Fix CJK test broken by fpdf2 2.8.7 CFF font encoding change

fpdf2 >= 2.8.7 emits a custom begincidchar Encoding CMap for CFF-based
CID fonts (e.g. NotoSansCJK). pdfminer.six returns <CMap: None> for such
CMaps, so text extraction yields empty output. Switch to pdftotext (poppler)
which handles the new encoding correctly.

.github/workflows/build.yml

@@ -87,7 +87,7 @@ jobs:
           uv run --no-dev pytest --cov-report xml --cov=ocrmypdf --cov=tests/ -n0 tests/
 
       - name: Upload coverage to Codecov
-        uses: codecov/codecov-action@v5
+        uses: codecov/codecov-action@v6
         env:
           CODECOV_TOKEN: ${{ secrets.CODECOV_TOKEN }}
         with:
@@ -149,7 +149,7 @@ jobs:
           uv run --no-dev pytest --cov-report xml --cov=ocrmypdf --cov=tests/ -n0 tests/
 
       - name: Upload coverage to Codecov
-        uses: codecov/codecov-action@v5
+        uses: codecov/codecov-action@v6
         env:
           CODECOV_TOKEN: ${{ secrets.CODECOV_TOKEN }}
         with:
@@ -196,7 +196,7 @@ jobs:
           uv run --no-dev pytest --cov-report xml --cov=ocrmypdf --cov=tests/ -n0 tests/
 
       - name: Upload coverage to Codecov
-        uses: codecov/codecov-action@v5
+        uses: codecov/codecov-action@v6
         env:
           CODECOV_TOKEN: ${{ secrets.CODECOV_TOKEN }}
         with:
@@ -218,7 +218,7 @@ jobs:
         run: |
           uv build --sdist --wheel
 
-      - uses: actions/upload-artifact@v6
+      - uses: actions/upload-artifact@v7
         with:
           name: artifact
           path: |
@@ -235,7 +235,7 @@ jobs:
     steps:
       - uses: actions/checkout@v6
 
-      - uses: actions/download-artifact@v7
+      - uses: actions/download-artifact@v8
         with:
           name: artifact
           path: dist
@@ -284,17 +284,17 @@ jobs:
       - uses: actions/checkout@v6
 
       - name: Login to Docker Hub
-        uses: docker/login-action@v3
+        uses: docker/login-action@v4
         with:
           username: jbarlow83
           password: ${{ secrets.DOCKERHUB_TOKEN }}
 
       - name: Set up QEMU
-        uses: docker/setup-qemu-action@v3
+        uses: docker/setup-qemu-action@v4
 
       - name: Set up Docker Buildx
         id: buildx
-        uses: docker/setup-buildx-action@v3
+        uses: docker/setup-buildx-action@v4
 
       - name: Print image tag
         run: echo "Building image ${DOCKER_REPOSITORY}/${DOCKER_IMAGE_NAME}:${DOCKER_IMAGE_TAG}"
@@ -330,14 +330,14 @@ jobs:
       - uses: actions/checkout@v6
 
       - name: Login to Docker Hub
-        uses: docker/login-action@v3
+        uses: docker/login-action@v4
         with:
           username: jbarlow83
           password: ${{ secrets.DOCKERHUB_TOKEN }}
 
       - name: Set up Docker Buildx
         id: buildx
-        uses: docker/setup-buildx-action@v3
+        uses: docker/setup-buildx-action@v4
 
       - name: Print image tag
         run: echo "Building image ${DOCKER_REPOSITORY}/${DOCKER_IMAGE_NAME}:${DOCKER_IMAGE_TAG}"

.github/workflows/release.yml

@@ -34,7 +34,7 @@ jobs:
 
       # PyPI doesn't support sigstore publishing, so generate after publishing to PyPI
       - name: Sign the dists with Sigstore
-        uses: sigstore/gh-action-sigstore-python@v3.2.0
+        uses: sigstore/gh-action-sigstore-python@v3.3.0
         with:
           inputs: |
             ./dist/*.tar.gz
@@ -87,13 +87,13 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Login to Docker Hub
-        uses: docker/login-action@v3
+        uses: docker/login-action@v4
         with:
           username: jbarlow83
           password: ${{ secrets.DOCKERHUB_TOKEN }}
 
       - name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v3
+        uses: docker/setup-buildx-action@v4
 
       - name: Tag ocrmypdf (Ubuntu) image
         run: |

docs/releasenotes/version17.md

@@ -3,6 +3,41 @@
 
 # v17
 
+## v17.4.2
+
+- Fixed Python API unconditionally overriding ``PIL.Image.MAX_IMAGE_PIXELS``
+  when the caller did not explicitly set ``max_image_mpixels``. Host
+  applications (e.g. Paperless-NGX) that configure the PIL limit before
+  invoking ``ocrmypdf.ocr()`` now have their setting respected. The CLI
+  default of 250 megapixels is unchanged. {issue}`1665`
+- Updated uv.lock to avoid pinning a vulnerable version of Pillow. {issue}`1666`
+
+## v17.4.1
+
+- Fixed RTL text extraction order in the fpdf2 renderer. Arabic lam-alef
+  ligatures and other multi-character CMap entries were garbled by the bidi
+  algorithm during text extraction. {issue}`1655`
+- Fixed ``work_folder`` not being set in ``PdfContext`` options when using
+  the Python API. Thanks @bluebox-steven. {issue}`1613`
+- Updated Ghostscript JPEG corruption warning to include the detected version
+  number, confirming the bug persists in Ghostscript 10.7.0.
+- Internal refactoring.
+- CI dependency updates.
+
+## v17.4.0
+
+- Added ``--no-overwrite`` / ``-n`` option to prevent overwriting output files.
+  If the destination file already exists, OCRmyPDF exits with code 5
+  (``OutputFileAccessError``). {issue}`1642`
+- Fixed text layer stretching in the fpdf2 renderer for widely-spaced words.
+  The horizontal scaling (Tz) was incorrectly stretched to fill inter-word gaps
+  instead of relying on Td positioning, causing text selection to highlight far
+  beyond the actual word boundaries. {issue}`1635`
+- Fixed ``optimize=2`` or ``optimize=3`` crash when using the Python API without
+  explicitly setting ``jpg_quality`` or ``png_quality``. {issue}`1641`
+- Fixed ``verapdf`` availability check crashing with ``NotADirectoryError`` on
+  some platforms. {issue}`1638`
+
 ## v17.3.0
 
 - Fixed Python API ignoring the ``language`` parameter, always defaulting to

pyproject.toml

@@ -6,7 +6,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "ocrmypdf"
-version = "17.3.0"
+version = "17.4.2"
 description = "OCRmyPDF adds an OCR text layer to scanned PDF files, allowing them to be searched"
 readme = "README.md"
 license = "MPL-2.0"
@@ -156,7 +156,13 @@ quote-style = "preserve"
 
 [dependency-groups]
 # Developer-only tools - use `uv sync --group <name>`
-dev = ["mypy>=1.13.0", "ipykernel>=6.29.5", "reportlab>=4.4.4"]
+dev = [
+    "mypy>=1.13.0",
+    "ipykernel>=6.29.5",
+    "reportlab>=4.4.4",
+    "cyclopts>=4.5.1",
+    "pygithub>=2.9.1",
+]
 test = [
   # Core testing framework
   "coverage[toml]>=6.2",

src/ocrmypdf/_exec/verapdf.py

@@ -36,7 +36,7 @@ def available() -> bool:
     """Check if verapdf is available."""
     try:
         version()
-    except MissingDependencyError:
+    except (MissingDependencyError, OSError):
         return False
     return True
 

src/ocrmypdf/_jobcontext.py

@@ -36,6 +36,7 @@ class PdfContext:
         plugin_manager,
     ):
         self.options = options
+        self.options.work_folder = work_folder
         self.work_folder = work_folder
         self.origin = origin
         self.pdfinfo = pdfinfo

src/ocrmypdf/_options.py

@@ -188,8 +188,11 @@ class OcrOptions(BaseModel):
         """Compatibility alias for jpg_quality."""
         self.jpg_quality = value
 
+    # Output behavior
+    no_overwrite: bool = False
+
     # Advanced options
-    max_image_mpixels: float = 250.0
+    max_image_mpixels: float | None = None
     pdf_renderer: str = 'auto'
     ocr_engine: str = 'auto'
     rasterizer: str = 'auto'
@@ -298,7 +301,7 @@ class OcrOptions(BaseModel):
     @classmethod
     def validate_max_image_mpixels(cls, v):
         """Validate max image megapixels."""
-        if v < 0:
+        if v is not None and v < 0:
             raise ValueError("max_image_mpixels must be non-negative")
         return v
 

src/ocrmypdf/_pipelines/_common.py

@@ -329,10 +329,13 @@ def setup_pipeline(
     # Note: OcrOptions is immutable, so we can't modify options.jobs directly
     # The jobs field should already be set correctly during OcrOptions creation
 
-    # Apply PIL max image pixels side effect
+    # Apply PIL max image pixels side effect only when explicitly requested.
-    PIL.Image.MAX_IMAGE_PIXELS = int(options.max_image_mpixels * 1_000_000)
+    # When None, leave PIL.Image.MAX_IMAGE_PIXELS as the host application
-    if PIL.Image.MAX_IMAGE_PIXELS == 0:
+    # configured it. The CLI passes its own default (250.0) via argparse.
-        PIL.Image.MAX_IMAGE_PIXELS = None  # type: ignore
+    if options.max_image_mpixels is not None:
+        PIL.Image.MAX_IMAGE_PIXELS = int(options.max_image_mpixels * 1_000_000)
+        if PIL.Image.MAX_IMAGE_PIXELS == 0:
+            PIL.Image.MAX_IMAGE_PIXELS = None  # type: ignore
 
     pikepdf_enable_mmap()
     executor = setup_executor(plugin_manager)

src/ocrmypdf/_validation.py

@@ -217,6 +217,17 @@ def check_requested_output_file(options: OcrOptions) -> None:
             f"Output file location ({options.output_file}) is not a writable file."
         )
 
+    if (
+        options.no_overwrite
+        and not hasattr(options.output_file, 'writable')
+        and options.output_file != '-'
+        and Path(str(options.output_file)).exists()
+    ):
+        raise OutputFileAccessError(
+            f"Output file already exists: {options.output_file}\n"
+            "To overwrite it, omit the --no-overwrite / -n option."
+        )
+
 
 def report_output_file_size(
     options: OcrOptions,

src/ocrmypdf/_version.py

@@ -1,3 +1,3 @@
 # SPDX-FileCopyrightText: 2022 James R. Barlow
 # SPDX-License-Identifier: MPL-2.0
-__version__ = "17.3.0"
+__version__ = "17.4.2"

src/ocrmypdf/api.py

@@ -440,6 +440,7 @@ def ocr(
     continue_on_soft_render_error: bool | None = None,
     invalidate_digital_signatures: bool | None = None,
     tagged_pdf_mode: str | None = None,
+    no_overwrite: bool | None = None,
     plugins: Iterable[Path | str] | None = None,
     plugin_manager: OcrmypdfPluginManager | None = None,
     keep_temporary_files: bool | None = None,
@@ -502,6 +503,7 @@ def ocr(  # noqa: D417
     continue_on_soft_render_error: bool | None = None,
     invalidate_digital_signatures: bool | None = None,
     tagged_pdf_mode: str | None = None,
+    no_overwrite: bool | None = None,
     plugins: Iterable[Path | str] | None = None,
     plugin_manager: OcrmypdfPluginManager | None = None,
     keep_temporary_files: bool | None = None,

src/ocrmypdf/builtin_plugins/ghostscript.py

@@ -131,10 +131,10 @@ def check_options(options):
             )
         if gs_version >= Version('10.6.0'):
             log.warning(
-                "Ghostscript 10.6.x contains JPEG encoding errors that may corrupt "
+                "Ghostscript %s contains JPEG encoding errors that may corrupt "
-                "images. OCRmyPDF will attempt to mitigate, but this version is "
+                "images. OCRmyPDF will attempt to mitigate, but versions 10.6.0+ "
-                "strongly not recommended. Please upgrade to a newer version. "
+                "are strongly not recommended until this is fixed upstream.",
-                "As of 2025-12, 10.6.0 is the latest version of Ghostscript."
+                gs_version,
             )
         if options.output_type == 'pdfa':
             options.output_type = 'pdfa-2'

src/ocrmypdf/cli.py

@@ -137,8 +137,9 @@ Online documentation is located at:
         'output_file',
         metavar="output_pdf",
         help="Output searchable PDF file (or '-' to write to standard output). "
-        "Existing files will be overwritten. If same as input file, the "
+        "Existing files will be overwritten (use --no-overwrite to prevent this). "
-        "input file will be updated only if processing is successful.",
+        "If same as input file, the input file will be updated only if "
+        "processing is successful.",
     )
     parser.add_argument(
         '-l',
@@ -190,6 +191,15 @@ Online documentation is located at:
         "may not both use stdout at the same time.",
     )
 
+    parser.add_argument(
+        '-n',
+        '--no-overwrite',
+        action='store_true',
+        default=False,
+        help="If the output file already exists, exit with an error instead of "
+        "overwriting it.",
+    )
+
     parser.add_argument(
         '--version',
         action='version',

src/ocrmypdf/fpdf_renderer/renderer.py

@@ -10,6 +10,7 @@ OCR text layers.
 from __future__ import annotations
 
 import logging
+import unicodedata
 from dataclasses import dataclass
 from math import atan, cos, degrees, radians, sin, sqrt
 from pathlib import Path
@@ -24,6 +25,21 @@ from ocrmypdf.models.ocr_element import OcrClass, OcrElement
 log = logging.getLogger(__name__)
 
 
+def _is_rtl_text(text: str) -> bool:
+    """Check if text is right-to-left based on Unicode bidi properties.
+
+    Looks for the first character with a strong directional type
+    (R, AL, or L) to determine the text's base direction.
+    """
+    for char in text:
+        bidi = unicodedata.bidirectional(char)
+        if bidi in ('R', 'AL'):
+            return True
+        if bidi == 'L':
+            return False
+    return False
+
+
 def transform_point(matrix: Matrix, x: float, y: float) -> tuple[float, float]:
     """Transform a point (x, y) by a matrix.
 
@@ -66,6 +82,17 @@ def transform_box(
     )
 
 
+@dataclass
+class WordRenderData:
+    """Rendering parameters for a single word on a line."""
+
+    text: str
+    x_baseline: float
+    font_family: str
+    word_tz: float
+    is_rtl: bool
+
+
 @dataclass
 class DebugRenderOptions:
     """Options for debug visualization during rendering.
@@ -426,8 +453,7 @@ class Fpdf2PdfRenderer:
         ):
             return
 
-        # Collect word rendering data: (text, x_baseline, font_family, word_tz)
+        word_render_data: list[WordRenderData] = []
-        word_render_data: list[tuple[str, float, str, float]] = []
         for word in words:
             if word is None or not word.text or word.bbox is None:
                 continue
@@ -459,13 +485,35 @@ class Fpdf2PdfRenderer:
             )
             font_family = self._register_font(pdf, font_manager)
             pdf.set_font(font_family, size=font_size)
-            natural_width = pdf.get_string_width(word.text)
+
+            # For RTL words with invisible text, we use encode_text()
+            # (which maps characters 1:1 in logical order) combined with
+            # a -1 x-scale text matrix. This avoids an fpdf2 issue where
+            # shaped RTL ligature glyphs (e.g. lam-alef) get multi-char
+            # CMap entries whose character order is reversed by the bidi
+            # algorithm during text extraction.
+            # Since the text is invisible, glyph mirroring is harmless.
+            # Compute Tz using unshaped widths to match encode_text().
+            word_is_rtl = self.invisible_text and _is_rtl_text(word.text)
+            if word_is_rtl:
+                saved_shaping = pdf.text_shaping
+                pdf.text_shaping = None
+                natural_width = pdf.get_string_width(word.text)
+                pdf.text_shaping = saved_shaping
+            else:
+                natural_width = pdf.get_string_width(word.text)
             if natural_width > 0 and word_width_pt > 0:
                 word_tz = (word_width_pt / natural_width) * 100
             else:
                 word_tz = 100.0
 
-            word_render_data.append((word.text, box_llx, font_family, word_tz))
+            word_render_data.append(WordRenderData(
+                text=word.text,
+                x_baseline=box_llx,
+                font_family=font_family,
+                word_tz=word_tz,
+                is_rtl=word_is_rtl,
+            ))
 
         if not word_render_data:
             return
@@ -564,7 +612,7 @@ class Fpdf2PdfRenderer:
     def _emit_line_bt_block(
         self,
         pdf: FPDF,
-        word_render_data: list[tuple[str, float, str, float]],
+        word_render_data: list[WordRenderData],
         baseline_matrix: Matrix,
         font_size: float,
         total_rotation_deg: float,
@@ -580,8 +628,7 @@ class Fpdf2PdfRenderer:
 
         Args:
             pdf: FPDF instance
-            word_render_data: List of (text, x_baseline, font_family, word_tz)
+            word_render_data: List of WordRenderData, one per word on this line
-                tuples, one per word on this line
             baseline_matrix: Transform from baseline coords to page coords
             font_size: Font size in points
             total_rotation_deg: Total rotation angle (textangle + slope)
@@ -626,7 +673,7 @@ class Fpdf2PdfRenderer:
         ops.append(f'{tr} Tr')
 
         # Initial text position
-        first_x_baseline = word_render_data[0][1]
+        first_x_baseline = word_render_data[0].x_baseline
         if has_rotation:
             # In the cm-transformed space, origin is at the baseline start
             ops.append(f'{first_x_baseline:.2f} 0 Td')
@@ -641,14 +688,12 @@ class Fpdf2PdfRenderer:
         prev_font_family: str | None = None
         prev_x_baseline = first_x_baseline
 
-        for i, (text, x_baseline, font_family, word_tz) in enumerate(
+        for i, word in enumerate(word_render_data):
-            word_render_data
-        ):
             is_last = i == len(word_render_data) - 1
 
             # Set font if changed
-            if font_family != prev_font_family:
+            if word.font_family != prev_font_family:
-                pdf.set_font(font_family, size=font_size)
+                pdf.set_font(word.font_family, size=font_size)
                 # Register font resource on this page
                 pdf._resource_catalog.add(
                     PDFResourceType.FONT, pdf.current_font.i, pdf.page
@@ -656,13 +701,13 @@ class Fpdf2PdfRenderer:
                 ops.append(
                     f'/F{pdf.current_font.i} {pdf.font_size_pt:.2f} Tf'
                 )
-                prev_font_family = font_family
+                prev_font_family = word.font_family
 
             # Relative positioning (for words after the first)
             if i > 0:
                 if has_rotation:
                     # In rotated space, advance is purely along x-axis
-                    dx_baseline = x_baseline - prev_x_baseline
+                    dx_baseline = word.x_baseline - prev_x_baseline
                     ops.append(f'{dx_baseline:.2f} 0 Td')
                 else:
                     # Non-rotated: compute delta in PDF coordinates
@@ -670,44 +715,40 @@ class Fpdf2PdfRenderer:
                         baseline_matrix, prev_x_baseline, 0
                     )
                     px_curr, py_curr_f = transform_point(
-                        baseline_matrix, x_baseline, 0
+                        baseline_matrix, word.x_baseline, 0
                     )
                     dx_pdf = px_curr - px_prev
                     # Flip y delta for PDF coordinates (y-up)
                     dy_pdf = -(py_curr_f - py_prev_f)
                     ops.append(f'{dx_pdf:.2f} {dy_pdf:.2f} Td')
 
-            # Determine text to render and compute Tz
+            # Determine text to render
             if not is_last:
-                next_text, next_x_baseline, _, _ = word_render_data[i + 1]
+                next_word = word_render_data[i + 1]
-                advance = next_x_baseline - x_baseline
+                advance = next_word.x_baseline - word.x_baseline
 
-                # Add trailing space unless both words are CJK-only
+                # Add trailing space for text extraction unless both are CJK
                 if (
                     advance > 0
                     and not (
-                        self._is_cjk_only(text)
+                        self._is_cjk_only(word.text)
-                        and self._is_cjk_only(next_text)
+                        and self._is_cjk_only(next_word.text)
                     )
                 ):
-                    text_to_render = text + ' '
+                    text_to_render = word.text + ' '
-                    natural_w = pdf.get_string_width(text_to_render)
-                    render_tz = (
-                        (advance / natural_w) * 100
-                        if natural_w > 0
-                        else word_tz
-                    )
                 else:
-                    text_to_render = text
+                    text_to_render = word.text
-                    render_tz = word_tz
             else:
-                text_to_render = text
+                text_to_render = word.text
-                render_tz = word_tz
 
-            ops.append(f'{render_tz:.2f} Tz')
+            # Use word_tz (fits word into its hOCR bbox) — Td handles
-            ops.append(self._encode_shaped_text(pdf, text_to_render))
+            # inter-word gaps, so Tz should not stretch to fill them.
+            ops.append(f'{word.word_tz:.2f} Tz')
+            ops.append(
+                self._encode_shaped_text(pdf, text_to_render, word.is_rtl)
+            )
 
-            prev_x_baseline = x_baseline
+            prev_x_baseline = word.x_baseline
 
         # End text object
         ops.append('ET')
@@ -721,15 +762,35 @@ class Fpdf2PdfRenderer:
         # don't think Tz is still set from our raw operators
         pdf.font_stretching = 100
 
-    def _encode_shaped_text(self, pdf: FPDF, text: str) -> str:
+    def _encode_shaped_text(
+        self, pdf: FPDF, text: str, is_rtl: bool = False
+    ) -> str:
         """Encode text using HarfBuzz text shaping for complex script support.
 
         Unlike font.encode_text() which maps unicode characters one-by-one to
         glyph IDs, this uses HarfBuzz to handle BiDi reordering, Arabic joining
         forms, Devanagari conjuncts, and other complex script shaping. Falls
         back to encode_text() when text shaping is not enabled.
+
+        For RTL words with invisible text, we use encode_text() instead of
+        shape_text(). fpdf2's shape_text() produces RTL ligature glyphs
+        (e.g. lam-alef) with multi-character CMap entries whose character
+        order gets reversed by the bidi algorithm during text extraction,
+        producing garbled output (e.g. "سالح" instead of "سلاح").
+        encode_text() maps characters 1:1 in logical order, giving correct
+        extraction. Since the text is invisible (Tr=3), the lack of proper
+        joining forms and ligature shaping is harmless.
         """
         font = pdf.current_font
+        if is_rtl:
+            # Reverse the text so that after bidi reversal by the text
+            # extractor, the characters end up in correct logical order.
+            # The text cursor advances left-to-right from the word's left
+            # edge (set by Td), so characters are positioned left-to-right
+            # in the PDF. The extractor sees RTL characters in L-to-R
+            # positions and applies bidi reversal, which reverses them.
+            # By pre-reversing, the double reversal yields the original.
+            return font.encode_text(text[::-1])
         if pdf.text_shaping and pdf.text_shaping.get("use_shaping_engine"):
             shaped = font.shape_text(text, pdf.font_size_pt, pdf.text_shaping)
             if shaped:

src/ocrmypdf/optimize.py

@@ -681,9 +681,9 @@ def optimize(
         safe_symlink(input_file, output_file)
         return output_file
 
-    if options.jpg_quality == 0:
+    if not options.jpg_quality:
         options.jpg_quality = DEFAULT_JPEG_QUALITY if options.optimize < 3 else 40
-    if options.png_quality == 0:
+    if not options.png_quality:
         options.png_quality = DEFAULT_PNG_QUALITY if options.optimize < 3 else 30
 
     with Pdf.open(input_file) as pdf:

tests/conftest.py

@@ -19,14 +19,11 @@ from ocrmypdf.exceptions import ExitCode
 
 
 class Gs106WarningFilter(logging.Filter):
-    """Filter out expected Ghostscript 10.6.x warning from test logs."""
+    """Filter out expected Ghostscript 10.6+ JPEG encoding warning from test logs."""
 
     def filter(self, record: logging.LogRecord) -> bool:
-        # Allow all records except the expected Ghostscript 10.6.x warning
+        # Allow all records except the expected Ghostscript 10.6+ warning
-        return (
+        return "contains JPEG encoding errors" not in record.getMessage()
-            "Ghostscript 10.6.x contains JPEG encoding errors"
-            not in record.getMessage()
-        )
 
 
 @pytest.fixture(autouse=True)

tests/test_fpdf_renderer.py

@@ -5,6 +5,8 @@
 
 from __future__ import annotations
 
+import shutil
+import subprocess
 from pathlib import Path
 
 import pytest
@@ -37,6 +39,26 @@ def resources():
     return Path(__file__).parent / "resources"
 
 
+@pytest.fixture
+def pdftotext():
+    """Return a function to extract text from PDF using pdftotext.
+
+    Skips the test if pdftotext is not available.
+    """
+    pdftotext_path = shutil.which('pdftotext')
+    if pdftotext_path is None:
+        pytest.skip("pdftotext not available")
+
+    def extract_text(pdf_path: Path) -> str:
+        return subprocess.check_output(
+            ['pdftotext', '-enc', 'UTF-8', str(pdf_path), '-'],
+            text=True,
+            encoding='utf-8',
+        )
+
+    return extract_text
+
+
 class TestFpdf2RendererImports:
     """Test that all fpdf2 renderer modules can be imported."""
 
@@ -443,14 +465,16 @@ class TestWordSegmentation:
         assert "World" in words_found
         assert "Test" in words_found
 
-    def test_cjk_no_spurious_spaces(self, multi_font_manager, tmp_path):
+    def test_cjk_no_spurious_spaces(self, multi_font_manager, tmp_path, pdftotext):
         """Test that CJK text does not get spurious spaces inserted.
 
         CJK scripts don't use spaces between characters/words, so we should
         not insert spaces between adjacent CJK words.
-        """
-        from pdfminer.high_level import extract_text
 
+        Uses pdftotext (poppler) instead of pdfminer.six because the latter
+        cannot decode the custom Encoding CMap that fpdf2 >= 2.8.7 emits for
+        subsetted CFF-based CID fonts (e.g. NotoSansCJK).
+        """
         from ocrmypdf.models.ocr_element import BoundingBox, OcrElement
 
         # Create a page with CJK words (Chinese characters)
@@ -487,15 +511,15 @@ class TestWordSegmentation:
         output_path = tmp_path / "test_cjk_segmentation.pdf"
         renderer.render(output_path)
 
-        # Extract text using pdfminer.six
+        extracted_text = pdftotext(output_path)
-        extracted_text = extract_text(str(output_path))
 
         # CJK text should be present
         assert "你好" in extracted_text
         assert "世界" in extracted_text
 
         # There should NOT be spaces between CJK characters
-        # (but pdfminer may add some whitespace, so we check the raw chars)
+        # (a space between the two words is acceptable, since they are
+        # separated horizontally on the rendered page)
         extracted_chars = extracted_text.replace(" ", "").replace("\n", "")
         assert "你好世界" in extracted_chars or (
             "你好" in extracted_chars and "世界" in extracted_chars

tests/test_pdf_renderer.py

@@ -5,9 +5,11 @@
 
 from __future__ import annotations
 
+import re
 from io import StringIO
 from pathlib import Path
 
+import pikepdf
 import pytest
 from pdfminer.converter import TextConverter
 from pdfminer.layout import LAParams
@@ -597,3 +599,312 @@ class TestFpdf2PdfRendererLineTypes:
         check_pdf(str(output_pdf))
         extracted_text = text_from_pdf(output_pdf)
         assert "Caption" in extracted_text
+
+
+def create_rtl_page(
+    words: list[tuple[str, tuple[float, float, float, float]]],
+    language: str = "ara",
+    width: float = 1000,
+    height: float = 500,
+) -> OcrElement:
+    """Create an OcrElement page with a single RTL paragraph/line.
+
+    Args:
+        words: List of (text, (left, top, right, bottom)) tuples.
+        language: Language code for the paragraph.
+        width: Page width in pixels.
+        height: Page height in pixels.
+
+    Returns:
+        OcrElement page.
+    """
+    word_elements = [
+        OcrElement(
+            ocr_class=OcrClass.WORD,
+            text=text,
+            bbox=BoundingBox(
+                left=bbox[0], top=bbox[1], right=bbox[2], bottom=bbox[3]
+            ),
+        )
+        for text, bbox in words
+    ]
+    line = OcrElement(
+        ocr_class=OcrClass.LINE,
+        bbox=BoundingBox(left=50, top=100, right=950, bottom=200),
+        baseline=Baseline(slope=0.0, intercept=0),
+        direction="rtl",
+        children=word_elements,
+    )
+    paragraph = OcrElement(
+        ocr_class=OcrClass.PARAGRAPH,
+        bbox=BoundingBox(left=50, top=100, right=950, bottom=200),
+        direction="rtl",
+        language=language,
+        children=[line],
+    )
+    return OcrElement(
+        ocr_class=OcrClass.PAGE,
+        bbox=BoundingBox(left=0, top=0, right=width, bottom=height),
+        children=[paragraph],
+    )
+
+
+def _tounicode_map(pdf_path: Path) -> dict[int, str]:
+    """Extract all ToUnicode CMap entries from the first page's OCR overlay.
+
+    Returns a dict mapping subset glyph index -> unicode string.
+    """
+    pdf = pikepdf.open(pdf_path)
+    page = pdf.pages[0]
+    resources = page.get('/Resources', {})
+
+    # Collect fonts from the page and from any Form XObjects (OCR overlay)
+    fonts: dict[str, pikepdf.Object] = {}
+    if '/Font' in resources:
+        for name, obj in resources['/Font'].items():
+            fonts[str(name)] = obj
+    for xobj in resources.get('/XObject', {}).values():
+        if xobj.get('/Subtype') == '/Form':
+            for name, obj in xobj.get('/Resources', {}).get('/Font', {}).items():
+                fonts[str(name)] = obj
+
+    result: dict[int, str] = {}
+    for fobj in fonts.values():
+        tounicode = fobj.get('/ToUnicode')
+        if tounicode is None:
+            continue
+        cmap = bytes(tounicode.read_bytes()).decode('latin-1', errors='replace')
+        for m in re.finditer(r'<([0-9A-Fa-f]+)>\s*<([0-9A-Fa-f]+)>', cmap):
+            src_int = int(m.group(1), 16)
+            dst_hex = m.group(2)
+            chars = ''.join(
+                chr(int(dst_hex[i : i + 4], 16))
+                for i in range(0, len(dst_hex), 4)
+                if int(dst_hex[i : i + 4], 16) > 0
+            )
+            if src_int > 0 and chars:
+                result[src_int] = chars
+    return result
+
+
+def _decode_tounicode_stream(
+    pdf_path: Path,
+) -> tuple[dict[int, str], list[int]]:
+    """Extract ToUnicode CMap and Tj glyph stream from a test PDF.
+
+    Searches the page content stream and any Form XObjects for fonts
+    and Tj operations.
+
+    Returns:
+        (cmap, glyph_ids) where *cmap* maps subset index -> Unicode string
+        and *glyph_ids* is the flat list of 2-byte glyph indices found in
+        the first Tj string.
+    """
+    pdf = pikepdf.open(pdf_path)
+    page = pdf.pages[0]
+    resources = page.get('/Resources', {})
+
+    # Collect fonts from page and from Form XObjects
+    cmap: dict[int, str] = {}
+    for font_dict in [resources.get('/Font', {})]:
+        for fobj in font_dict.values():
+            tounicode = fobj.get('/ToUnicode')
+            if tounicode is None:
+                continue
+            raw = bytes(tounicode.read_bytes()).decode('latin-1', errors='replace')
+            for m in re.finditer(r'<([0-9A-Fa-f]+)>\s*<([0-9A-Fa-f]+)>', raw):
+                src = int(m.group(1), 16)
+                dst_hex = m.group(2)
+                chars = ''.join(
+                    chr(int(dst_hex[i : i + 4], 16))
+                    for i in range(0, len(dst_hex), 4)
+                    if int(dst_hex[i : i + 4], 16) > 0
+                )
+                if src > 0 and chars:
+                    cmap[src] = chars
+    for xobj in resources.get('/XObject', {}).values():
+        if xobj.get('/Subtype') != '/Form':
+            continue
+        for fobj in xobj.get('/Resources', {}).get('/Font', {}).values():
+            tounicode = fobj.get('/ToUnicode')
+            if tounicode is None:
+                continue
+            raw = bytes(tounicode.read_bytes()).decode('latin-1', errors='replace')
+            for m in re.finditer(r'<([0-9A-Fa-f]+)>\s*<([0-9A-Fa-f]+)>', raw):
+                src = int(m.group(1), 16)
+                dst_hex = m.group(2)
+                chars = ''.join(
+                    chr(int(dst_hex[i : i + 4], 16))
+                    for i in range(0, len(dst_hex), 4)
+                    if int(dst_hex[i : i + 4], 16) > 0
+                )
+                if src > 0 and chars:
+                    cmap[src] = chars
+
+    # Find first Tj glyph IDs from page content or XObject streams
+    glyph_ids: list[int] = []
+    streams: list[bytes] = []
+    contents = page.get('/Contents')
+    if contents:
+        streams.append(bytes(contents.read_bytes()))
+    for xobj in resources.get('/XObject', {}).values():
+        if xobj.get('/Subtype') == '/Form':
+            streams.append(bytes(xobj.read_bytes()))
+    for data in streams:
+        if glyph_ids:
+            break
+        tj = re.search(rb'\(([^\)]+)\)\s*Tj', data)
+        if tj:
+            raw_bytes = tj.group(1)
+            for j in range(0, len(raw_bytes) - 1, 2):
+                glyph_ids.append((raw_bytes[j] << 8) | raw_bytes[j + 1])
+    return cmap, glyph_ids
+
+
+class TestRtlTextExtraction:
+    """Verify that RTL text is extracted in correct logical order.
+
+    The fpdf2 renderer must produce PDF text layers where text extractors
+    (pdftotext, pdfminer) return characters in correct logical (reading)
+    order for Arabic, Hebrew, and Farsi scripts.
+
+    These tests exercise invisible_text=True (the production path) to
+    catch issues like the lam-alef ligature CMap ordering bug (issue #1655).
+    """
+
+    def test_arabic_lam_alef_extraction_order(self, tmp_path, multi_font_manager):
+        """Arabic words with lam-alef ligature extract in correct order.
+
+        The lam-alef (لا) ligature was the primary trigger for issue #1655:
+        fpdf2's shape_text() produced a multi-char CMap entry whose
+        character order was reversed by the bidi algorithm during
+        extraction, giving "سالم" instead of "سلام".
+        """
+        # سلام contains lam-alef: sin(س) lam(ل) alef(ا) meem(م)
+        page = create_rtl_page(
+            [("سلام", (600, 100, 900, 200))],
+            language="fas",
+        )
+        output_pdf = tmp_path / "rtl_lam_alef.pdf"
+        renderer = Fpdf2PdfRenderer(
+            page=page,
+            dpi=72.0,
+            multi_font_manager=multi_font_manager,
+            invisible_text=True,
+        )
+        renderer.render(output_pdf)
+
+        cmap, glyph_ids = _decode_tounicode_stream(output_pdf)
+        # Decode the glyph stream via the CMap
+        decoded = ''.join(cmap.get(g, '') for g in glyph_ids)
+        # The stream is pre-reversed for RTL, so reversing it back
+        # must yield the original logical text
+        logical = decoded[::-1]
+        assert logical == 'سلام', (
+            f"Expected logical text 'سلام', got {logical!r} "
+            f"(stream: {decoded!r}, glyph_ids: {glyph_ids})"
+        )
+
+    def test_arabic_multiple_words_extraction(self, tmp_path, multi_font_manager):
+        """Multiple Arabic words produce correct Unicode mappings."""
+        page = create_rtl_page(
+            [
+                ("مرحبا", (600, 100, 900, 200)),
+                ("بالعالم", (100, 100, 500, 200)),
+            ],
+            language="ara",
+        )
+        output_pdf = tmp_path / "rtl_arabic_words.pdf"
+        renderer = Fpdf2PdfRenderer(
+            page=page,
+            dpi=72.0,
+            multi_font_manager=multi_font_manager,
+            invisible_text=True,
+        )
+        renderer.render(output_pdf)
+
+        cmap, _ = _decode_tounicode_stream(output_pdf)
+        # Every CMap value should contain valid Arabic characters
+        arabic_chars = {c for chars in cmap.values() for c in chars}
+        expected = set('مرحبابالعالم')
+        assert expected.issubset(arabic_chars | {' '}), (
+            f"CMap missing Arabic characters; got {arabic_chars}"
+        )
+
+    def test_hebrew_extraction_order(self, tmp_path, multi_font_manager):
+        """Hebrew text produces correct stream order for extraction."""
+        page = create_rtl_page(
+            [("שלום", (600, 100, 900, 200))],
+            language="heb",
+        )
+        output_pdf = tmp_path / "rtl_hebrew.pdf"
+        renderer = Fpdf2PdfRenderer(
+            page=page,
+            dpi=72.0,
+            multi_font_manager=multi_font_manager,
+            invisible_text=True,
+        )
+        renderer.render(output_pdf)
+
+        cmap, glyph_ids = _decode_tounicode_stream(output_pdf)
+        decoded = ''.join(cmap.get(g, '') for g in glyph_ids)
+        logical = decoded[::-1]
+        assert logical == 'שלום', (
+            f"Expected logical text 'שלום', got {logical!r} "
+            f"(stream: {decoded!r})"
+        )
+
+    def test_rtl_tounicode_one_to_one(self, tmp_path, multi_font_manager):
+        """RTL invisible text produces 1:1 glyph-to-Unicode CMap entries.
+
+        When using encode_text() for RTL words, each glyph maps to exactly
+        one Unicode character. Multi-char ligature CMap entries (produced by
+        shape_text()) are the root cause of the extraction order bug, so
+        their absence confirms the fix.
+        """
+        page = create_rtl_page(
+            [("سلام", (600, 100, 900, 200))],
+            language="ara",
+        )
+        output_pdf = tmp_path / "rtl_tounicode.pdf"
+        renderer = Fpdf2PdfRenderer(
+            page=page,
+            dpi=72.0,
+            multi_font_manager=multi_font_manager,
+            invisible_text=True,
+        )
+        renderer.render(output_pdf)
+
+        cmap, _ = _decode_tounicode_stream(output_pdf)
+        # Every CMap entry should map to exactly one Unicode character
+        for glyph_id, chars in cmap.items():
+            assert len(chars) == 1, (
+                f"Glyph {glyph_id} maps to {len(chars)} chars {chars!r}; "
+                f"expected 1:1 mapping for RTL invisible text"
+            )
+
+    def test_visible_rtl_still_uses_shaping(self, tmp_path, multi_font_manager):
+        """Visible RTL text (debug mode) still uses text shaping.
+
+        The encode_text() bypass is only for invisible text. When
+        invisible_text=False, shaping must remain active for correct
+        glyph rendering (joining forms, ligatures).
+        """
+        page = create_rtl_page(
+            [("سلام", (600, 100, 900, 200))],
+            language="ara",
+        )
+        output_pdf = tmp_path / "rtl_visible.pdf"
+        renderer = Fpdf2PdfRenderer(
+            page=page,
+            dpi=72.0,
+            multi_font_manager=multi_font_manager,
+            invisible_text=False,
+        )
+        renderer.render(output_pdf)
+
+        # Shaped text may have multi-char CMap entries (ligatures);
+        # just verify the PDF is valid and non-empty
+        check_pdf(str(output_pdf))
+        text = text_from_pdf(output_pdf)
+        assert len(text.strip()) > 0, "Visible RTL should produce extractable text"

tests/test_validation.py

@@ -102,6 +102,45 @@ def test_pillow_options():
     with pytest.raises(ValueError, match="max_image_mpixels must be non-negative"):
         make_ocr_opts(max_image_mpixels=-1)
 
+    # Default is None, meaning "do not override host-set PIL.Image.MAX_IMAGE_PIXELS"
+    opts = make_ocr_opts()
+    assert opts.max_image_mpixels is None
+
+
+def test_pillow_max_image_pixels_not_overridden_when_unset():
+    """Issue #1665: respect host-set PIL.Image.MAX_IMAGE_PIXELS.
+
+    API callers (e.g. Paperless-NGX) that set PIL.Image.MAX_IMAGE_PIXELS
+    before invoking ocrmypdf should not have their setting clobbered when
+    max_image_mpixels is not explicitly passed.
+    """
+    import PIL.Image
+
+    from ocrmypdf._pipelines._common import setup_pipeline
+
+    parser = get_parser()
+    pm = setup_plugin_infrastructure(plugins=[])
+    pm.add_options(parser=parser)
+
+    saved = PIL.Image.MAX_IMAGE_PIXELS
+    try:
+        PIL.Image.MAX_IMAGE_PIXELS = None  # host disables the limit
+        opts = make_ocr_opts()
+        assert opts.max_image_mpixels is None
+        setup_pipeline(opts, pm)
+        assert PIL.Image.MAX_IMAGE_PIXELS is None
+
+        PIL.Image.MAX_IMAGE_PIXELS = 1_000_000_000  # host sets a high limit
+        setup_pipeline(opts, pm)
+        assert PIL.Image.MAX_IMAGE_PIXELS == 1_000_000_000
+
+        # When explicitly passed, it still takes effect.
+        opts = make_ocr_opts(max_image_mpixels=100)
+        setup_pipeline(opts, pm)
+        assert PIL.Image.MAX_IMAGE_PIXELS == 100_000_000
+    finally:
+        PIL.Image.MAX_IMAGE_PIXELS = saved
+
 
 def test_output_tty():
     with patch('sys.stdout.isatty', return_value=True), pytest.raises(BadArgsError):