Skip to content

RAGFlow is an agent-native database: a Retrieval-Augmented Generation (RAG) engine that grew into context infrastructure.

feat: locate Excel citation sources in document preview (#18853)

天海蒼灆61298397f691
diff --git a/deepdoc/parser/excel_parser.py b/deepdoc/parser/excel_parser.py
index bc9edf57c2..eef73fb97e 100644
--- a/deepdoc/parser/excel_parser.py
+++ b/deepdoc/parser/excel_parser.py
@@ -213,7 +213,7 @@ class RAGFlowExcelParser:
return ""
return str(v).strip()
- for sheetname in wb.sheetnames:
+ for sheet_idx, sheetname in enumerate(wb.sheetnames):
ws = wb[sheetname]
try:
rows = RAGFlowExcelParser._get_rows_limited(ws)
@@ -228,6 +228,7 @@ class RAGFlowExcelParser:
for t in list(rows[0]):
tb_rows_0 += f"<th>{escape(_fmt(t.value))}</th>"
tb_rows_0 += "</tr>"
+ col_max = max((len(r) for r in rows), default=1)
# rows[0] is the header; split the remaining data rows into
# ceil(n_data / chunk_rows) chunks. Using +1 here over-counts by one
@@ -235,10 +236,12 @@ class RAGFlowExcelParser:
# a spurious header-only chunk.
n_data_rows = len(rows) - 1
for chunk_i in range((n_data_rows + chunk_rows - 1) // chunk_rows):
+ row_start = 2 + chunk_i * chunk_rows
+ row_end = min(1 + (chunk_i + 1) * chunk_rows, len(rows))
tb = ""
tb += f"<table><caption>{sheetname}</caption>"
tb += tb_rows_0
- for r in list(rows[1 + chunk_i * chunk_rows : min(1 + (chunk_i + 1) * chunk_rows, len(rows))]):
+ for r in list(rows[1 + chunk_i * chunk_rows : row_end]):
tb += "<tr>"
for i, c in enumerate(r):
if c.value is None:
@@ -247,7 +250,9 @@ class RAGFlowExcelParser:
tb += f"<td>{escape(_fmt(c.value))}</td>"
tb += "</tr>"
tb += "</table>\n"
- tb_chunks.append(tb)
+ # position: (sheet_idx 0-based, row_start, row_end, col_start, col_end)
+ # for add_positions which increments the first component to 1-based.
+ tb_chunks.append((tb, (sheet_idx, row_start, row_end, 1, col_max)))
return tb_chunks
@@ -270,7 +275,7 @@ class RAGFlowExcelParser:
wb = RAGFlowExcelParser._load_excel_to_workbook(file_like_object)
res = []
- for sheetname in wb.sheetnames:
+ for sheet_idx, sheetname in enumerate(wb.sheetnames):
ws = wb[sheetname]
try:
rows = RAGFlowExcelParser._get_rows_limited(ws)
@@ -280,11 +285,15 @@ class RAGFlowExcelParser:
if not rows:
continue
ti = list(rows[0])
- for r in list(rows[1:]):
+ for row_idx, r in enumerate(list(rows[1:]), start=2):
fields = []
+ col_min, col_max = None, None
for i, c in enumerate(r):
if c.value is None or str(c.value).strip() == "":
continue
+ col = i + 1
+ col_min = col if col_min is None else min(col_min, col)
+ col_max = col if col_max is None else max(col_max, col)
t = str(ti[i].value) if i < len(ti) else ""
t += (":" if t else "") + str(c.value)
fields.append(t)
@@ -293,7 +302,7 @@ class RAGFlowExcelParser:
line = "; ".join(fields)
if sheetname.lower().find("sheet") < 0:
line += " ——" + sheetname
- res.append(line)
+ res.append((line, (sheet_idx, row_idx, row_idx, col_min or 1, col_max or 1)))
return res
@staticmethod
diff --git a/internal/parser/parser/office_table_render.go b/internal/parser/parser/office_table_render.go
index bc8153c05d..3e63539362 100644
--- a/internal/parser/parser/office_table_render.go
+++ b/internal/parser/parser/office_table_render.go
@@ -151,6 +151,16 @@ func buildHeaderRow(row []string) string {
return b.String()
}
+// htmlTableChunk is one self-contained <table> chunk with 1-based sheet/row/col
+// coordinates for source location (mirrors Excel position_int semantics).
+type htmlTableChunk struct {
+ HTML string
+ RowStart int // inclusive, 1-based Excel row of first data row (header-only chunks use headerRowAbs)
+ RowEnd int // inclusive, 1-based Excel row of last data row
+ ColStart int
+ ColEnd int
+}
+
// recordsToHTMLTableChunks renders records as one or more self-contained HTML
// <table> chunks. The first row is always the header (<th>). Data rows are
// split into chunks of chunkRows, each chunk being a complete <table> with
@@ -161,22 +171,48 @@ func buildHeaderRow(row []string) string {
// <thead>/<tbody>, so every <table> is one atomic chunk that downstream
// chunkers can consume independently.
func recordsToHTMLTableChunks(records [][]string, chunkRows int, caption string) string {
- return strings.Join(recordsToHTMLTableChunkList(records, chunkRows, caption), "")
+ chunks := recordsToHTMLTableChunkList(records, chunkRows, caption, 1)
+ parts := make([]string, len(chunks))
+ for i, ch := range chunks {
+ parts[i] = ch.HTML
+ }
+ return strings.Join(parts, "")
}
-func recordsToHTMLTableChunkList(records [][]string, chunkRows int, caption string) []string {
+// recordsToHTMLTableChunkList is the structured form of recordsToHTMLTableChunks.
+// headerRowAbs is the 1-based workbook row number of records[0] (normally 1).
+func recordsToHTMLTableChunkList(records [][]string, chunkRows int, caption string, headerRowAbs int) []htmlTableChunk {
+ if headerRowAbs <= 0 {
+ headerRowAbs = 1
+ }
+ colEnd := 1
+ for _, row := range records {
+ if n := len(row); n > colEnd {
+ colEnd = n
+ }
+ }
if len(records) == 0 {
- return []string{"<table><caption>" + html.EscapeString(caption) + "</caption></table>"}
+ return []htmlTableChunk{{
+ HTML: "<table><caption>" + html.EscapeString(caption) + "</caption></table>",
+ RowStart: headerRowAbs,
+ RowEnd: headerRowAbs,
+ ColStart: 1,
+ ColEnd: colEnd,
+ }}
}
- // Build the header row once — repeated in every chunk.
headerHTML := buildHeaderRow(records[0])
dataRows := records[1:]
nData := len(dataRows)
if nData == 0 {
- // Only a header row exists.
- return []string{"<table><caption>" + html.EscapeString(caption) + "</caption>\n" + headerHTML + "</table>"}
+ return []htmlTableChunk{{
+ HTML: "<table><caption>" + html.EscapeString(caption) + "</caption>\n" + headerHTML + "</table>",
+ RowStart: headerRowAbs,
+ RowEnd: headerRowAbs,
+ ColStart: 1,
+ ColEnd: colEnd,
+ }}
}
if chunkRows <= 0 {
@@ -184,7 +220,7 @@ func recordsToHTMLTableChunkList(records [][]string, chunkRows int, caption stri
}
nChunks := (nData + chunkRows - 1) / chunkRows
- chunks := make([]string, 0, nChunks)
+ out := make([]htmlTableChunk, 0, nChunks)
for ci := 0; ci < nChunks; ci++ {
start := ci * chunkRows
end := start + chunkRows
@@ -208,9 +244,16 @@ func recordsToHTMLTableChunkList(records [][]string, chunkRows int, caption stri
b.WriteString("</tr>\n")
}
b.WriteString("</table>\n")
- chunks = append(chunks, b.String())
+
+ out = append(out, htmlTableChunk{
+ HTML: b.String(),
+ RowStart: headerRowAbs + start + 1,
+ RowEnd: headerRowAbs + end,
+ ColStart: 1,
+ ColEnd: colEnd,
+ })
}
- return chunks
+ return out
}
// ──────────────────────────────────────────────────────────── axis helpers
@@ -597,10 +640,15 @@ func decodeChunkRows(setup map[string]any) int {
// data into chunkRows-sized atomic tables each repeating the header. An empty
// or unreadable sheet yields an empty string.
func renderSheetTables(f *excelize.File, sheet string, chunkRows int) string {
- return strings.Join(renderSheetTableChunks(f, sheet, chunkRows), "")
+ chunks := renderSheetTableChunks(f, sheet, chunkRows)
+ parts := make([]string, len(chunks))
+ for i, ch := range chunks {
+ parts[i] = ch.HTML
+ }
+ return strings.Join(parts, "")
}
-func renderSheetTableChunks(f *excelize.File, sheet string, chunkRows int) []string {
+func renderSheetTableChunks(f *excelize.File, sheet string, chunkRows int) []htmlTableChunk {
rows, err := f.GetRows(sheet)
if err != nil || len(rows) == 0 {
return nil
@@ -627,11 +675,32 @@ func renderSheetTableChunks(f *excelize.File, sheet string, chunkRows int) []str
// data. For the common case (header on row 1) this is a no-op.
records := make([][]string, 0, len(rows))
records = append(records, rows[headerRow-1])
+ absDataRows := make([]int, 0, len(rows)-1)
for i, r := range rows {
if i == headerRow-1 {
continue
}
records = append(records, r)
+ absDataRows = append(absDataRows, i+1)
}
- return recordsToHTMLTableChunkList(records, chunkRows, sheet)
+
+ chunks := recordsToHTMLTableChunkList(records, chunkRows, sheet, headerRow)
+ if len(chunks) == 0 || len(absDataRows) == 0 {
+ return chunks
+ }
+ if chunkRows <= 0 {
+ chunkRows = defaultTableChunkRows
+ }
+ for ci := range chunks {
+ start := ci * chunkRows
+ end := start + chunkRows
+ if end > len(absDataRows) {
+ end = len(absDataRows)
+ }
+ if start < end {
+ chunks[ci].RowStart = absDataRows[start]
+ chunks[ci].RowEnd = absDataRows[end-1]
+ }
+ }
+ return chunks
}
diff --git a/internal/parser/parser/office_table_render_test.go b/internal/parser/parser/office_table_render_test.go
index a9b0c8c3b8..956afb5e62 100644
--- a/internal/parser/parser/office_table_render_test.go
+++ b/internal/parser/parser/office_table_render_test.go
@@ -147,6 +147,38 @@ func TestRecordsToHTMLTableChunks_Chunking(t *testing.T) {
}
}
+func TestRecordsToHTMLTableChunkList_RowRange(t *testing.T) {
+ records := make([][]string, 0, 25)
+ records = append(records, []string{"H"})
+ for i := 0; i < 24; i++ {
+ records = append(records, []string{"x"})
+ }
+ chunks := recordsToHTMLTableChunkList(records, 12, "S", 1)
+ if len(chunks) != 2 {
+ t.Fatalf("want 2 chunks, got %d", len(chunks))
+ }
+ if chunks[0].RowStart != 2 || chunks[0].RowEnd != 13 {
+ t.Fatalf("chunk0 rows = %d-%d, want 2-13", chunks[0].RowStart, chunks[0].RowEnd)
+ }
+ if chunks[1].RowStart != 14 || chunks[1].RowEnd != 25 {
+ t.Fatalf("chunk1 rows = %d-%d, want 14-25", chunks[1].RowStart, chunks[1].RowEnd)
+ }
+}
+
+func TestRecordsToHTMLTableChunkList_ColEndUsesWidestRow(t *testing.T) {
+ records := [][]string{
+ {"H"},
+ {"a", "b", "c"},
+ }
+ chunks := recordsToHTMLTableChunkList(records, 12, "S", 1)
+ if len(chunks) != 1 {
+ t.Fatalf("want 1 chunk, got %d", len(chunks))
+ }
+ if chunks[0].ColEnd != 3 {
+ t.Fatalf("ColEnd = %d, want 3", chunks[0].ColEnd)
+ }
+}
+
// TestXLSXParser_HeaderAndCaption asserts the XLSX parser emits a <caption> and
// renders the first row as <th>, and that the header text appears only in <th>.
func TestXLSXParser_HeaderAndCaption(t *testing.T) {
diff --git a/internal/parser/parser/xlsx_parser.go b/internal/parser/parser/xlsx_parser.go
index 08c29351a4..fa16b6c398 100644
--- a/internal/parser/parser/xlsx_parser.go
+++ b/internal/parser/parser/xlsx_parser.go
@@ -127,13 +127,20 @@ func (p *XLSXParser) ParseWithResult(ctx context.Context, filename string, data
items := make([]map[string]any, 0)
warnings := make([]string, 0)
- for _, sheet := range sheets {
+ for sheetIdx, sheet := range sheets {
for _, table := range renderSheetTableChunks(f, sheet, chunkRows) {
items = append(items, map[string]any{
- "text": table,
+ "text": table.HTML,
"doc_type_kwd": "table",
"ck_type": "table",
"sheet": sheet,
+ "positions": [][]float64{{
+ float64(sheetIdx + 1),
+ float64(table.RowStart),
+ float64(table.RowEnd),
+ float64(table.ColStart),
+ float64(table.ColEnd),
+ }},
})
}
images, imageWarnings := extractXLSXImages(f, sheet)
diff --git a/rag/app/naive.py b/rag/app/naive.py
index e59f2ba168..bce88e08e5 100644
--- a/rag/app/naive.py
+++ b/rag/app/naive.py
@@ -58,6 +58,7 @@ from rag.nlp import (
naive_merge_docx,
rag_tokenizer,
tokenize_chunks,
+ tokenize_chunks_with_positions,
doc_tokenize_chunks_with_images,
tokenize_table,
append_context2table_image4pdf,
@@ -114,6 +115,53 @@ def _normalize_section_text_for_rtl_presentation_forms(sections):
return normalized_sections
+def _merge_excel_items(items, chunk_token_num=128):
+ """Merge consecutive Excel rows within the same sheet by token budget.
+
+ Each item is (text, (sheet_idx, row_start, row_end, col_start, col_end)).
+ When chunk_token_num <= 0, items are returned unchanged (html4excel).
+ """
+ if not items:
+ return []
+ if chunk_token_num <= 0:
+ return items
+
+ merged = []
+ cur_text = ""
+ cur_pos = None
+ cur_tokens = 0
+
+ for text, pos in items:
+ sheet_idx, r1, r2, c1, c2 = pos
+ tok = num_tokens_from_string(text)
+ same_sheet = cur_pos is not None and cur_pos[0] == sheet_idx
+ if cur_text and (not same_sheet or cur_tokens + tok > chunk_token_num):
+ merged.append((cur_text, cur_pos))
+ cur_text = ""
+ cur_pos = None
+ cur_tokens = 0
+
+ if not cur_text:
+ cur_text = text
+ cur_pos = (sheet_idx, r1, r2, c1, c2)
+ cur_tokens = tok
+ continue
+
+ cur_text = cur_text + "\n" + text
+ cur_pos = (
+ sheet_idx,
+ min(cur_pos[1], r1),
+ max(cur_pos[2], r2),
+ min(cur_pos[3], c1),
+ max(cur_pos[4], c2),
+ )
+ cur_tokens += tok
+
+ if cur_text:
+ merged.append((cur_text, cur_pos))
+ return merged
+
+
def by_deepdoc(filename, binary=None, from_page=0, to_page=MAXIMUM_PAGE_NUMBER, lang="Chinese", callback=None, pdf_cls=None, **kwargs):
pdf_parser = pdf_cls() if pdf_cls else Pdf()
sections, tables = pdf_parser(filename if binary is None else binary, from_page=from_page, to_page=to_page, callback=callback)
@@ -1187,16 +1235,28 @@ def chunk(filename, binary=None, from_page=0, to_page=MAXIMUM_PAGE_NUMBER, lang=
sections = _normalize_section_text_for_rtl_presentation_forms(sections)
parser_config["chunk_token_num"] = 0
res = tokenize_table(tables, doc, is_english, language=lang)
+ sections = []
callback(0.8, "Finish parsing.")
else:
# Default DeepDOC parser
excel_parser = ExcelParser()
if parser_config.get("html4excel"):
- sections = [(_, "") for _ in excel_parser.html(binary, 12) if _]
+ excel_items = [item for item in excel_parser.html(binary, 12) if item and item[0]]
parser_config["chunk_token_num"] = 0
else:
- sections = [(_, "") for _ in excel_parser(binary) if _]
- sections = _normalize_section_text_for_rtl_presentation_forms(sections)
+ excel_items = [item for item in excel_parser(binary) if item and item[0]]
+ excel_items = [(normalize_arabic_presentation_forms(text), pos) for text, pos in excel_items]
+ res.extend(
+ tokenize_chunks_with_positions(
+ _merge_excel_items(excel_items, int(parser_config.get("chunk_token_num", 128))),
+ doc,
+ is_english,
+ child_delimiters_pattern=child_deli,
+ language=lang,
+ )
+ )
+ sections = []
+ callback(0.8, "Finish parsing.")
elif re.search(r"\.(txt|py|js|java|c|cpp|h|php|go|ts|sh|cs|kt|sql)$", filename, re.IGNORECASE):
callback(0.1, "Start to parse.")
diff --git a/rag/app/one.py b/rag/app/one.py
index 1a08d2c9cb..7cde2485e8 100644
--- a/rag/app/one.py
+++ b/rag/app/one.py
@@ -145,6 +145,9 @@ def chunk(filename, binary=None, from_page=0, to_page=MAXIMUM_PAGE_NUMBER, lang=
callback(0.1, "Start to parse.")
excel_parser = ExcelParser()
sections = excel_parser.html(binary, MAXIMUM_TASK_PAGE_NUMBER)
+ # One parser tokenizes the whole file as a single chunk, so per-cell
+ # coordinates cannot be cited. Keep the HTML text only.
+ sections = [s[0] if isinstance(s, tuple) else s for s in sections]
elif re.search(r"\.(txt|md|markdown|mdx)$", filename, re.IGNORECASE):
callback(0.1, "Start to parse.")
diff --git a/rag/flow/parser/parser.py b/rag/flow/parser/parser.py
index eda668727d..d8742083ed 100644
--- a/rag/flow/parser/parser.py
+++ b/rag/flow/parser/parser.py
@@ -880,9 +880,22 @@ class Parser(ProcessBase):
spreadsheet_parser = ExcelParser()
if conf.get("output_format") == "html":
htmls = spreadsheet_parser.html(blob, 1000000000)
- self.set_output("html", htmls[0])
+ self.set_output("html", htmls[0][0] if htmls else "")
elif conf.get("output_format") == "json":
- self.set_output("json", [{"text": txt, "doc_type_kwd": "text"} for txt in spreadsheet_parser(blob) if txt])
+ self.set_output(
+ "json",
+ [
+ {
+ "text": txt,
+ "doc_type_kwd": "text",
+ # 0-based sheet. TaskExecutor and dataflow_service
+ # call add_positions, which stores pn+1 (1-based).
+ "positions": [[sheet, r1, r2, c1, c2]],
+ }
+ for txt, (sheet, r1, r2, c1, c2) in spreadsheet_parser(blob)
+ if txt
+ ],
+ )
elif conf.get("output_format") == "markdown":
self.set_output("markdown", spreadsheet_parser.markdown(blob))
diff --git a/rag/nlp/__init__.py b/rag/nlp/__init__.py
index fe625b01fa..e6e48d7ce2 100644
--- a/rag/nlp/__init__.py
+++ b/rag/nlp/__init__.py
@@ -451,6 +451,27 @@ def tokenize_chunks(chunks, doc, eng, pdf_parser=None, child_delimiters_pattern=
return res
+def tokenize_chunks_with_positions(chunks_with_pos, doc, eng, child_delimiters_pattern=None, language="English"):
+ """Tokenize chunks that already carry real positions (Excel sheet/row).
+
+ chunks_with_pos: iterable of (text, position) where position is a 5-tuple
+ suitable for add_positions (first component 0-based).
+ """
+ res = []
+ for ck, pos in chunks_with_pos:
+ if not ck or not str(ck).strip():
+ continue
+ d = copy.deepcopy(doc)
+ add_positions(d, [pos])
+ if child_delimiters_pattern:
+ d["mom_with_weight"] = ck.removeprefix("\n")
+ res.extend(split_with_pattern(d, child_delimiters_pattern, ck, eng, language=language))
+ continue
+ tokenize(d, ck, eng, language=language)
+ res.append(d)
+ return res
+
+
def doc_tokenize_chunks_with_images(chunks, doc, eng, child_delimiters_pattern=None, batch_size=10, language="English"):
res = []
for ii, ck in enumerate(chunks):
diff --git a/test/unit_test/deepdoc/parser/test_excel_parser.py b/test/unit_test/deepdoc/parser/test_excel_parser.py
index 9e3a2aa281..b137779ca4 100644
--- a/test/unit_test/deepdoc/parser/test_excel_parser.py
+++ b/test/unit_test/deepdoc/parser/test_excel_parser.py
@@ -73,7 +73,8 @@ def test_exact_multiple_does_not_emit_header_only_chunk():
# 12 data rows with chunk_rows=12 (the value rag/app/naive.py uses).
chunks = RAGFlowExcelParser().html(_make_xlsx(12), chunk_rows=12)
assert len(chunks) == 1
- assert all(not _chunk_has_no_data_cells(c) for c in chunks)
+ assert all(not _chunk_has_no_data_cells(c[0]) for c in chunks)
+ assert chunks[0][1][0] == 0 and chunks[0][1][1] == 2 and chunks[0][1][2] == 13
@pytest.mark.p2
@@ -81,7 +82,24 @@ def test_multiple_of_chunk_rows_splits_without_spurious_chunk():
# 24 data rows with chunk_rows=12 -> exactly 2 data chunks, no trailing header-only chunk.
chunks = RAGFlowExcelParser().html(_make_xlsx(24), chunk_rows=12)
assert len(chunks) == 2
- assert all(not _chunk_has_no_data_cells(c) for c in chunks)
+ assert all(not _chunk_has_no_data_cells(c[0]) for c in chunks)
+ assert chunks[0][1][1] == 2 and chunks[0][1][2] == 13
+ assert chunks[1][1][1] == 14 and chunks[1][1][2] == 25
+
+
+def test_html_col_max_uses_widest_row():
+ from openpyxl import Workbook
+
+ wb = Workbook()
+ ws = wb.active
+ ws.append(["H1"])
+ ws.append(["a", "b", "c"])
+ buf = BytesIO()
+ wb.save(buf)
+ buf.seek(0)
+ chunks = RAGFlowExcelParser().html(buf.read(), chunk_rows=12)
+ assert chunks[0][1][3] == 1 and chunks[0][1][4] == 3
@pytest.mark.p2
@@ -89,7 +107,7 @@ def test_non_multiple_unchanged():
# 13 data rows with chunk_rows=12 -> 2 chunks (12 + 1).
chunks = RAGFlowExcelParser().html(_make_xlsx(13), chunk_rows=12)
assert len(chunks) == 2
- assert all(not _chunk_has_no_data_cells(c) for c in chunks)
+ assert all(not _chunk_has_no_data_cells(c[0]) for c in chunks)
def _make_xlsx_with_values(header, row):
@@ -111,13 +129,42 @@ def test_call_keeps_zero_valued_cells():
# is real data, not an empty cell, so it must survive. The header is only
# emitted alongside a kept value, so a dropped 0 also loses its "stock" label.
lines = RAGFlowExcelParser()(_make_xlsx_with_values(["name", "stock"], ["widget", 0]))
- joined = " ".join(lines)
+ joined = " ".join(text for text, _ in lines)
assert "stock" in joined and "0" in joined, lines
+ assert lines[0][1][1] == 2
@pytest.mark.p2
def test_call_skips_truly_empty_cells():
# None / empty-string cells carry no value and should still be skipped.
lines = RAGFlowExcelParser()(_make_xlsx_with_values(["name", "note"], ["widget", None]))
- joined = " ".join(lines)
+ joined = " ".join(text for text, _ in lines)
assert "note" not in joined, lines
+
+
+def _make_two_sheet_xlsx():
+ from openpyxl import Workbook
+
+ wb = Workbook()
+ ws1 = wb.active
+ ws1.title = "s1"
+ ws1.append(["H"])
+ ws1.append(["a"])
+ ws2 = wb.create_sheet("s2")
+ ws2.append(["H"])
+ ws2.append(["b"])
+ buf = BytesIO()
+ wb.save(buf)
+ buf.seek(0)
+ return buf.read()
+
+
+def test_call_emits_zero_based_sheet_index():
+ # Flow JSON stores this index unchanged; add_positions later does pn+1 so
+ # sheet 2 is stored as 2 and the preview selects xs.datas[1].
+ lines = RAGFlowExcelParser()(_make_two_sheet_xlsx())
+ sheets = {pos[0] for _, pos in lines}
+ assert sheets == {0, 1}
+ second = next(pos for _, pos in lines if pos[0] == 1)
+ assert second == (1, 2, 2, 1, 1)
diff --git a/test/unit_test/rag/nlp/test_tokenize_chunks.py b/test/unit_test/rag/nlp/test_tokenize_chunks.py
index 594305d708..09b0fd4ab4 100644
--- a/test/unit_test/rag/nlp/test_tokenize_chunks.py
+++ b/test/unit_test/rag/nlp/test_tokenize_chunks.py
@@ -57,3 +57,44 @@ def test_tokenize_chunks_with_images_removes_leading_newline_from_parent():
assert [doc["mom_with_weight"] for doc in docs] == ["first\nsecond", "first\nsecond"]
assert [doc["content_with_weight"] for doc in docs] == ["first\n", "second"]
+
+
+def test_tokenize_chunks_with_positions_increments_zero_based_sheet():
+ # Flow JSON / naive Excel emit 0-based sheet; add_positions stores pn+1.
+ docs = nlp.tokenize_chunks_with_positions(
+ [("row", (1, 2, 4, 1, 3))],
+ {"docnm_kwd": "a.xlsx"},
+ True,
+ )
+ assert docs[0]["position_int"] == [(2, 2, 4, 1, 3)]
+ assert docs[0]["page_num_int"] == [2]
+ assert docs[0]["content_with_weight"] == "row"
+
+
+def test_tokenize_chunks_with_positions_skips_blank_and_splits_children():
+ docs = nlp.tokenize_chunks_with_positions(
+ [(" ", (0, 2, 2, 1, 1)), ("first\nsecond", (0, 3, 4, 1, 2))],
+ {},
+ True,
+ child_delimiters_pattern="\n",
+ )
+ assert [doc["content_with_weight"] for doc in docs] == ["first\n", "second"]
+ assert all(doc["position_int"][0] == (1, 3, 4, 1, 2) for doc in docs)
+
+
+def test_flow_json_sheet_is_one_based_after_add_positions():
+ # rag/flow/parser/parser.py output_format=json stores 0-based sheet.
+ # task_executor / dataflow_service then call add_positions (pn+1).
+ json_out = [
+ {"text": "s1", "positions": [[0, 2, 2, 1, 1]]},
+ {"text": "s2", "positions": [[1, 2, 2, 1, 1]]},
+ ]
+ sheets = []
+ for ck in json_out:
+ nlp.add_positions(ck, ck["positions"])
+ del ck["positions"]
+ sheets.append(ck["position_int"][0][0])
+ assert sheets == [1, 2]
diff --git a/test/unit_test/rag/test_naive_dispatch.py b/test/unit_test/rag/test_naive_dispatch.py
index dd4649d367..f7e0cac2f7 100644
--- a/test/unit_test/rag/test_naive_dispatch.py
+++ b/test/unit_test/rag/test_naive_dispatch.py
@@ -221,6 +221,7 @@ def naive_module():
naive_merge_with_images=lambda *a, **k: [],
naive_merge_docx=lambda *a, **k: [],
tokenize_chunks=lambda *a, **k: [],
+ tokenize_chunks_with_positions=lambda *a, **k: [],
doc_tokenize_chunks_with_images=lambda *a, **k: [],
tokenize_chunks_with_images=lambda *a, **k: [],
tokenize_table=lambda *a, **k: [],
@@ -374,3 +375,29 @@ def test_dispatch_uses_resolved_layout_recognize_via_override_for_opendataloader
assert name == "opendataloader"
assert parser is _ByOpenDataLoader
assert op_name == resolved
+
+
+def test_merge_excel_items_keeps_sheets_separate(naive_module):
+ items = [
+ ("a b", (0, 2, 2, 1, 2)),
+ ("c d", (0, 3, 3, 1, 2)),
+ ("e f", (1, 2, 2, 1, 3)),
+ ]
+ merged = naive_module._merge_excel_items(items, chunk_token_num=10)
+ assert [pos for _, pos in merged] == [(0, 2, 3, 1, 2), (1, 2, 2, 1, 3)]
+ assert merged[0][0] == "a b\nc d"
+
+
+def test_merge_excel_items_splits_on_token_budget(naive_module):
+ items = [
+ ("one two", (0, 2, 2, 1, 1)),
+ ("three four", (0, 3, 3, 1, 1)),
+ ]
+ merged = naive_module._merge_excel_items(items, chunk_token_num=2)
+ assert [pos for _, pos in merged] == [(0, 2, 2, 1, 1), (0, 3, 3, 1, 1)]
+
+
+def test_merge_excel_items_passthrough_when_budget_disabled(naive_module):
+ items = [("a", (0, 2, 2, 1, 1)), ("b", (0, 3, 3, 1, 1))]
+ assert naive_module._merge_excel_items(items, chunk_token_num=0) == items
+ assert naive_module._merge_excel_items([], chunk_token_num=128) == []
diff --git a/web/src/components/document-preview/excel-preview.tsx b/web/src/components/document-preview/excel-preview.tsx
index f336cf3403..038022fa9e 100644
--- a/web/src/components/document-preview/excel-preview.tsx
+++ b/web/src/components/document-preview/excel-preview.tsx
@@ -14,21 +14,22 @@
* limitations under the License.
*/
-// import { useFetchExcel } from '@/pages/document-viewer/hooks';
import classNames from 'classnames';
import { useFetchExcel } from './hooks';
interface ExcelCsvPreviewerProps {
className?: string;
url: string;
+ /** [sheetIndex(1-based), rowStart, rowEnd, colStart, colEnd] */
+ positions?: number[][];
}
export const ExcelCsvPreviewer: React.FC<ExcelCsvPreviewerProps> = ({
className,
url,
+ positions,
}) => {
- // const url = useGetDocumentUrl();
- const { containerRef } = useFetchExcel(url);
+ const { containerRef } = useFetchExcel(url, positions);
return (
<div
diff --git a/web/src/components/document-preview/hooks.test.ts b/web/src/components/document-preview/hooks.test.ts
new file mode 100644
index 0000000000..a1c029d052
--- /dev/null
+++ b/web/src/components/document-preview/hooks.test.ts
@@ -0,0 +1,132 @@
+import { applyExcelSourceLocate } from './hooks';
+
+jest.mock('@js-preview/excel', () => ({
+ __esModule: true,
+ default: { init: jest.fn() },
+}));
+jest.mock('ahooks', () => ({
+ useDebounceFn: () => ({ run: jest.fn() }),
+ useSize: () => ({ width: 800, height: 600 }),
+}));
+jest.mock('axios', () => ({ get: jest.fn() }));
+jest.mock('jszip');
+jest.mock('xlsx', () => ({ read: jest.fn(), write: jest.fn() }));
+jest.mock('@/hooks/route-hook', () => ({
+ useGetKnowledgeSearchParams: () => ({}),
+}));
+jest.mock('@/pages/dataflow-result/hooks', () => ({
+ useGetPipelineResultSearchParams: () => ({}),
+}));
+jest.mock('@/utils/api', () => ({
+ default: {},
+ restAPIv1: '',
+}));
+jest.mock('@/utils/authorization-util', () => ({
+ getAuthorization: () => '',
+}));
+jest.mock('@/constants/authorization', () => ({
+ Authorization: 'Authorization',
+}));
+
+function mockPreviewer(sheetCount = 2) {
+ const makeData = () => ({
+ addStyle: jest.fn(() => 7),
+ rows: {
+ setCell: jest.fn(),
+ getCellOrNew: jest.fn(),
+ getHeight: jest.fn(() => 20),
+ },
+ scroll: { x: 0, y: 0 },
+ });
+ const datas = Array.from({ length: sheetCount }, makeData);
+ const items = datas.map(() => document.createElement('div'));
+ return {
+ xs: {
+ datas,
+ reRender: jest.fn(),
+ sheet: {
+ resetData: jest.fn(),
+ data: datas[0],
+ reload: jest.fn(),
+ table: { render: jest.fn() },
+ },
+ bottombar: {
+ items,
+ clickSwap2: jest.fn(),
+ },
+ },
+ };
+}
+
+describe('applyExcelSourceLocate', () => {
+ beforeEach(() => {
+ jest.spyOn(window, 'getComputedStyle').mockReturnValue({
+ getPropertyValue: (name: string) => {
+ if (name === '--background-highlight') {
+ return 'rgba(76, 164, 231, 0.1)';
+ }
+ if (name === '--text-title') {
+ return 'rgba(22, 22, 24, 1)';
+ }
+ return '';
+ },
+ } as CSSStyleDeclaration);
+ });
+
+ afterEach(() => {
+ jest.restoreAllMocks();
+ });
+
+ it('selects the 1-based sheet and paints the cell range with theme tokens', () => {
+ const previewer = mockPreviewer();
+ applyExcelSourceLocate(previewer as never, [2, 3, 4, 1, 2]);
+
+ expect(previewer.xs.bottombar.clickSwap2).toHaveBeenCalledWith(
+ previewer.xs.bottombar.items[1],
+ );
+ expect(previewer.xs.datas[1].addStyle).toHaveBeenCalledWith({
+ bgcolor: 'rgba(76, 164, 231, 0.1)',
+ color: 'rgba(22, 22, 24, 1)',
+ });
+ expect(previewer.xs.datas[1].rows.setCell).toHaveBeenCalledTimes(4);
+ expect(previewer.xs.datas[1].rows.setCell).toHaveBeenCalledWith(
+ 2,
+ 0,
+ { style: 7 },
+ 'format',
+ );
+ expect(previewer.xs.datas[1].rows.setCell).toHaveBeenCalledWith(
+ 3,
+ 1,
+ { style: 7 },
+ 'format',
+ );
+ });
+
+ it('is a no-op when the position tuple is short', () => {
+ const previewer = mockPreviewer();
+ applyExcelSourceLocate(previewer as never, [1, 2]);
+ expect(previewer.xs.datas[0].addStyle).not.toHaveBeenCalled();
+ });
+
+ it('falls back to resetData when clickSwap2 is missing', () => {
+ const previewer = mockPreviewer();
+ delete previewer.xs.bottombar.clickSwap2;
+ applyExcelSourceLocate(previewer as never, [1, 2, 2, 1, 1]);
+ expect(previewer.xs.sheet.resetData).toHaveBeenCalledWith(
+ previewer.xs.datas[0],
+ );
+ });
+
+ it('uses hex fallbacks when theme tokens are empty', () => {
+ jest.spyOn(window, 'getComputedStyle').mockReturnValue({
+ getPropertyValue: () => ' ',
+ } as CSSStyleDeclaration);
+ const previewer = mockPreviewer();
+ applyExcelSourceLocate(previewer as never, [1, 2, 2, 1, 1]);
+ expect(previewer.xs.datas[0].addStyle).toHaveBeenCalledWith({
+ bgcolor: '#ffe58f',
+ color: '#1a1a1a',
+ });
+ });
+});
diff --git a/web/src/components/document-preview/hooks.ts b/web/src/components/document-preview/hooks.ts
index 3caa7b2d16..5a981f313c 100644
--- a/web/src/components/document-preview/hooks.ts
+++ b/web/src/components/document-preview/hooks.ts
@@ -209,7 +209,117 @@ async function normalizeXlsxForExcelJS(data: ArrayBuffer): Promise<ArrayBuffer>
}
}
-export const useFetchExcel = (filePath: string) => {
+type ExcelLocatePos = number[];
+
+/** x-spreadsheet Data instance used by @js-preview/excel (not a plain object). */
+type XsData = {
+ addStyle: (style: Record<string, unknown>) => number;
+ rows: {
+ setCell: (
+ ri: number,
+ ci: number,
+ cell: { style?: number; text?: string },
+ type?: 'all' | 'text' | 'format',
+ ) => void;
+ getCellOrNew: (
+ ri: number,
+ ci: number,
+ ) => { style?: number; text?: string };
+ getHeight?: (ri: number) => number;
+ len?: number;
+ };
+ cols?: { len?: number };
+ scroll?: { x: number; y: number };
+};
+
+type XsInstance = {
+ datas: XsData[];
+ reRender?: () => void;
+ sheet: {
+ resetData: (data: XsData) => void;
+ data?: XsData & { row?: { height?: number } };
+ reload?: () => void;
+ table?: { render?: () => void };
+ };
+ bottombar?: {
+ items: HTMLElement[];
+ clickSwap2?: (el: HTMLElement) => void;
+ };
+};
+
+/** Locate inside @js-preview/excel: switch sheet, paint range via Data API, scroll. */
+export function applyExcelSourceLocate(
+ previewer: ReturnType<typeof jsPreviewExcel.init>,
+ pos: ExcelLocatePos,
+) {
+ if (!pos || pos.length < 5) return;
+ const xs = (previewer as { xs?: XsInstance }).xs;
+ if (!xs?.datas?.length) return;
+
+ const sheetIdx = Math.min(
+ xs.datas.length - 1,
+ Math.max(0, (pos[0] || 1) - 1),
+ );
+ let r1 = Math.max(0, (pos[1] || 1) - 1);
+ let r2 = Math.max(r1, (pos[2] || pos[1] || 1) - 1);
+ let c1 = (pos[3] || 1) - 1;
+ let c2 = (pos[4] || pos[3] || 1) - 1;
+ if (c1 < 0) c1 = 0;
+ if (c2 < c1) c2 = c1;
+
+ const data = xs.datas[sheetIdx];
+ if (!data?.addStyle || !data.rows?.setCell) return;
+
+ if (typeof data.rows.len === 'number') {
+ const lastRow = Math.max(0, data.rows.len - 1);
+ r1 = Math.min(r1, lastRow);
+ r2 = Math.min(r2, lastRow);
+ }
+ if (typeof data.cols?.len === 'number') {
+ const lastCol = Math.max(0, data.cols.len - 1);
+ c1 = Math.min(c1, lastCol);
+ c2 = Math.min(c2, lastCol);
+ }
+
+ const theme = getComputedStyle(document.documentElement);
+ if (r2 >= r1 && c2 >= c1) {
+ const styleIdx = data.addStyle({
+ bgcolor:
+ theme.getPropertyValue('--background-highlight').trim() || '#ffe58f',
+ color: theme.getPropertyValue('--text-title').trim() || '#1a1a1a',
+ });
+ for (let r = r1; r <= r2; r++) {
+ for (let c = c1; c <= c2; c++) {
+ data.rows.setCell(r, c, { style: styleIdx }, 'format');
+ }
+ }
+ }
+
+ const tab = xs.bottombar?.items?.[sheetIdx];
+ if (tab && typeof xs.bottombar?.clickSwap2 === 'function') {
+ xs.bottombar.clickSwap2(tab);
+ } else {
+ xs.sheet.resetData(data);
+ }
+
+ const rowHeight = data.rows.getHeight?.(0) || xs.sheet.data?.row?.height || 24;
+ let y = 0;
+ for (let r = 0; r < r1; r++) {
+ y += data.rows.getHeight?.(r) ?? rowHeight;
+ }
+ data.scroll = { ...(data.scroll || { x: 0, y: 0 }), x: 0, y };
+ if (xs.sheet.data && xs.sheet.data !== data) {
+ xs.sheet.data.scroll = { ...(xs.sheet.data.scroll || { x: 0, y: 0 }), x: 0, y };
+ }
+ xs.reRender?.();
+ xs.sheet.table?.render?.();
+ xs.sheet.reload?.();
+}
+
+export const useFetchExcel = (
+ filePath: string,
+ positions?: number[][],
+) => {
const [status, setStatus] = useState(true);
const { fetchDocument } = useFetchDocument();
const [containerEl, setContainerEl] = useState<HTMLDivElement | null>(null);
@@ -222,6 +332,10 @@ export const useFetchExcel = (filePath: string) => {
// Cache the fetched ArrayBuffer so we don't re-fetch on every resize.
const dataRef = useRef<ArrayBuffer | null>(null);
const [dataReady, setDataReady] = useState(false);
+ const positionsRef = useRef(positions);
+ positionsRef.current = positions;
+ const locateKey = positions?.[0]?.join(',') ?? '';
+ const locateTimerRef = useRef<ReturnType<typeof setTimeout> | null>(null);
// @js-preview/excel reads the container's width/height at init time and
// exposes no public resize method, so the only way to make the spreadsheet
@@ -229,6 +343,10 @@ export const useFetchExcel = (filePath: string) => {
const renderPreview = useCallback(async () => {
if (!containerEl || !dataRef.current) return;
+ if (locateTimerRef.current != null) {
+ clearTimeout(locateTimerRef.current);
+ locateTimerRef.current = null;
+ }
if (previewerRef.current) {
previewerRef.current.destroy();
previewerRef.current = null;
@@ -239,6 +357,18 @@ export const useFetchExcel = (filePath: string) => {
try {
await previewer.preview(dataRef.current);
+ const pos = positionsRef.current?.[0];
+ if (pos) {
+ // bottombar tabs are created during loadData; apply after paint.
+ const run = () => {
+ if (previewerRef.current !== previewer) return;
+ applyExcelSourceLocate(previewer, pos);
+ };
+ requestAnimationFrame(() => {
+ run();
+ locateTimerRef.current = setTimeout(run, 50);
+ });
+ }
setStatus(true);
} catch (e) {
// oxlint-disable-next-line no-console
@@ -305,15 +435,26 @@ export const useFetchExcel = (filePath: string) => {
};
}, [filePath, fetchDocument]);
- // Initial render + debounced re-render on container resize.
+ // Initial render + debounced re-render on container resize / locate target change.
useEffect(() => {
if (!dataReady || !containerEl) return;
debouncedRender();
- }, [dataReady, containerEl, size?.width, size?.height, debouncedRender]);
+ }, [
+ dataReady,
+ containerEl,
+ size?.width,
+ size?.height,
+ locateKey,
+ debouncedRender,
+ ]);
// Tear down the previewer on unmount.
useEffect(() => {
return () => {
+ if (locateTimerRef.current != null) {
+ clearTimeout(locateTimerRef.current);
+ locateTimerRef.current = null;
+ }
if (previewerRef.current) {
previewerRef.current.destroy();
previewerRef.current = null;
diff --git a/web/src/components/document-preview/index.tsx b/web/src/components/document-preview/index.tsx
index de3361fc62..2d9920bcf9 100644
--- a/web/src/components/document-preview/index.tsx
+++ b/web/src/components/document-preview/index.tsx
@@ -31,6 +31,7 @@ type PreviewProps = {
fileType: string;
className?: string;
url: string;
+ positions?: number[][];
};
const DocumentPreview = function ({
fileType,
@@ -38,6 +39,7 @@ const DocumentPreview = function ({
highlights,
setWidthAndHeight,
url,
+ positions,
}: PreviewProps & Partial<IProps>) {
const isPdf = fileType === 'pdf';
@@ -90,9 +92,13 @@ const DocumentPreview = function ({
<PptPreviewer className={className} url={url} />
</section>
)}
- {['xlsx'].indexOf(fileType) > -1 && (
- <section>
- <ExcelCsvPreviewer className={className} url={url} />
+ {['xlsx', 'xls'].indexOf(fileType) > -1 && (
+ <section className="h-full">
+ <ExcelCsvPreviewer
+ className={className}
+ url={url}
+ positions={positions}
+ />
</section>
)}
{['csv'].indexOf(fileType) > -1 && (
diff --git a/web/src/components/floating-chat-widget-markdown.tsx b/web/src/components/floating-chat-widget-markdown.tsx
index 7e1afc3957..6f20c1d590 100644
--- a/web/src/components/floating-chat-widget-markdown.tsx
+++ b/web/src/components/floating-chat-widget-markdown.tsx
@@ -34,6 +34,7 @@ import {
} from '@/utils/chat';
import { citationMarkerReg } from '@/utils/citation-utils';
import { getExtension } from '@/utils/document-util';
+import { supportsSourceLocate } from '@/utils/source-locate';
import { getDirAttribute } from '@/utils/text-direction';
import { InfoCircleOutlined } from '@ant-design/icons';
import * as HoverCardPrimitive from '@radix-ui/react-hover-card';
@@ -108,12 +109,15 @@ const FloatingChatWidgetMarkdown = ({
) =>
() => {
if (!documentId) return;
- if (fileExtension !== 'pdf' && documentUrl) {
- const nextLink = `/document/${documentId}?ext=${fileExtension}&resource=${'document'}`;
- window.open(nextLink, '_blank');
- } else if (clickDocumentButton) {
+ if (supportsSourceLocate(fileExtension) && clickDocumentButton) {
clickDocumentButton(documentId, chunk);
+ return;
}
+ if (!documentUrl) return;
+ window.open(
+ `/document/${documentId}?ext=${fileExtension}&resource=${'document'}`,
+ '_blank',
+ );
},
[clickDocumentButton],
);
@@ -236,7 +240,9 @@ const FloatingChatWidgetMarkdown = ({
fileExtension,
documentUrl,
)}
- disabled={!documentUrl && fileExtension !== 'pdf'}
+ disabled={
+ !documentUrl && !supportsSourceLocate(fileExtension)
+ }
style={{ whiteSpace: 'normal' }}
>
<span className="truncate">
@@ -245,7 +251,7 @@ const FloatingChatWidgetMarkdown = ({
</Button>
</TooltipTrigger>
<TooltipContent>
- {!documentUrl && fileExtension !== 'pdf'
+ {!documentUrl && !supportsSourceLocate(fileExtension)
? 'Document link unavailable'
: document.doc_name}
</TooltipContent>
diff --git a/web/src/components/markdown-content/index.tsx b/web/src/components/markdown-content/index.tsx
index 408e6abc44..ae5d3bde0e 100644
--- a/web/src/components/markdown-content/index.tsx
+++ b/web/src/components/markdown-content/index.tsx
@@ -20,6 +20,7 @@ import { MarkdownRemarkPlugins } from '@/constants/markdown-remark-plugins';
import { IReference, IReferenceChunk } from '@/interfaces/database/chat';
import { citationMarkerReg } from '@/utils/citation-utils';
import { getExtension } from '@/utils/document-util';
+import { supportsSourceLocate } from '@/utils/source-locate';
import { getDirAttribute } from '@/utils/text-direction';
import DOMPurify from 'dompurify';
import { memo, useCallback, useEffect, useMemo } from 'react';
@@ -185,15 +186,15 @@ const MarkdownContent = ({
documentUrl?: string,
) =>
() => {
- if (fileExtension !== 'pdf') {
- if (!documentUrl) {
- return;
- }
- const nextLink = `/document/${documentId}?ext=${fileExtension}&resource=${'document'}`;
- window.open(nextLink, '_blank');
- } else {
- clickDocumentButton?.(documentId, chunk);
+ if (supportsSourceLocate(fileExtension) && clickDocumentButton) {
+ clickDocumentButton(documentId, chunk);
+ return;
}
+ if (!documentUrl) return;
+ window.open(
+ `/document/${documentId}?ext=${fileExtension}&resource=${'document'}`,
+ '_blank',
+ );
},
[clickDocumentButton],
);
diff --git a/web/src/components/next-markdown-content/index.tsx b/web/src/components/next-markdown-content/index.tsx
index 68a5b5304f..acc9a6283c 100644
--- a/web/src/components/next-markdown-content/index.tsx
+++ b/web/src/components/next-markdown-content/index.tsx
@@ -20,6 +20,7 @@ import { SafeImg } from '@/components/safe-img';
import { MarkdownRemarkPlugins } from '@/constants/markdown-remark-plugins';
import { IReferenceChunk, IReferenceObject } from '@/interfaces/database/chat';
import { getExtension } from '@/utils/document-util';
+import { supportsSourceLocate } from '@/utils/source-locate';
import { downloadFileFromBlob } from '@/utils/file-util';
import request from '@/utils/request';
import DOMPurify from 'dompurify';
@@ -230,15 +231,15 @@ function MarkdownContent({
documentUrl?: string,
) =>
() => {
- if (fileExtension !== 'pdf') {
- if (!documentUrl) {
- return;
- }
- const nextLink = `/document/${documentId}?ext=${fileExtension}&resource=${'document'}`;
- window.open(nextLink, '_blank');
- } else {
- clickDocumentButton?.(documentId, chunk);
+ if (supportsSourceLocate(fileExtension) && clickDocumentButton) {
+ clickDocumentButton(documentId, chunk);
+ return;
}
+ if (!documentUrl) return;
+ window.open(
+ `/document/${documentId}?ext=${fileExtension}&resource=${'document'}`,
+ '_blank',
+ );
},
[clickDocumentButton],
);
diff --git a/web/src/components/pdf-drawer/index.tsx b/web/src/components/pdf-drawer/index.tsx
index a5df43f042..f6b35247d8 100644
--- a/web/src/components/pdf-drawer/index.tsx
+++ b/web/src/components/pdf-drawer/index.tsx
@@ -22,7 +22,9 @@ import { IModalProps } from '@/interfaces/common';
import { IReferenceChunk } from '@/interfaces/database/chat';
import { IChunk } from '@/interfaces/database/dataset';
import { cn } from '@/lib/utils';
-import PdfPreview from '../document-preview/pdf-preview';
+import { getExtension } from '@/utils/document-util';
+import { get } from 'lodash';
+import DocumentPreview from '../document-preview';
import { Sheet, SheetContent, SheetHeader, SheetTitle } from '../ui/sheet';
interface IProps extends IModalProps<any> {
@@ -42,6 +44,11 @@ export const PdfSheet = ({
const getDocumentUrl = useGetDocumentUrl(documentId);
const url = getDocumentUrl(documentId);
const { highlights, setWidthAndHeight } = useGetChunkHighlights(chunk);
+ const fileType = getExtension(
+ get(chunk, 'document_name', '') || get(chunk, 'docnm_kwd', '') || 'pdf',
+ );
+ const positions = Array.isArray(chunk?.positions) ? chunk.positions : [];
+
return (
<Sheet open onOpenChange={hideModal}>
<SheetContent
@@ -55,12 +62,14 @@ export const PdfSheet = ({
<SheetTitle>Document Previewer</SheetTitle>
</SheetHeader>
{url && documentId && (
- <PdfPreview
+ <DocumentPreview
className={'p-0 !h-[calc(100vh-80px)] w-full'}
+ fileType={fileType || 'pdf'}
highlights={highlights}
setWidthAndHeight={setWidthAndHeight}
url={url}
- ></PdfPreview>
+ positions={positions}
+ />
)}
</SheetContent>
</Sheet>
diff --git a/web/src/constants/common.ts b/web/src/constants/common.ts
index 0bce29efef..9e3674b63a 100644
--- a/web/src/constants/common.ts
+++ b/web/src/constants/common.ts
@@ -218,6 +218,8 @@ export const ExceptiveType = [
'txt',
'csv',
'pptx',
+ 'html',
+ 'htm',
...Images,
];
diff --git a/web/src/pages/chunk/parsed-result/add-knowledge/components/knowledge-chunk/components/document-view-switch/index.tsx b/web/src/pages/chunk/parsed-result/add-knowledge/components/knowledge-chunk/components/document-view-switch/index.tsx
index 44118464b0..c19d66374e 100644
--- a/web/src/pages/chunk/parsed-result/add-knowledge/components/knowledge-chunk/components/document-view-switch/index.tsx
+++ b/web/src/pages/chunk/parsed-result/add-knowledge/components/knowledge-chunk/components/document-view-switch/index.tsx
@@ -24,6 +24,7 @@ interface DocumentViewSwitchProps {
highlights: IHighlight[];
setWidthAndHeight: (width: number, height: number) => void;
url: string;
+ positions?: number[][];
onChunkIdsChange?: (chunkIds: string[]) => void;
}
@@ -33,6 +34,7 @@ export default function DocumentViewSwitch({
highlights,
setWidthAndHeight,
url,
+ positions,
onChunkIdsChange,
}: DocumentViewSwitchProps) {
const { t } = useTranslation();
@@ -100,6 +102,7 @@ export default function DocumentViewSwitch({
highlights={highlights}
setWidthAndHeight={setWidthAndHeight}
url={url}
+ positions={positions}
/>
) : (
<Representation onNodeClick={handleNodeClick} />
diff --git a/web/src/pages/chunk/parsed-result/add-knowledge/components/knowledge-chunk/index.tsx b/web/src/pages/chunk/parsed-result/add-knowledge/components/knowledge-chunk/index.tsx
index 046bb504e0..2f1278de8e 100644
--- a/web/src/pages/chunk/parsed-result/add-knowledge/components/knowledge-chunk/index.tsx
+++ b/web/src/pages/chunk/parsed-result/add-knowledge/components/knowledge-chunk/index.tsx
@@ -10,6 +10,7 @@ import {
useChangeChunkTextMode,
useDeleteChunkByIds,
useGetChunkHighlights,
+ useGetSelectedChunk,
useHandleChunkCardClick,
useUpdateChunk,
} from './hooks';
@@ -38,6 +39,7 @@ import {
QueryStringMap,
useNavigatePage,
} from '@/hooks/logic-hooks/navigate-hooks';
+import { getExtension } from '@/utils/document-util';
import { LucideArrowBigLeft } from 'lucide-react';
function Chunk() {
@@ -166,19 +168,26 @@ function Chunk() {
const { highlights, setWidthAndHeight } =
useGetChunkHighlights(selectedChunkId);
+ const selectedChunk = useGetSelectedChunk(selectedChunkId);
+ const positions = Array.isArray(selectedChunk?.positions)
+ ? selectedChunk.positions
+ : [];
const fileType = useMemo(() => {
+ const name = documentInfo?.name || '';
+ if (name.includes('.')) {
+ return getExtension(name);
+ }
switch (documentInfo?.type) {
case 'doc':
- return documentInfo?.name.split('.').pop() || 'doc';
case 'visual':
- return documentInfo?.name.split('.').pop() || 'visual';
+ return documentInfo?.name?.split('.').pop() || documentInfo.type;
case 'docx':
case 'txt':
case 'md':
case 'mdx':
case 'pdf':
- return documentInfo?.type;
+ return documentInfo.type;
}
return 'unknown';
}, [documentInfo]);
@@ -208,6 +217,7 @@ function Chunk() {
highlights={highlights}
setWidthAndHeight={setWidthAndHeight}
url={fileUrl}
+ positions={positions}
onChunkIdsChange={handleChunkIdsChange}
/>
</article>
diff --git a/web/src/pages/next-search/document-preview-modal/index.tsx b/web/src/pages/next-search/document-preview-modal/index.tsx
index be25e822a9..2f7800abd3 100644
--- a/web/src/pages/next-search/document-preview-modal/index.tsx
+++ b/web/src/pages/next-search/document-preview-modal/index.tsx
@@ -57,6 +57,7 @@ const PdfDrawer = ({
const fileType = documentName ? getFileExtensionRegex(documentName) : '';
const isWebPage = !fileType && !!chunk.document_url;
const url = isWebPage ? (chunk.document_url as string) : getDocumentUrl();
+ const positions = Array.isArray(chunk.positions) ? chunk.positions : [];
return (
<Modal
title={
@@ -88,6 +89,7 @@ const PdfDrawer = ({
highlights={highlights}
setWidthAndHeight={setWidthAndHeight}
url={url}
+ positions={positions}
></DocumentPreview>
</Modal>
);
diff --git a/web/src/utils/__tests__/source-locate.test.ts b/web/src/utils/__tests__/source-locate.test.ts
new file mode 100644
index 0000000000..a3c1088a1a
--- /dev/null
+++ b/web/src/utils/__tests__/source-locate.test.ts
@@ -0,0 +1,15 @@
+import { supportsSourceLocate } from '../source-locate';
+
+describe('supportsSourceLocate', () => {
+ it('accepts pdf and excel extensions case-insensitively', () => {
+ expect(supportsSourceLocate('pdf')).toBe(true);
+ expect(supportsSourceLocate('XLSX')).toBe(true);
+ expect(supportsSourceLocate('xls')).toBe(true);
+ });
+
+ it('rejects unsupported or missing extensions', () => {
+ expect(supportsSourceLocate('html')).toBe(false);
+ expect(supportsSourceLocate('docx')).toBe(false);
+ expect(supportsSourceLocate(undefined)).toBe(false);
+ });
+});
diff --git a/web/src/utils/source-locate.ts b/web/src/utils/source-locate.ts
new file mode 100644
index 0000000000..1518bfe5f7
--- /dev/null
+++ b/web/src/utils/source-locate.ts
@@ -0,0 +1,23 @@
+/*
+ * Copyright 2026 The InfiniFlow Authors. All Rights Reserved.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+/** File types that support in-app citation source location. */
+export const SourceLocateExtensions = ['pdf', 'xlsx', 'xls'] as const;
+
+/** True when the citation should open the in-app preview locate drawer. */
+export const supportsSourceLocate = (ext?: string) =>
+ !!ext &&
+ (SourceLocateExtensions as readonly string[]).includes(ext.toLowerCase());