Bugs in pyodide

[SF73]

Description of the bug

Hi,
First of all I really like handling pdfs with pymupdf.
Right now, I'm trying to port some small basic tools to pyodide.

I don't know how much effort is put into building if for pyodide but I noticed several bugs or discrepancies.

For example, I encountered #4423 and in order to fix I had to adapt a bit the solution from this comment as doc.xref_object(xref) never raises an exception in pyodide (instead the object is 'null')

// Adapted from https://github.com/pymupdf/PyMuPDF/issues/4423#issuecomment-2773807348
// Should be removed when pymupdf 1.26.0 is released ?

export function fixXref(doc) {
    for (let xref = 1; xref < doc.xref_length(); xref++) {
            const o = doc.xref_object(xref); // never raises exceptions but returns 'null'
            if (o === 'null') {
                doc.update_object(xref, "<<>>");
            }
    }
}

Another issue I noticed is that page.get_image_info(xrefs=True) doesn't contain the xref. Typically :

export async function compress({ pymupdf, pyodide, buffer }) {
    const doc = pymupdf.Document.callKwargs({ stream: pyodide.toPy(buffer) });
    const pageCount = doc.page_count;
    for (const page of doc) {
      console.log("Page:", page);
      const images = page.get_image_info({xrefs:true}).toJs({ dict_converter: Object.fromEntries });
      for (const img of images) {
        console.log("Image:", img);       
  }
}
}

outputs

{
  "number": 5,
  "bbox": [
    192.4720001220703,
    46.254425048828125,
    433.41748046875,
    181.43099975585938
  ],
  "transform": [
    240.94546508789062,
    0,
    0,
    135.17657470703125,
    192.4720001220703,
    46.254425048828125
  ],
  "width": 975,
  "height": 547,
  "colorspace": 3,
  "cs-name": "DeviceRGB",
  "xres": 96,
  "yres": 96,
  "bpc": 8,
  "size": 45700,
  "digest": {
    "0": 130,
    "1": 115,
    "2": 83,
    "3": 79,
    "4": 72,
    "5": 42,
    "6": 35,
    "7": 50,
    "8": 136,
    "9": 201,
    "10": 53,
    "11": 103,
    "12": 95,
    "13": 15,
    "14": 63,
    "15": 136
  },
  "has-mask": false
}

How to reproduce the bug

Info :

• I'm using artifact from github action pymupdf-1.25.5-cp312-abi3-pyodide_2024_0_wasm32.whl
• Pyodide 27.5

PyMuPDF version

1.25.5

Operating system

Other

Python version

3.12

[julian-smith-artifex-com]

Thanks for this.

Unfortunately we don't currently have a way to run our test suite on Pyodide builds.

It looks like cibuildwheel may be getting support for Pyodide which might help with this. But currently it fails to build MuPDF for Pyodide.

[julian-smith-artifex-com]

We've recently pushed some changes that allow cibuildwheel to build and test PyMuPDF on Pyodide.

• We've fixed various Pyodide test failures that became apparent. Most of these were caused by Pyodide's inability to run child processes; we've just disabled the tests that required this.

• Requires some fixes to MuPDF, these are on the current MuPDF master and 1.26.x branches.

• Requires Python-3.13 in order for Pyodide to handle MuPDF's setmp/longjmp-based exceptions.

Here's an example command to be run on Linux that builds and tests with cibuildwheel on Pyodide, taking care of the above requirements:

PyMuPDF/scripts/test.py --cibw-pyodide 1 -m :master cibw

Run PyMuPDF/scripts/test.py -h for details.

We've started to regularly test Pyodide builds on Github so hopefully things will work ok for you in future.

Please let us know how you get on.

[julian-smith-artifex-com]

Closing this as no response, and we now regularly test pyodide builds.