pdfviewer using PyMuPDF broken with recent versions of PyMuPDF. PATCH included

[jmoraleda]

The following snippet of code (extracted from https://wxpython.org/Phoenix/docs/html/wx.lib.pdfviewer.html?highlight=pdf#usage by removing the button panel) demonstrates the problem:

import wx
import wx.lib.sized_controls as sc

from wx.lib.pdfviewer import pdfViewer

class PDFViewer(sc.SizedFrame):
    def __init__(self, parent, **kwds):
        super(PDFViewer, self).__init__(parent, **kwds)

        paneCont = self.GetContentsPane()

        self.viewer = pdfViewer(paneCont, wx.ID_ANY, wx.DefaultPosition,
                                wx.DefaultSize,
                                wx.HSCROLL|wx.VSCROLL|wx.SUNKEN_BORDER)

        self.viewer.SetSizerProps(expand=True, proportion=1)


if __name__ == '__main__':
    import wx.lib.mixins.inspection as WIT
    app = WIT.InspectableApp(redirect=False)

    pdfV = PDFViewer(None, size=(800, 600))
    pdfV.viewer.LoadFile(r'path_to_any_pdf_file.pdf')
    pdfV.Show()

    app.MainLoop()

The code works correctly with older versions of PyMuPDF(up to 1.14) but fails with more recent versions (1.15 and above). This is the complete output written to the console:

python .\pdftest.py
pdfviewer using PyMuPDF (GPL)
Traceback (most recent call last):
  File "C:\Program Files\Python37\lib\site-packages\wx\core.py", line 3259, in <lambda>
    lambda event: event.callable(*event.args, **event.kw) )
  File "C:\Program Files\Python37\lib\site-packages\wx\lib\pdfviewer\viewer.py", line 433, in Render
    self.pdfdoc.RenderPage(gc, pageno, scale=self.scale)
  File "C:\Program Files\Python37\lib\site-packages\wx\lib\pdfviewer\viewer.py", line 518, in RenderPage
    bmp = wx.Bitmap.FromBufferRGBA(pix.width, pix.height, pix.samples)
ValueError: Invalid data buffer size.

FIX

The issue is that PyMuPDF used to produce RGBA values, but now produces RGB values, thus the code as it stands works on older installations, but fails on newer installations. In particular, Bitmap.FromBuffer should be used instead of Bitmap.FromBufferRGBA.

This is a patch that will correctly handle both the older and newer versions of PyMuPDF.

diff --git a/wx/lib/pdfviewer/viewer.py b/wx/lib/pdfviewer/viewer.py
index 7b75724b..284069e1 100644
--- a/wx/lib/pdfviewer/viewer.py
+++ b/wx/lib/pdfviewer/viewer.py
@@ -516,7 +516,11 @@ class mupdfProcessor(object):
         matrix = fitz.Matrix(scale, scale)
         try:
             pix = page.getPixmap(matrix=matrix)   # MUST be keyword arg(s)
-            bmp = wx.Bitmap.FromBufferRGBA(pix.width, pix.height, pix.samples)
+            if [int(v) for v in fitz.version[1].split('.')] >= [1,15,0]:
+                # See https://github.com/wxWidgets/Phoenix/issues/1350
+                bmp = wx.Bitmap.FromBuffer(pix.width, pix.height, pix.samples)
+            else:
+                bmp = wx.Bitmap.FromBufferRGBA(pix.width, pix.height, pix.samples)
             gc.DrawBitmap(bmp, 0, 0, pix.width, pix.height)
             self.zoom_error = False
         except (RuntimeError, MemoryError):

[Derriick]

I confirm that this patch works perfectly for me.
@jmoraleda maybe if you made a pull request this change would be accepted? (I can do it, but it would be a bit of a theft of idea)

[jmoraleda]

@Metallicow Thank you. Feel free to remove the comment in the patch if you decide to apply it. I feel it is too small a contribution for me to make a pull request.

[jmoraleda]

I have found that the change in default behavior of getPixmap in MuPDF with respect to whether getPixmap returns an image with or without an alpha channel did not happen exactly between 1.14 and 1.15 but in between versions, so my previous patch (#1516) to the pdfviewer will not work for some versions of MuPDF.

I think an altogether cleaner solution to this problem that is also future proof against future changes in MuPDF default behavior is to not rely on the default and instead explicitly indicate in the call to getPixmap whether an alpha channel should be returned.

I just submitting a patch to fix this issue.