open stream can raise a FzErrorFormat error instead of FileDataError

[cbm755]

Description of the bug

If I feed a .csv file to pymupdf.open, I get an FileDataError, as documented:

If you attempt to open an unsupported file then PyMuPDF will throw a file data error.

But if I instead pass the bytes of the same file to stream= I get an FzErrorFormat, which I was not expecting from the docs.

How to reproduce the bug

with open('myfile.csv', 'rb') as f:
    file_bytes = f.read()

It probably doesn't matter what's in csv but here's mine:

>> file_bytes
b'A,B,C,D\r\n1,2,1,2\r\n2,2,1,2\r\n'

Now we try to open this:

>> pymupdf.open(stream=file_bytes)
---------------------------------------------------------------------------
FzErrorFormat                             Traceback (most recent call last)
<ipython-input-21-668e9798a921> in ?()
----> 1 pymupdf.open(stream=file_bytes)

~/.local/lib/python3.12/site-packages/pymupdf/__init__.py in ?(self, filename, stream, filetype, rect, width, height, fontsize)
   2884                     self.page_count2 = extra.page_count_pdf
   2885                 else:
   2886                     self.page_count2 = extra.page_count_fz
   2887         finally:
-> 2888             JM_mupdf_show_errors = JM_mupdf_show_errors_old

~/.local/lib/python3.12/site-packages/pymupdf/mupdf.py in ?(magic, stream)
  44292 
  44293         NOTE: The caller retains ownership of 'stream' - the document will take its
  44294         own reference if required.
  44295     """
> 44296     return _mupdf.fz_open_document_with_stream(magic, stream)

FzErrorFormat: code=7: no objects found

---

Contrast this with what happens when I open the file directly:

pymupdf.open("myfile.csv")
---------------------------------------------------------------------------
FzErrorUnsupported                        Traceback (most recent call last)
~/.local/lib/python3.12/site-packages/pymupdf/__init__.py in ?(self, filename, stream, filetype, rect, width, height, fontsize)
   2886                     self.page_count2 = extra.page_count_fz
   2887         finally:
-> 2888             JM_mupdf_show_errors = JM_mupdf_show_errors_old

~/.local/lib/python3.12/site-packages/pymupdf/mupdf.py in ?(filename)
  44271         filename: a path to a file as it would be given to open(2).
  44272     """
> 44273     return _mupdf.fz_open_document(filename)

FzErrorUnsupported: code=6: cannot find document handler for file: myfile.csv

The above exception was the direct cause of the following exception:

FileDataError                             Traceback (most recent call last)
<ipython-input-22-b19d9e4e2772> in ?()
----> 1 pymupdf.open("myfile.csv")

~/.local/lib/python3.12/site-packages/pymupdf/__init__.py in ?(self, filename, stream, filetype, rect, width, height, fontsize)
   2884                     self.page_count2 = extra.page_count_pdf
   2885                 else:
   2886                     self.page_count2 = extra.page_count_fz
   2887         finally:
-> 2888             JM_mupdf_show_errors = JM_mupdf_show_errors_old

FileDataError: Failed to open file 'myfile.csv'.

(we can see it still fails with FzErrorUnSupported but this ultimately raises FileDataError as documented).

PyMuPDF version

1.24.10

Operating system

Linux

Python version

3.12

[julian-smith-artifex-com]

The difference here is that in your file case the filename suffix .csv is available so PyMuPDF tries to open the file as a comma-separated file.

Where as your stream case just specifies the data with no filetype information, so PyMuPDF tries to open the stream as a PDF by default.

Instead you can specify the filetype explicitly with pymupdf.open(stream=file_bytes, filetype='csv').

[cbm755]

Right, but my concern here is about the exception I get. From the docs, I expected FileDataError.

I could certainly change my code to also check for pymupdf.mupdf.FzErrorBase (and try to send you a docs patch) but that seems a bit "private" to me (?)

For example, I have changed my code as follows:

         try:
             ...
         except (pymupdf.FileDataError, KeyError) as e:
            raise ValidationError(f"Unable to open file: {e}") from e
+        except pymupdf.mupdf.FzErrorBase as e:
+            # https://github.com/pymupdf/PyMuPDF/issues/3905
+            raise ValidationError(
+                f"Perhaps not a pdf file?  Unexpected error: {e}"
+            ) from e

[cbm755]

(I have edited to issue title to try to clarify I'm asking about what sort of Exception I should get here)

[julian-smith-artifex-com]

Ah, i see what you mean. The problem is that we don't wrap the internal call to fz_open_document_with_stream() like we do fz_open_document(), so the underlying MuPDF exception leaks out.

I have a fix in my tree.

[julian-smith-artifex-com]

Fixed in 1.24.11.