Fix regression: Disable Ghostscript JPEG passthrough entirely

src/ocrmypdf/exec/ghostscript.py

@@ -32,6 +32,18 @@ def version():
     return get_version('gs')
 
 
+def jpeg_passthrough_available():
+    """
+    Ghostscript 9.23 introduced JPEG passthrough but it seems to corrupt the
+    last two bytes of certain images, for now we disable it for 9.23 and
+    do not mention it for < 9.23.
+
+    https://bugs.ghostscript.com/show_bug.cgi?id=699216
+
+    """
+    return False
+
+
 def _gs_error_reported(stream):
     return re.search(r'error', stream, flags=re.IGNORECASE)
 
@@ -133,7 +145,8 @@ def generate_pdfa(pdf_pages, output_file, compression, log,
     strategy = 'RGB' if version() >= '9.19' else '/RGB'
 
     if version() == '9.23':
-        # 9.23: JPEG passthrough broken for image masks?
+        # 9.23: new feature JPEG passthrough is broken in some cases, best to
+        # disable it always
         # https://bugs.ghostscript.com/show_bug.cgi?id=699216
         compression_args.append('-dPassThroughJPEGImages=false')
 

tests/test_main.py

@@ -610,10 +610,6 @@ def test_masks(spoof_tesseract_noop, resources, outpdf):
     p, out, err = run_ocrmypdf(
         resources / 'masks.pdf', outpdf, env=spoof_tesseract_noop)
 
-    if ghostscript.version() == '9.23' and \
-            p.returncode == ExitCode.invalid_output_pdf:
-        pytest.xfail('https://bugs.ghostscript.com/show_bug.cgi?id=699216')
-
     assert p.returncode == ExitCode.ok
 
 
@@ -906,7 +902,7 @@ def test_compression_changed(spoof_tesseract_noop, ocrmypdf_exec,
     if compression == "jpeg":
         assert pdfimage.enc == Encoding.jpeg
     else:
-        if ghostscript.version() >= '9.23':
+        if ghostscript.jpeg_passthrough_available():
             # Ghostscript 9.23 adds JPEG passthrough, which allows a JPEG to be
             # copied without transcoding - so report
             if image.endswith('jpg'):