feat(pipeline): feed the current source page into the extraction prompt (#106) #132

Merged
mathias merged 4 commits from feat/106-page-aware-extraction-v2 into main 2026-09-23 05:45:47 +00:00
2 changed files with 59 additions and 14 deletions
Showing only changes of commit bdba49b756 - Show all commits
+29 -14
View File
@@ -65,29 +65,44 @@ func MergeReport(a, b Page) (Page, []string) {
var conflicts []string
// First occurrence wins on a collision — a page can already carry two
// on-disk sections that only differ by case or whitespace (the exact
// corpus drift this normalization exists to catch going forward, found
// pre-existing in review of PR #130): a plain last-write assignment
// here would silently and deterministically route every future update
// to whichever duplicate happens to sort last in the file, freezing
// the other one forever with nothing to say so.
idx := make(map[string]int, len(secsA))
// Sections in secsA sharing a normalized heading (pre-existing corpus
// drift — "## Notes" and "## NOTES" both already on disk, the exact
// pattern this normalization exists to catch going forward, found in
// review of PR #130/#132) are matched to b's occurrences of that same
// heading POSITIONALLY: b's 1st occurrence of the key pairs with a's
// 1st, b's 2nd with a's 2nd, and so on. A single "last write wins" (or
// even "first write wins") map from key to one index would instead
// route every one of b's occurrences to the SAME on-disk section —
// which either freezes every duplicate but one (never receiving
// updates again) or, worse, compares mismatched pairs and reports a
// disagreement between two sections that were never supposed to be
// compared, false-positiving on the plain self-merge case a revision
// pass produces when the model echoes a page back unchanged (found in
// review of PR #132).
byKey := make(map[string][]int, len(secsA))
for i, s := range secsA {
key := normalizeHeading(s.heading)
if _, exists := idx[key]; !exists {
idx[key] = i
}
byKey[key] = append(byKey[key], i)
}
cursor := make(map[string]int, len(byKey))
for _, sB := range secsB {
key := normalizeHeading(sB.heading)
i, exists := idx[key]
if !exists {
idx[key] = len(secsA)
indices := byKey[key]
pos := cursor[key]
cursor[key] = pos + 1
if pos >= len(indices) {
// b has more occurrences of this heading than a does (including
// the case where a has none at all) — the extra is new content,
// appended as its own section and registered so a THIRD
// occurrence of the same heading in b pairs with this one
// rather than appending yet another duplicate.
byKey[key] = append(byKey[key], len(secsA))
secsA = append(secsA, sB)
continue
}
i := indices[pos]
sA := secsA[i]
switch {
case bulletSectionsNorm[key]:
+30
View File
@@ -367,6 +367,36 @@ func TestMerge_PreExistingCaseDuplicateHeadingsRouteToTheFirstOccurrence(t *test
"the second duplicate is untouched by this update, not merged into or overwritten")
}
func TestMergeReport_SelfMergeOfPreExistingDuplicateHeadingsReportsNoConflict(t *testing.T) {
// Found in review of PR #132: first-occurrence-wins routes ALL of b's
// occurrences of a normalized heading to a's FIRST occurrence, not to
// their own positional counterpart. So merging a page against itself
// (or an echoed-back copy, exactly what #106's revision prompt can
// produce) pairs b's second "## NOTES" against a's first "## Notes" —
// different content, same normalized key — and MergeReport wrongly
// reports a disagreement where none exists. Fixed by pairing
// positionally: b's Nth occurrence of a normalized heading matches a's
// Nth occurrence, not always the first.
page := Page{Path: "p.md", Content: "---\ntitle: X\n---\n\n## Notes\n\nFirst note.\n\n## NOTES\n\nSecond note.\n"}
_, conflicts := MergeReport(page, page)
assert.Empty(t, conflicts, "merging a page with pre-existing duplicate headings against itself must report no conflicts")
}
func TestMergeReport_DuplicateHeadingsPairPositionallyNotAllToFirst(t *testing.T) {
// Companion case: b has genuinely different content in its second
// occurrence of the normalized heading — that must compare against a's
// second occurrence (and report if they disagree), not against a's
// first (which would either false-positive if a's first differs, or
// false-negative-mask a real disagreement in the second pair).
a := Page{Path: "p.md", Content: "---\ntitle: X\n---\n\n## Notes\n\nFirst note.\n\n## NOTES\n\nSecond note.\n"}
b := Page{Path: "p.md", Content: "---\ntitle: X\n---\n\n## Notes\n\nFirst note.\n\n## NOTES\n\nSecond note, revised.\n"}
_, conflicts := MergeReport(a, b)
assert.Equal(t, []string{"NOTES"}, conflicts,
"only the second occurrence, which actually disagrees, is reported")
}
func TestRefreshOwned_CommentLinesSurviveUnmangled(t *testing.T) {
// Third review round on PR #126: parseFrontmatterBlock dropped the
// sibling api package's comment guard (strings.HasPrefix(trimmed, "#")).