maskantech/pdf
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

fix(reflow): kill bulge/merge + font-change on edit; add center/right, IME; harden Raw Text

Browse Source 
Reflow (primary editor):
- Geometric backstop in reflow_paragraph: adopt any text object fully inside the paragraph
  bbox that the model's objectIndices omitted (PDFium glyph->object map can return -1), so it's
  deleted + its original font is resolved. Fixes both the leftover-glyph "bulge/merge" and the
  font-substitution-on-edit (same root cause). No-op on correctly-indexed paragraphs (gate-proven
  byte-identical: overlay diff unchanged at 1.69%/1.70%).
- Preserve data-fid on edited contentEditable nodes (extractFlatRuns climbs to nearest styled
  ancestor / inherits from adjacent run) so edits keep their real font instead of the dominant.
- Converge preview & commit through one buildReflowData (no preview-OK/commit-wrong drift).
- Center/right alignment: additive emission branch (greedy path only) + heading-align inference.
- IME composition handling (suppress render mid-composition; don't commit on composing Enter).
- Fix WASM document-handle leak (free superseded versions on documentId change/unmount).
- Quiet the backstop instrumentation (warn -> debug) now the root cause is confirmed.

Raw Text / StreamEditor (beta companion):
- Permission-gate both text_objects endpoints behind canModify; label tool "(beta)".
- P1a: preserve TJ kerning numbers for same-length edits (redistribute into original slots),
  fall back to single-string collapse otherwise (never worse than before).
- P1b: reject edits with characters unencodable in the run's font (clear 400 instead of corruption).
- P2a: commit Raw Text edits as a new document version (add_document) and adopt the id via
  pushHistory, so they join undo/redo instead of mutating in place.
- P2b: exact hit-boxes from real font metrics (re-measured once @font-face loads).

WASM preview rebuilt with the reflow changes; cache version bumped.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This commit is contained in:
Furqan-14
2026-06-17 17:12:53 +05:30
co-authored by Claude Opus 4.8
parent 21f0bfab9a
commit 660649a562
17 changed files with 707 additions and 98 deletions
Download Patch File Download Diff File Expand all files Collapse all files
engine/src/parser/pdfium_document.cpp
+62 -7
View File
@@ -2683,6 +2683,51 @@ std::expected<void, EngineError> PdfiumDocument::applyEdits(const std::string& e
return std::unexpected(EngineError::Unknown);
}
// --- Geometric backstop (fixes BOTH the "bulge/merge" and the font-substitution bug) ---
// The frontend-supplied objectIndices can omit a paragraph object when its glyphs map to
// pageObjectIndex == -1 (load-dependent, and the map shifts after a prior edit re-serializes
// the stream). A missing index means that object is (a) NOT deleted -> its glyphs stay UNDER
// the new text ("bulge/merge"), and (b) NOT seen by resolveOrigFont/getFontDataFromObjects
// -> the run re-emits in a base-14 substitute ("font changes on edit"). Backstop: widen the
// working set to any TEXT object whose bbox is FULLY inside the union bbox of the supplied
// indices. Conservative (contained text only) so a neighbouring paragraph / image / rule line
// is never touched. When the supplied indices are already complete (the correct common case)
// NO extra object is contained -> paragraphSet == objectIndices -> byte-identical behaviour.
std::vector<int> paragraphSet = objectIndices;
{
float ul = 0, ub = 0, ur = 0, ut = 0; bool haveUnion = false;
for (int idx : objectIndices) {
FPDF_PAGEOBJECT o = FPDFPage_GetObject(page, idx);
if (!o) continue;
float l = 0, b = 0, r = 0, t = 0;
if (!FPDFPageObj_GetBounds(o, &l, &b, &r, &t)) continue;
if (!haveUnion) { ul = l; ub = b; ur = r; ut = t; haveUnion = true; }
else { // plain comparisons (Windows headers #define min/max macros)
if (l < ul) ul = l; if (b < ub) ub = b;
if (r > ur) ur = r; if (t > ut) ut = t;
}
}
if (haveUnion) {
const float eps = 0.5f; // tolerate sub-pixel bbox slop
int nObjs = FPDFPage_CountObjects(page);
int adopted = 0;
for (int k = 0; k < nObjs; ++k) {
if (std::find(objectIndices.begin(), objectIndices.end(), k) != objectIndices.end()) continue;
FPDF_PAGEOBJECT o = FPDFPage_GetObject(page, k);
if (!o || FPDFPageObj_GetType(o) != FPDF_PAGEOBJ_TEXT) continue;
float l = 0, b = 0, r = 0, t = 0;
if (!FPDFPageObj_GetBounds(o, &l, &b, &r, &t)) continue;
if (l >= ul - eps && b >= ub - eps && r <= ur + eps && t <= ut + eps) {
paragraphSet.push_back(k);
adopted++;
spdlog::debug("reflow_paragraph: adopted leftover text object idx={} inside paragraph bbox (not in objectIndices) -> prevents bulge/merge + font substitution", k);
}
}
if (adopted > 0)
spdlog::debug("reflow_paragraph: geometric backstop adopted {} object(s) the model omitted", adopted);
}
}
// Resolve + load each run's emission font (subset over that run's codepoints).
std::vector<EmissionFont> runFonts(runs.size());
auto toCodepoints = [](const std::string& s) {
@@ -2717,7 +2762,7 @@ std::expected<void, EngineError> PdfiumDocument::applyEdits(const std::string& e
// -> "Cuwi Le,e"). The handle stays valid because `page` is open through emission.
auto resolveOrigFont = [&](const std::string& fid) -> FPDF_FONT {
const std::string expected = baseNameFromInternalFontId(fid);
for (int idx : objectIndices) {
for (int idx : paragraphSet) {
FPDF_PAGEOBJECT o = FPDFPage_GetObject(page, idx);
if (!o || FPDFPageObj_GetType(o) != FPDF_PAGEOBJ_TEXT) continue;
FPDF_FONT fo = FPDFTextObj_GetFont(o);
@@ -2733,7 +2778,7 @@ std::expected<void, EngineError> PdfiumDocument::applyEdits(const std::string& e
std::unordered_map<std::string, EmissionFont> fontByFid;
for (const auto& rs : runs) {
if (!fontByFid.count(rs.internalFontId))
fontByFid[rs.internalFontId] = loadEmissionFont(pageIndex, rs.internalFontId, rs.fontSize, fontCps[rs.internalFontId], objectIndices, resolveOrigFont(rs.internalFontId));
fontByFid[rs.internalFontId] = loadEmissionFont(pageIndex, rs.internalFontId, rs.fontSize, fontCps[rs.internalFontId], paragraphSet, resolveOrigFont(rs.internalFontId));
}
for (size_t ri = 0; ri < runs.size(); ++ri) runFonts[ri] = fontByFid[runs[ri].internalFontId];
@@ -2882,7 +2927,7 @@ std::expected<void, EngineError> PdfiumDocument::applyEdits(const std::string& e
int nObjs = FPDFPage_CountObjects(page);
int pushed = 0;
for (int k = 0; k < nObjs; ++k) {
if (std::find(objectIndices.begin(), objectIndices.end(), k) != objectIndices.end()) continue;
if (std::find(paragraphSet.begin(), paragraphSet.end(), k) != paragraphSet.end()) continue;
FPDF_PAGEOBJECT o = FPDFPage_GetObject(page, k);
if (!o) continue;
float l = 0, bo = 0, rr = 0, tt = 0;
@@ -2896,10 +2941,12 @@ std::expected<void, EngineError> PdfiumDocument::applyEdits(const std::string& e
spdlog::info("reflow_paragraph: lines {}->{}, deltaH={}, pushed {} objects", oldLineCount, newLineCount, deltaH, pushed);
}
// Delete the old paragraph objects (descending so indices stay valid).
std::sort(objectIndices.begin(), objectIndices.end(), std::greater<int>());
int minIndex = objectIndices.back();
for (int idx : objectIndices) {
// Delete the old paragraph objects (descending so indices stay valid). Uses paragraphSet
// so any object the model omitted but that the geometric backstop adopted is also removed
// (otherwise its glyphs would remain UNDER the freshly-emitted text -> "bulge/merge").
std::sort(paragraphSet.begin(), paragraphSet.end(), std::greater<int>());
int minIndex = paragraphSet.back();
for (int idx : paragraphSet) {
FPDF_PAGEOBJECT o = FPDFPage_GetObject(page, idx);
if (o) { FPDFPage_RemoveObject(page, o); FPDFPageObj_Destroy(o); }
}
@@ -2927,6 +2974,14 @@ std::expected<void, EngineError> PdfiumDocument::applyEdits(const std::string& e
std::vector<double> adv; // advance (PDF units) of each char in lineText
double lineFontSize = 0.0;
double x = (li < lineX.size()) ? lineX[li] : columnLeft; // exact source left when provided
// Center/right alignment: shift the whole line within the column. Only on the
// greedy/re-wrapped path (no provided lineX) so the unchanged "lines" path and the
// existing left/justify behaviour are byte-identical. justifyThis is already false
// for center/right (it only triggers on align=="justify"), so no gap distribution.
if (li >= lineX.size()) {
if (align == "right") x = columnLeft + (columnWidth - naturalW);
else if (align == "center") x = columnLeft + (columnWidth - naturalW) / 2.0;
}
for (size_t k = 0; k < lw.size(); ++k) {
size_t wi = lw[k];
if (k > 0) {