docvortex 0.4.16__tar.gz → 0.4.18__tar.gz

{docvortex-0.4.16/src/docvortex.egg-info → docvortex-0.4.18}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: docvortex
-Version: 0.4.16
+Version: 0.4.18
 Summary: A fast, multi-format document parsing and conversion engine
 License-Expression: MIT
 Project-URL: Repository, https://github.com/myhloli/DocVortex

{docvortex-0.4.16 → docvortex-0.4.18}/src/docvortex/analyzers/native/office/spreadsheet/projector.py

@@ -708,10 +708,11 @@ class SpreadsheetProjector:
             if cell.alignment.vertical:
                 style["vertical-align"] = cell.alignment.vertical
 
-        if cell.fill and cell.fill.patternType == "solid" and cell.fill.fgColor:
-            # handle bg color
-            color = cell.fill.fgColor.rgb
-            if hasattr(cell.fill.fgColor, "type") and cell.fill.fgColor.type == "rgb" and color:
+        # 渐变填充没有 patternType；只提取既有纯色背景，其他填充仍保留内容及其余样式。
+        fill = cell.fill
+        if getattr(fill, "patternType", None) == "solid" and fill.fgColor:
+            color = fill.fgColor.rgb
+            if hasattr(fill.fgColor, "type") and fill.fgColor.type == "rgb" and color:
                 if isinstance(color, str) and len(color) == 8:
                     style["background-color"] = "#" + color[2:]
         return style

{docvortex-0.4.16 → docvortex-0.4.18}/src/docvortex/analyzers/native/pdf/line_layout.py

@@ -679,78 +679,65 @@ def _reattach_span_lane_continuations(
             span_lane.lines.sort(key=lambda item: (item[1][1], item[1][0], item[0].source_index))
 
 
+def _lane_accepts_short_tail(
+    candidate: tuple[_LineItem, BBox],
+    lane: _TextLane,
+    median_height: float,
+) -> bool:
+    """用同一组前序行证据判断当前栏或其他栏是否能够接纳正文短尾。"""
+
+    candidate_line, candidate_bbox = candidate
+    preceding = [
+        item for item in lane.lines if item[0].semantic_type == candidate_line.semantic_type and item[1][1] < candidate_bbox[1]
+    ]
+    if not preceding:
+        return False
+    # 同位置前序行按栏内稳定排序取 source_index 较小者，避免后续插入触发排序后改变已用证据。
+    previous = max(preceding, key=lambda item: (item[1][1], item[1][0], -item[0].source_index))
+    previous_line, previous_bbox = previous
+    pair_height = max(
+        _line_effective_height(*previous),
+        _line_effective_height(*candidate),
+        median_height,
+    )
+    lane_width = max(0.1, lane.right - lane.left)
+    if (
+        previous_bbox[2] - previous_bbox[0] < 0.65 * lane_width
+        or candidate_bbox[2] - candidate_bbox[0] > 0.85 * lane_width
+        or candidate_bbox[0] < lane.left - 0.75 * pair_height
+        or candidate_bbox[2] > lane.right + 0.75 * pair_height
+        or abs(candidate_bbox[0] - previous_bbox[0]) > 0.75 * pair_height
+        or not _title_fonts_compatible(previous_line, candidate_line)
+    ):
+        return False
+    gap = _effective_body_text_row_gap(previous, candidate)
+    if not -0.25 * pair_height <= gap <= 0.9 * pair_height:
+        return False
+    if (
+        previous_line.visual_row_id is not None
+        and candidate_line.visual_row_id is not None
+        and not 0 < candidate_line.visual_row_id - previous_line.visual_row_id <= 2
+    ):
+        return False
+    return True
+
+
 def _reattach_cross_lane_short_tails(
     lanes: list[_TextLane],
     median_height: float,
 ) -> None:
-    """把误入另一栏带、但完整落在唯一前序栏内的正文短尾迁回原栏。"""
-
-    while True:
-        moves: list[
-            tuple[
-                float,
-                _TextLane,
-                _TextLane,
-                tuple[_LineItem, BBox],
-            ]
-        ] = []
-        for source_lane in lanes:
-            for candidate in source_lane.lines:
-                candidate_line, candidate_bbox = candidate
-                if candidate_line.semantic_type is not None:
-                    continue
-                matches: list[tuple[float, _TextLane]] = []
-                for target_lane in lanes:
-                    if target_lane is source_lane or not target_lane.lines:
-                        continue
-                    preceding = [
-                        item
-                        for item in target_lane.lines
-                        if item[0].semantic_type == candidate_line.semantic_type and item[1][1] < candidate_bbox[1]
-                    ]
-                    if not preceding:
-                        continue
-                    previous = max(
-                        preceding,
-                        key=lambda item: (item[1][1], item[1][0]),
-                    )
-                    previous_line, previous_bbox = previous
-                    pair_height = max(
-                        _line_effective_height(*previous),
-                        _line_effective_height(*candidate),
-                        median_height,
-                    )
-                    lane_width = max(0.1, target_lane.right - target_lane.left)
-                    if (
-                        previous_bbox[2] - previous_bbox[0] < 0.65 * lane_width
-                        or candidate_bbox[2] - candidate_bbox[0] > 0.85 * lane_width
-                        or candidate_bbox[0] < target_lane.left - 0.75 * pair_height
-                        or candidate_bbox[2] > target_lane.right + 0.75 * pair_height
-                        or abs(candidate_bbox[0] - previous_bbox[0]) > 0.75 * pair_height
-                        or not _title_fonts_compatible(previous_line, candidate_line)
-                    ):
-                        continue
-                    gap = _effective_body_text_row_gap(previous, candidate)
-                    if not -0.25 * pair_height <= gap <= 0.9 * pair_height:
-                        continue
-                    if (
-                        previous_line.visual_row_id is not None
-                        and candidate_line.visual_row_id is not None
-                        and not 0 < candidate_line.visual_row_id - previous_line.visual_row_id <= 2
-                    ):
-                        continue
-                    matches.append((max(0.0, gap), target_lane))
-                if len(matches) == 1:
-                    gap, target_lane = matches[0]
-                    moves.append((candidate_bbox[1] + gap, source_lane, target_lane, candidate))
-        if not moves:
-            return
-        _score, source_lane, target_lane, candidate = min(
-            moves,
-            key=lambda item: item[0],
-        )
-        if candidate not in source_lane.lines:
+    """按前序依赖顺序，把具有唯一栏归属的正文短尾迁回对应栏。"""
+
+    # 栏边界固定，且匹配只依赖 y0 严格更小的行；先确定上方行后，每行至多迁移一次。
+    # 保留原始对象及其来源栏，不能用 source_index 合并或跳过不同的文本行。
+    pending = [(lane, candidate) for lane in lanes for candidate in lane.lines if candidate[0].semantic_type is None]
+    pending.sort(key=lambda item: (item[1][1][1], item[1][1][0], item[1][0].source_index))
+    for source_lane, candidate in pending:
+        # 当前栏也必须参与唯一性判断；多个栏均能接纳时保留原位，避免双向迁移。
+        matches = [lane for lane in lanes if _lane_accepts_short_tail(candidate, lane, median_height)]
+        if len(matches) != 1 or matches[0] is source_lane:
             continue
+        target_lane = matches[0]
         source_lane.lines.remove(candidate)
         target_lane.lines.append(candidate)
         target_lane.lines.sort(

{docvortex-0.4.16 → docvortex-0.4.18}/src/docvortex/render/_internal/html/renderer.py

@@ -70,14 +70,14 @@ from ....schema import (
 )
 
 _STYLE_RESOURCE_NAME = "docvortex.min.css"
-_MATHJAX_URL = "https://cdn.jsdelivr.net/npm/mathjax@4.1.2/tex-chtml.js"
+_MATHJAX_URL = "https://gcore.jsdelivr.net/npm/mathjax@4.1.2/tex-chtml.js"
 _MATHJAX_INTEGRITY = "sha384-zAhQQhdaMeHsMProNntGGg6nOUVcfuF9F22C3d1qJ9NZAVzCplXk1X85D2O5iufn"
-_PRISM_CORE_URL = "https://cdn.jsdelivr.net/npm/prismjs@1.30.0/components/prism-core.min.js"
+_PRISM_CORE_URL = "https://gcore.jsdelivr.net/npm/prismjs@1.30.0/components/prism-core.min.js"
 _PRISM_CORE_INTEGRITY = "sha384-zLRFO4dwowZvh8kzutOb5AWhH7f39HeJp+N7PtHF1SQtTBnifRx0AtmvTYs3F4YV"
-_PRISM_AUTOLOADER_URL = "https://cdn.jsdelivr.net/npm/prismjs@1.30.0/plugins/autoloader/prism-autoloader.min.js"
+_PRISM_AUTOLOADER_URL = "https://gcore.jsdelivr.net/npm/prismjs@1.30.0/plugins/autoloader/prism-autoloader.min.js"
 _PRISM_AUTOLOADER_INTEGRITY = "sha384-Uq05+JLko69eOiPr39ta9bh7kld5PKZoU+fF7g0EXTAriEollhZ+DrN8Q/Oi8J2Q"
-_PRISM_LANGUAGES_PATH = "https://cdn.jsdelivr.net/npm/prismjs@1.30.0/components/"
-_MERMAID_URL = "https://cdn.jsdelivr.net/npm/mermaid@11.16.1/dist/mermaid.min.js"
+_PRISM_LANGUAGES_PATH = "https://gcore.jsdelivr.net/npm/prismjs@1.30.0/components/"
+_MERMAID_URL = "https://gcore.jsdelivr.net/npm/mermaid@11.16.1/dist/mermaid.min.js"
 _MERMAID_INTEGRITY = "sha384-aBQXj4hK6Jm05i7aQAsUV3bLdSUrHX1BGYfMB0166TtWt/RRaw+h0Eelme9OCOvy"
 _MERMAID_MAX_TEXT_SIZE = 50_000
 _MERMAID_FLOWCHART_HEADER_RE = re.compile(

{docvortex-0.4.16 → docvortex-0.4.18}/src/docvortex/version.py

@@ -1,3 +1,3 @@
-__version__ = "0.4.16"
+__version__ = "0.4.18"
 
 __all__ = ["__version__"]

{docvortex-0.4.16 → docvortex-0.4.18/src/docvortex.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: docvortex
-Version: 0.4.16
+Version: 0.4.18
 Summary: A fast, multi-format document parsing and conversion engine
 License-Expression: MIT
 Project-URL: Repository, https://github.com/myhloli/DocVortex

{docvortex-0.4.16 → docvortex-0.4.18}/src/docvortex.egg-info/SOURCES.txt

@@ -567,6 +567,7 @@ tests/unittest/test_flash_pdf_reviewed_history.py
 tests/unittest/test_flash_pdf_samples.py
 tests/unittest/test_flash_pdf_scripts.py
 tests/unittest/test_flash_pdf_semantics.py
+tests/unittest/test_flash_pdf_short_tails.py
 tests/unittest/test_flash_pdf_tables.py
 tests/unittest/test_flash_pdf_text_blocks.py
 tests/unittest/test_flash_pdf_text_visibility.py

docvortex-0.4.18/tests/unittest/test_flash_pdf_short_tails.py

@@ -0,0 +1,184 @@
+"""覆盖短尾迁移的终止性、唯一归属及前序依赖；死循环反例在可回收子进程内运行。"""
+
+from __future__ import annotations
+
+import subprocess
+import sys
+from collections import Counter
+from pathlib import Path
+
+import pytest
+
+from docvortex.analyzers.native.pdf.line_layout import _infer_text_lanes, _reattach_cross_lane_short_tails
+from docvortex.analyzers.native.pdf.models import _LineItem, _TextLane
+from docvortex.schema import BBox
+
+
+def _line(text: str, bbox: BBox, index: int) -> tuple[_LineItem, BBox]:
+    """构造具有明确有效行高的文本行及局部几何。"""
+    return _LineItem(text=text, bbox=bbox, angle=0, source_index=index, effective_height=10.0), bbox
+
+
+def _ambiguous_lanes(offset: float = 0.0, tails: int = 1) -> list[_TextLane]:
+    """构造两栏均可接纳的短尾组，并用横向偏移隔离不同组。"""
+    first = _TextLane(left=offset, right=offset + 200.0)
+    second = _TextLane(left=offset, right=offset + 200.0)
+    first.lines.append(_line("first predecessor", (offset + 10, 100, offset + 150, 110), 0))
+    second.lines.append(_line("second predecessor", (offset + 10, 100, offset + 150, 110), 1))
+    for index in range(tails):
+        first.lines.append(_line(f"tail {index}", (offset + 15, 118, offset + 140, 128), index + 2))
+    return [first, second]
+
+
+def _membership(lanes: list[_TextLane]) -> tuple[tuple[int, ...], ...]:
+    """按对象身份记录栏内顺序，避免重复文本或 source_index 掩盖丢行。"""
+    return tuple(tuple(id(line) for line, _bbox in lane.lines) for lane in lanes)
+
+
+def _run_and_check_stability(lanes: list[_TextLane]) -> None:
+    """验证迁移不丢行、不重复、不改边界，且再次执行保持同一结果。"""
+    before = Counter(id(line) for lane in lanes for line, _bbox in lane.lines)
+    bounds = [(lane.left, lane.right, lane.is_span) for lane in lanes]
+    _reattach_cross_lane_short_tails(lanes, 10.0)
+    assert Counter(id(line) for lane in lanes for line, _bbox in lane.lines) == before
+    assert [(lane.left, lane.right, lane.is_span) for lane in lanes] == bounds
+    settled = _membership(lanes)
+    _reattach_cross_lane_short_tails(lanes, 10.0)
+    assert _membership(lanes) == settled
+
+
+def _case_ambiguous_current_lane() -> None:
+    """验证当前栏与其他栏同时匹配时保留原位，不发生往返振荡。"""
+    lanes = _ambiguous_lanes()
+    original = _membership(lanes)
+    _run_and_check_stability(lanes)
+    assert _membership(lanes) == original
+
+
+def _case_independent_groups_and_valid_tail() -> None:
+    """验证多个独立歧义组不会耗尽轮数或阻止稍后的合法短尾迁回。"""
+    lanes = _ambiguous_lanes(tails=8) + _ambiguous_lanes(offset=400.0, tails=8)
+    original = _membership(lanes)
+    destination = _TextLane(left=810.0, right=1000.0)
+    previous = _line("late predecessor", (815, 200, 965, 210), 0)
+    tail = _line("late valid tail", (815, 218, 965, 228), 1)
+    destination.lines.append(previous)
+    lanes[0].lines.append(tail)
+    lanes.append(destination)
+    _run_and_check_stability(lanes)
+    assert _membership(lanes[:-1]) == original
+    assert destination.lines == [previous, tail]
+
+
+def _case_inferred_body_and_span_lanes() -> None:
+    """验证正常栏推断生成的正文栏和跨栏栏带不会互相争抢短尾。"""
+    geometry = [_line("wide predecessor", (0, 100, 300, 110), 0)]
+    for y in (103.0, 140.0, 160.0):
+        for x in (0.0, 160.0):
+            geometry.append(_line(f"column {x} {y}", (x, y, x + 140, y + 10), len(geometry)))
+    tail = _line("tail", (0, 118, 100, 128), len(geometry))
+    geometry.append(tail)
+    lanes = _infer_text_lanes(geometry, 300.0, 10.0)
+    assert len(lanes) == 3
+    assert any(lane.is_span for lane in lanes)
+    assert tail in lanes[0].lines
+    _run_and_check_stability(lanes)
+
+
+def _case_chained_tails() -> None:
+    """验证逆序输入的连续短尾依次迁回，重复 source_index 不会漏掉后续行。"""
+    previous = _line("predecessor", (10, 100, 150, 110), 0)
+    first = _line("first tail", (10, 118, 150, 128), 1)
+    second = _line("second tail", (10, 136, 130, 146), 1)
+    source = _TextLane(left=210, right=400, lines=[second, first])
+    target = _TextLane(left=0, right=200, lines=[previous])
+    _run_and_check_stability([source, target])
+    assert source.lines == []
+    assert target.lines == [previous, first, second]
+
+
+def _case_predecessor_changes_membership() -> None:
+    """验证先迁移上方行后再判断短尾，不能用最初的栏状态一次性计算所有目标。"""
+    anchor = _line("anchor", (0, 100, 200, 110), 0)
+    first = _line("first tail", (0, 118, 140, 128), 1)
+    second = _line("second tail", (0, 136, 100, 146), 2)
+    source = _TextLane(left=0, right=300, lines=[second, first])
+    target = _TextLane(left=0, right=200, lines=[anchor])
+    _run_and_check_stability([source, target])
+    assert source.lines == []
+    assert target.lines == [anchor, first, second]
+
+
+def _case_equal_height_peers() -> None:
+    """验证同高度的先处理行不会成为另一个同行片段的前序证据。"""
+    anchor = _line("anchor", (10, 100, 150, 110), 0)
+    first = _line("first peer", (10, 118, 150, 128), 1)
+    second = _line("second peer", (10, 118, 130, 128), 2)
+    first[0].visual_row_id = 2
+    second[0].visual_row_id = 2
+    anchor[0].visual_row_id = 1
+    source = _TextLane(left=210, right=400, lines=[second, first])
+    target = _TextLane(left=0, right=200, lines=[anchor])
+    _run_and_check_stability([source, target])
+    assert source.lines == []
+    assert target.lines == [anchor, first, second]
+
+
+def _case_non_unique_or_missing_target() -> None:
+    """验证两个其他栏匹配、没有匹配和非正文行均保持原位。"""
+    targets = _ambiguous_lanes(tails=0)
+    ambiguous = _line("two targets", (15, 118, 140, 128), 2)
+    missing = _line("no target", (415, 118, 540, 128), 3)
+    title = _line("title", (15, 118, 140, 128), 4)
+    title[0].semantic_type = "paragraph_title"
+    source = _TextLane(left=600, right=800, lines=[ambiguous, missing, title])
+    lanes = [source, *targets]
+    original = _membership(lanes)
+    _run_and_check_stability(lanes)
+    assert _membership(lanes) == original
+
+
+def _case_same_position_predecessors() -> None:
+    """验证同坐标前序行的平局选择不随迁入后排序改变，首次调用即得到稳定归属。"""
+    first_anchor = _line("A anchor", (10, 100, 150, 110), 1)
+    second_anchor = _line("B anchor", (10, 100, 150, 110), 2)
+    first_tail = _line("A tail", (10, 118, 150, 128), 3)
+    second_tail = _line("B tail", (10, 118.5, 150, 128.5), 4)
+    for item, family in ((first_anchor, "A"), (second_anchor, "B"), (first_tail, "A"), (second_tail, "B")):
+        item[0].font_signature = (family, 0)
+        item[0].font_coverage = 1.0
+    target = _TextLane(left=0, right=200, lines=[second_anchor, first_anchor])
+    source = _TextLane(left=300, right=500, lines=[first_tail, second_tail])
+    _run_and_check_stability([target, source])
+    assert target.lines == [first_anchor, second_anchor, first_tail]
+    assert source.lines == [second_tail]
+
+
+@pytest.mark.parametrize(
+    "case",
+    [
+        "ambiguous_current_lane",
+        "independent_groups_and_valid_tail",
+        "inferred_body_and_span_lanes",
+        "chained_tails",
+        "predecessor_changes_membership",
+        "equal_height_peers",
+        "non_unique_or_missing_target",
+        "same_position_predecessors",
+    ],
+)
+def test_short_tail_assignment_terminates_and_preserves_members(case: str) -> None:
+    """在可强制终止并回收的子进程内检查归属，任何异常通过退出码和标准错误上报。"""
+    completed = subprocess.run(
+        [sys.executable, str(Path(__file__).resolve()), case],
+        capture_output=True,
+        text=True,
+        encoding="utf-8",
+        timeout=20,
+        check=False,
+    )
+    assert completed.returncode == 0, completed.stdout + completed.stderr
+
+
+if __name__ == "__main__":
+    globals()[f"_case_{sys.argv[1]}"]()

{docvortex-0.4.16 → docvortex-0.4.18}/tests/unittest/test_flash_spreadsheet_projector.py

@@ -1,13 +1,17 @@
 from __future__ import annotations
 
+from io import BytesIO
+
 import pytest
 from _span_test_utils import inline
 from bs4 import BeautifulSoup
 from openpyxl import Workbook
 from openpyxl.cell.rich_text import CellRichText, TextBlock
 from openpyxl.cell.text import InlineFont
+from openpyxl.styles import Alignment, Font, GradientFill, PatternFill
 from openpyxl.worksheet.worksheet import Worksheet
 
+from docvortex.analyzers.native import XlsxModel
 from docvortex.analyzers.native.office.spreadsheet.html import render_spreadsheet_table
 from docvortex.analyzers.native.office.spreadsheet.models import AnchoredBlock, ExcelCell, ExcelTable, SheetImage
 from docvortex.analyzers.native.office.spreadsheet.projector import SpreadsheetProjector
@@ -17,6 +21,72 @@ from docvortex.analyzers.native.office.xlsx.xlsx_converter import XlsxConverter
 from docvortex.schema import BlockType
 
 
+@pytest.mark.parametrize(
+    ("fill", "background"),
+    [
+        (PatternFill(), None),
+        (PatternFill(patternType="solid", fgColor="FF99CCFF"), "#99CCFF"),
+        (PatternFill(patternType="darkGrid", fgColor="FF99CCFF"), None),
+        (GradientFill(stop=("FFFFFF", "99CCFF")), None),
+        (GradientFill(type="path", stop=("FFFFFF", "99CCFF")), None),
+    ],
+    ids=["none", "solid", "pattern", "linear-gradient", "path-gradient"],
+)
+def test_cell_fill_preserves_content_and_supported_styles(fill: PatternFill | GradientFill, background: str | None) -> None:
+    """验证不同填充经过 StyleProxy 后均能提取内容，渐变不影响字体、对齐或纯色支持。"""
+    workbook = Workbook()
+    sheet = workbook.active
+    assert sheet is not None
+    sheet["A1"] = 240
+    sheet["A1"].fill = fill
+    sheet["A1"].font = Font(b=True, i=True, color="FF112233")
+    sheet["A1"].alignment = Alignment(horizontal="center", vertical="top")
+    styles = SpreadsheetProjector()._extract_cell_style(sheet["A1"])
+    expected = {
+        "font-weight": "bold",
+        "font-style": "italic",
+        "color": "#112233",
+        "text-align": "center",
+        "vertical-align": "top",
+    }
+    if background is not None:
+        expected["background-color"] = background
+    assert styles == expected
+
+    # 单格也会经过样式提取；通过公共入口锁定 issue 中的最小失败路径。
+    stream = BytesIO()
+    workbook.save(stream)
+    pages = XlsxModel().predict(BytesIO(stream.getvalue()))
+    assert pages == [[{"type": BlockType.TEXT, "content": inline("240")}]]
+
+
+def test_xlsx_mixed_fills_preserve_all_sheets_and_merged_cells() -> None:
+    """验证两类渐变与纯色混用时仍输出全部工作表、单元格内容和合并跨度。"""
+    workbook = Workbook()
+    sheet = workbook.active
+    assert sheet is not None
+    sheet.title = "Mixed"
+    sheet.append(["linear", "path", "solid"])
+    sheet.append(["merged", None, "plain"])
+    sheet.merge_cells("A2:B2")
+    sheet["A1"].fill = GradientFill(stop=("FFFFFF", "99CCFF"))
+    sheet["B1"].fill = GradientFill(type="path", stop=("FFFFFF", "99CCFF"))
+    sheet["C1"].fill = PatternFill(patternType="solid", fgColor="FF99CCFF")
+    sheet["A2"].fill = GradientFill(stop=("FFFFFF", "99CCFF"))
+    workbook.create_sheet("Following")["A1"] = "still parsed"
+    stream = BytesIO()
+    workbook.save(stream)
+
+    pages = XlsxModel().predict(BytesIO(stream.getvalue()))
+    assert len(pages) == 2
+    table = next(block for block in pages[0] if block["type"] == BlockType.TABLE)
+    soup = BeautifulSoup(table["content"], "html.parser")
+    cells = soup.find_all(["th", "td"])
+    assert [cell.get_text() for cell in cells] == ["linear", "path", "solid", "merged", "plain"]
+    assert cells[3]["colspan"] == "2"
+    assert pages[1][-1] == {"type": BlockType.TEXT, "content": inline("still parsed")}
+
+
 def _cell(
     row: int,
     col: int,