 76 | 69 | } |
| 77 | 70 | |
| 78 | 71 | for _, c := range rows { |
| 79 | | - m, _ := getMap(c.FilePath, string(c.Side)) |
| 72 | + newRef := newHeadOID |
| 73 | + if c.Side == pullsdb.PrReviewSideLeft { |
| 74 | + newRef = newBaseOID |
| 75 | + } |
| 76 | + original := loadBlob(c.OriginalCommitSha, c.FilePath) |
| 77 | + current := loadBlob(newRef, c.FilePath) |
| 78 | + |
| 80 | 79 | var newPos pgtype.Int4 |
| 81 | | - if m != nil { |
| 82 | | - if pos, ok := m[int(c.OriginalLine)]; ok { |
| 83 | | - newPos = pgtype.Int4{Int32: int32(pos), Valid: true} |
| 80 | + if line, ok := lineAt(original, int(c.OriginalLine)); ok { |
| 81 | + if cur, ok := lineAt(current, int(c.OriginalLine)); ok && bytesEqual(line, cur) { |
| 82 | + newPos = pgtype.Int4{Int32: c.OriginalLine, Valid: true} |
| 84 | 83 | } |
| 85 | 84 | } |
| 86 | 85 | if err := q.SetPRReviewCommentCurrentPosition(ctx, deps.Pool, pullsdb.SetPRReviewCommentCurrentPositionParams{ |
@@ -93,25 +92,41 @@ func RemapAllForPR(ctx context.Context, deps Deps, gitDir string, prID int64, ne |
| 93 | 92 | return nil |
| 94 | 93 | } |
| 95 | 94 | |
| 96 | | -// buildLineMap returns line-number → position-index for a blob. v1 |
| 97 | | -// just maps line N to position N — line numbers in the blob *are* |
| 98 | | -// the positions for the simple "line still exists" check. The map |
| 99 | | -// will be replaced with a content-aware mapping when we add the |
| 100 | | -// blame-based variant. |
| 101 | | -func buildLineMap(blob []byte) map[int]int { |
| 102 | | - out := map[int]int{} |
| 103 | | - line := 1 |
| 104 | | - pos := 1 |
| 95 | +// lineAt returns the bytes of line N (1-indexed) in blob, excluding |
| 96 | +// the trailing newline. Returns (nil, false) when the blob is nil/empty |
| 97 | +// or N is out of range. |
| 98 | +func lineAt(blob []byte, n int) ([]byte, bool) { |
| 99 | + if blob == nil || n < 1 { |
| 100 | + return nil, false |
| 101 | + } |
| 102 | + start, line := 0, 1 |
| 105 | 103 | for i := 0; i < len(blob); i++ { |
| 106 | 104 | if blob[i] == '\n' { |
| 107 | | - out[line] = pos |
| 105 | + if line == n { |
| 106 | + return blob[start:i], true |
| 107 | + } |
| 108 | 108 | line++ |
| 109 | | - pos++ |
| 109 | + start = i + 1 |
| 110 | 110 | } |
| 111 | 111 | } |
| 112 | | - // Trailing line without newline. |
| 113 | | - if len(blob) > 0 && blob[len(blob)-1] != '\n' { |
| 114 | | - out[line] = pos |
| 112 | + // Trailing line without terminator. |
| 113 | + if line == n && start < len(blob) { |
| 114 | + return blob[start:], true |
| 115 | + } |
| 116 | + return nil, false |
| 117 | +} |
| 118 | + |
| 119 | +// bytesEqual is a tiny escape hatch so the test fixtures stay readable. |
| 120 | +// `bytes.Equal` would do, but pulling the whole package in for one call |
| 121 | +// adds noise to the position-map dependency graph. |
| 122 | +func bytesEqual(a, b []byte) bool { |
| 123 | + if len(a) != len(b) { |
| 124 | + return false |
| 125 | + } |
| 126 | + for i := range a { |
| 127 | + if a[i] != b[i] { |
| 128 | + return false |
| 129 | + } |
| 115 | 130 | } |
| 116 | | - return out |
| 131 | + return true |
| 117 | 132 | } |