Merge pull request #130 from jpirnay/fix-kosync-upload

fix: Reestablish 1.37 kosync upload behaviour

docs/contributing/koreader-sync-xpath-mapping.md

@@ -38,12 +38,16 @@ via a KOReader contributor mapping spine items to DocFragment numbers.
 Implemented in `ProgressMapper::toKOReader`.
 
 1. Compute overall `percentage` from chapter/page.
-2. If a paragraph index is available from the section cache LUT (`CrossPointPosition::hasParagraphIndex`),
+2. Generate XPath via byte-offset estimation (`ChapterXPathIndexer::findXPathForProgress`),
-   generate an XPath directly: `/body/DocFragment[spineIndex + 1]/body/p[paragraphIndex]`.
+   producing a `…/text()[K].M` anchor proportional to intra-spine progress.
-3. Otherwise, attempt byte-offset estimation via `ChapterXPathIndexer::findXPathForProgress`.
+3. If XPath extraction fails, fallback to synthetic chapter path:
-4. If XPath extraction fails, fallback to synthetic chapter path:
    - `/body/DocFragment[spineIndex + 1]/body`
 
+The paragraph LUT (see below) is intentionally **not** used for upload: snapping to the
+start of `p[N]` when the user is mid-paragraph causes pulled positions to land at the
+start of the paragraph (and at the start of the chapter when an opening paragraph spans
+many pages). The LUT remains in use for the reverse direction.
+
 ### KOReader -> CrossPoint
 
 Implemented in `ProgressMapper::toCrossPoint`.

lib/KOReaderSync/ProgressMapper.cpp

@@ -57,24 +57,11 @@ KOReaderPosition ProgressMapper::toKOReader(const std::shared_ptr<Epub>& epub, c
   // Calculate overall book progress (0.0-1.0)
   result.percentage = epub->calculateProgress(pos.spineIndex, intraSpineProgress);
 
-  // Generate XPath for the current position.
+  // Generate XPath for the current position via byte-offset scan. Targeting the
-  // When we have a paragraph index from the section LUT, target that specific <p> element
+  // paragraph LUT entry instead would snap to the start of the paragraph the user
-  // directly — this produces a structurally precise full-ancestry path even for chapters
+  // is inside, which causes pulled positions to land at the start of the chapter
-  // where paragraphs are nested inside divs/sections.  Fall back to the progress-based
+  // when an opening paragraph spans many pages.
-  // scan (which works for any content) when no paragraph index is available.
+  result.xpath = ChapterXPathIndexer::findXPathForProgress(epub, pos.spineIndex, intraSpineProgress);
-  if (pos.hasParagraphIndex && pos.paragraphIndex > 0) {
-    // When a seek hint is set, the LUT entry's paragraphIndex equals pos.paragraphIndex
-    // (both describe the same page). The byte offset now points at the body-child element
-    // that was current at the page break, so re-parsing from there will re-encounter that
-    // paragraph — seed startParagraphCount with paragraphIndex-1 to avoid double counting.
-    const uint16_t startCount =
-        pos.xhtmlSeekHint > 0 && pos.paragraphIndex > 0 ? static_cast<uint16_t>(pos.paragraphIndex - 1) : 0;
-    result.xpath = ChapterXPathIndexer::findXPathForParagraph(epub, pos.spineIndex, pos.paragraphIndex,
-                                                              pos.xhtmlSeekHint, startCount);
-  }
-  if (result.xpath.empty()) {
-    result.xpath = ChapterXPathIndexer::findXPathForProgress(epub, pos.spineIndex, intraSpineProgress);
-  }
   if (result.xpath.empty()) {
     result.xpath = generateXPath(pos.spineIndex);
   }