—output-type=pdf now outputs /UserUnit PDFs at the correct size

This currently distorts the output size because Tesseract assumes it
 knows the DPI better than we do.

Does not work for Ghostscript, because it emerges that Ghostscript
honors /UserUnit for rasterizing but not in pdfwrite (resolve/wontfix).

https://bugs.ghostscript.com/show_bug.cgi?id=690781

Ghostscript’s output would need to be patched in a PDF/A safe way for
this to work. Temporary route may be to block Ghostscript if
/UserUnit.

ocrmypdf/exec/ghostscript.py

@@ -64,7 +64,8 @@ def rasterize_pdf(input_file, output_file, xres, yres, raster_device, log,
             raise SubprocessOutputError()
 
 
-def generate_pdfa(pdf_pages, output_file, compression, log, threads=1):
+def generate_pdfa(*, pdf_version, pdf_pages, output_file, compression, log,
+                  threads=1):
     compression_args = []
     if compression == 'jpeg':
         compression_args = [
@@ -92,6 +93,7 @@ def generate_pdfa(pdf_pages, output_file, compression, log, threads=1):
             "-dQUIET",
             "-dBATCH",
             "-dNOPAUSE",
+            '-dCompatibilityLevel=' + str(pdf_version),
             '-dNumRenderingThreads=' + str(threads),
             "-sDEVICE=pdfwrite",
             "-dAutoRotatePages=/None",

ocrmypdf/exec/qpdf.py

@@ -120,13 +120,14 @@ def split_pages(input_file, work_folder, npages):
         run(args_qpdf, check=True)
 
 
-def merge(input_files, output_file):
+def merge(*, min_version, input_files, output_file):
     """Merge the list of input files (all filenames) into the output file.
 
     The input files may contain one or more pages.
     """
     args_qpdf = [
-        get_program('qpdf'), input_files[0], '--pages'
+        get_program('qpdf'), '--min-version={}'.format(min_version),
+        input_files[0], '--pages'
     ] + input_files + ['--', output_file]
     run(args_qpdf, check=True)
 

ocrmypdf/pipeline.py

@@ -205,10 +205,18 @@ def get_page_dpi(pageinfo, options):
 
 
 def get_page_square_dpi(pageinfo, options):
-    "Get the DPI when we require xres == yres"
+    "Get the working DPI when we require xres == yres"
     return float(max(
-        pageinfo.xres or VECTOR_PAGE_DPI,
-        pageinfo.yres or VECTOR_PAGE_DPI,
+        (pageinfo.xres * pageinfo.userunit) or VECTOR_PAGE_DPI,
+        (pageinfo.yres * pageinfo.userunit) or VECTOR_PAGE_DPI,
+        options.oversample or 0))
+
+
+def get_page_true_square_dpi(pageinfo, options):
+    "Get the true DPI when we require xres == yres, and scaled in userunits"
+    return float(max(
+        (pageinfo.xres) or VECTOR_PAGE_DPI,
+        (pageinfo.yres) or VECTOR_PAGE_DPI,
         options.oversample or 0))
 
 
@@ -414,11 +422,24 @@ def rasterize_with_ghostscript(
               os.path.basename(input_file), device))
 
     # Produce the page image with square resolution or else deskew and OCR
-    # will not work properly
-    dpi = get_page_square_dpi(pageinfo, options)
-    ghostscript.rasterize_pdf(
-        input_file, output_file, xres=dpi, yres=dpi, raster_device=device,
-        log=log)
+    # will not work properly.
+    true_dpi = get_page_true_square_dpi(pageinfo, options)
+    working_dpi = get_page_square_dpi(pageinfo, options)
+
+    if true_dpi == working_dpi:
+        dpi = true_dpi
+        ghostscript.rasterize_pdf(
+            input_file, output_file, xres=dpi, yres=dpi,
+            raster_device=device, log=log)
+    else:
+        # Ghostscript respects /UserUnit when rasterizing so replace the
+        # image's DPI with the working DPI
+        ghostscript.rasterize_pdf(
+            input_file, output_file + '.tmp', xres=true_dpi, yres=true_dpi,
+            raster_device=device, log=log)
+
+        with Image.open(output_file + ".tmp") as im:
+            im.save(output_file, dpi=(working_dpi, working_dpi))
 
 
 def preprocess_remove_background(
@@ -692,6 +713,12 @@ def combine_layers(
     pdf_output = pypdf.PdfFileWriter()
     pdf_output.addPage(page_text)
 
+    # If the input was scaled, re-apply the scaling
+    pageinfo = get_pageinfo(text, context)
+    if pageinfo.userunit != 1:
+        page_text[pypdf.generic.NameObject('/UserUnit')] = pageinfo.userunit
+        pdf_output._header = b'%PDF-1.6'  # Hack header to correct version
+
     with open(output_file, "wb") as out:
         pdf_output.write(out)
 
@@ -835,9 +862,14 @@ def merge_pages_ghostscript(
                    if not f.endswith('.txt'))
     pdf_pages = sorted(input_files, key=input_file_order)
     log.debug("Final pages: " + "\n".join(pdf_pages))
+    input_pdfinfo = context.get_pdfinfo()
     ghostscript.generate_pdfa(
-        pdf_pages, output_file, options.pdfa_image_compression,
-        log, options.jobs or 1)
+        pdf_version=input_pdfinfo.min_version,
+        pdf_pages=pdf_pages,
+        output_file=output_file,
+        compression=options.pdfa_image_compression,
+        log=log,
+        threads=options.jobs or 1)
 
 
 def merge_pages_qpdf(
@@ -879,7 +911,8 @@ def merge_pages_qpdf(
 
     pdf_pages[0] = writer_file
 
-    qpdf.merge(pdf_pages, output_file)
+    qpdf.merge(input_files=pdf_pages, output_file=output_file,
+               min_version=context.get_pdfinfo().min_version)
 
 
 def merge_sidecars(

tests/test_pageinfo.py

@@ -116,4 +116,10 @@ def test_no_contents(resources):
 
     pdf = pdfinfo.PdfInfo(filename)
     assert len(pdf[0].images) == 0
-    assert pdf[0].has_text == False
+    assert pdf[0].has_text == False
+
+
+def test_oversized_page(resources):
+    pdf = pdfinfo.PdfInfo(resources / 'poster.pdf')
+    image = pdf[0].images[0]
+    assert image.width * image.xres > 200, "this is supposed to be oversized"