Fix Font Processing Error

[7shi]

Overview

Fixed an issue where certain PDF files would fail with KeyError: 'china-ss' due to incorrect font handling during doc_zh construction.

Problem Details

I noticed that certain PDF files result in errors. For example:

• https://davidhestenes.net/geocalc/pdf/Tutorial%20on%20Geometric%20Calculus.pdf

The following error occurs:

$ wget https://davidhestenes.net/geocalc/pdf/Tutorial%20on%20Geometric%20Calculus.pdf
$ uv run pdf2zh "Tutorial on Geometric Calculus.pdf"
  6%|███▋                                                      | 1/16 [00:00<00:05,  2.64it/s]
Traceback (most recent call last):
  File "/home/7shi/repos/PDFMathTranslate/.venv/bin/pdf2zh", line 8, in <module>
    sys.exit(main())
  File "/home/7shi/repos/PDFMathTranslate/pdf2zh/pdf2zh.py", line 192, in main
    translate(**vars(parsed_args))
  File "/home/7shi/repos/PDFMathTranslate/pdf2zh/high_level.py", line 366, in translate
    s_mono, s_dual = translate_stream(
  File "/home/7shi/repos/PDFMathTranslate/pdf2zh/high_level.py", line 235, in translate_stream
    obj_patch: dict = translate_patch(fp, prompt=kwarg["prompt"], **locals())
  File "/home/7shi/repos/PDFMathTranslate/pdf2zh/high_level.py", line 165, in translate_patch
    interpreter.process_page(page)
  File "/home/7shi/repos/PDFMathTranslate/pdf2zh/pdfinterp.py", line 270, in process_page
    ops_new = self.device.end_page(page)
  File "/home/7shi/repos/PDFMathTranslate/pdf2zh/converter.py", line 61, in end_page
    return self.receive_layout(self.cur_item)
  File "/home/7shi/repos/PDFMathTranslate/pdf2zh/converter.py", line 409, in receive_layout
    adv = self.fontmap[fcur_].char_width(ord(ch)) * size
KeyError: 'china-ss'

Investigation revealed that fonts were not being added correctly in the translate_stream function. The following debug code shows the difference in font handling between the problematic and working cases:

diff --git a/pdf2zh/high_level.py b/pdf2zh/high_level.py
index 2b65458..289f0c2 100644
--- a/pdf2zh/high_level.py
+++ b/pdf2zh/high_level.py
@@ -230,6 +230,9 @@ def translate_stream(
             except Exception:
                 pass
+    for pnum, page in enumerate(doc_zh, 1):
+        print(pnum, [f[4] for f in page.get_fonts()])
+
     fp = io.BytesIO()
     doc_zh.save(fp)
     obj_patch: dict = translate_patch(fp, prompt=kwarg["prompt"], **locals())

Problematic case:

$ uv run pdf2zh "Tutorial on Geometric Calculus.pdf"
1 ['F2.0', 'F7.0', 'F6.1', 'F1.0', 'F3.0', 'F5.0', 'F4.0', 'F8.0']
MuPDF error: format error: object out of range (310 0 R); xref size 310

MuPDF error: format error: object out of range (311 0 R); xref size 310

2 ['F2.0', 'F1.0', 'F3.0']
MuPDF error: format error: object out of range (310 0 R); xref size 310

MuPDF error: format error: object out of range (311 0 R); xref size 310
(snip)

Working case:

$ wget -O attention.pdf https://arxiv.org/pdf/1706.03762
$ uv run pdf2zh attention.pdf
1 ['F29', 'F56', 'F65', 'F84', 'F87', 'arXivStAmP', 'tiro', 'china-ss']
2 ['F124', 'F23', 'F24', 'F26', 'F27', 'F29', 'F84', 'F87', 'tiro', 'china-ss']
(snip)

Solution

To address this issue, I modified the code to save doc_en to a temporary BytesIO stream before creating doc_zh. This approach was inspired by the existing pattern in the same function where doc_zh is saved to a BytesIO stream before calling translate_patch. Applying the same pattern earlier in the process ensures proper font initialization and prevents the KeyError while maintaining the original functionality.

Testing

I have verified that:

• The previously failing "Tutorial on Geometric Calculus.pdf" now processes successfully
• Existing working PDFs (like attention.pdf) continue to work as expected with no changes in behavior