_get_mathMl fails to get MathML

[davidcarlisle]

Steps to reproduce:

With MathCAT add on installed, read any PDF file with tagged MathML in acrobat or Foxit,

Actual behavior:

With read-all the math is not read and an exception is shown in the log

Expected behavior:

The PDDom query should succeed and pass the MathML to MathCAT

NVDA logs, crash dumps and other attachments:

System configuration

NVDA installed/portable/running from source:

NVDA installed version

NVDA version:

Log file from 2025.2 although same originally seen with 2025.3 beta

Windows version:

11

Name and version of other software in use when reproducing the issue:

Foxit 2025.2.0.33046
MathCAT 0.6.10 installed from nvda (also occurs with 0.7.x installed directly from @NSoiffer 's github)

Other information about your system:

I think the problem started after a foxit update but persists after uninstalling Foxit and testing with acrobat, and after completely uninstalling NVDA and re-installing

(I say "think" as I updated foxit and also nvda to 2025.3 beta at the same time, and things broke, but stayed broken even after reverting both programs, so I can not be sure exactly which update causes this)

Other questions

Does the issue still occur after restarting your computer?

yes

Have you tried any other versions of NVDA? If so, please report their behaviors.

Yes originally observed with 2025.3beta but still now shows with 2025.2

If NVDA add-ons are disabled, is your problem still occurring?

Only MathCAT is installed, which is needed to see the issue.

Does the issue still occur after you run the COM Registration Fixing Tool in NVDA's tools menu?

I think so, running this is a bit tricky as I'm not aware of a commandline call.

Due to company policies I need to install nvda under Windows "elevated access" but run it and other applications as normal user.

This means (for example) to uninstall nvda (as I have done while testing this) I can't use the "uninstall" menu as it prompts for admin password, but I can run the nvda uninstall program from program files x86 under elevated access.

the com registration fix appeared to run if I start nvda under elevated access just to select that option.

It had no observable effect on behaviour.

Given that this behaviour wasn't observable on this machine previously but now is observable even after full uninstall and re-install, is there a registry setting that is being used here or some other global setting that is not cleared by uninstall?

It is observable with any pdf file that has math tagged with mathml, but a small example is attached

It reads

text and math

when it should read

text and math fraction x over y end fraction

(or similar, depending on mathcat settings)

test-af.pdf

nvda.log

basically any math produces a log that looks like

Speaking [LangChangeCommand ('en_US'), 'August 11, 2025', CallbackCommand(name=say-all:lineReached)]
DEBUGWARNING - NVDAObjects.IAccessible.adobeAcrobat.AcrobatNode.initOverlayClass (10:41:51.981) - MainThread (15156):
FAILED: QueryService(SID_GetPDDomNode, IGetPDDomNode)
DEBUGWARNING - NVDAObjects.IAccessible.adobeAcrobat.AcrobatNode.initOverlayClass (10:41:51.981) - MainThread (15156):
FAILED: get_PDDomNode
DEBUGWARNING - NVDAObjects.IAccessible.adobeAcrobat.AcrobatNode._get_mathMl (10:41:51.981) - MainThread (15156):
_get_mathMl: self.pdDomNode is None!

So the PDDom query fails with both acrobat and foxit

[seanbudd]

cc @NSoiffer

[NSoiffer]

I also have this problem on my desktop computer for over a year. I have a python script that a Foxit developer gave me and it exhibits the same problem:
QueryService(SID_GetPDDomNode, IGetPDDomNode) fails with error (-2147024809, 'The parameter is incorrect.', (None, None, None, 0, None))

In the past, I spent days on trying to figuring out what has gone wrong with no success. @davidcarlisle's problems prompted me to tried again and write things down. Since my knowledge of COM has faded, I turned to AI for some help. The failure of the QueryService is happening because the object being queried doesn't know about SID_GetPDDomNode. AI helped me to understand that it is not looking in the registry for this, but it should be known to the object.

The full sequence of calls in the python script is:

    ia = msaa.AccessibleObjectFromWindow(window, msaa.OBJID_CLIENT)
    serv = ia.QueryInterface(IServiceProvider)

    try:
        pddomRoot = serv.QueryService(SID_GetPDDomNode, IGetPDDomNode)
    except COMError as ce:
        print(f"***serv.QueryService(SID_GetPDDomNode, IGetPDDomNode) fails with error {ce}")
        quit()

Note: the original code had print statements after each query and they all indicated success up to the QueryService call.

The object that is queried is AccessibleObjectFromWindows IServiceProvider interface. The calling code It should be coming from AcrobatAccess.dll according to ChatGPT. ChatGPT says this should be inC:\Program Files\Adobe\Acrobat [version]\Acrobat\AcrobatAccess.dll for a 64 bit system. I don't have that on either the working or non-working computer. So that is a bit of a dead end. If we can find the right dll, runing regsvr32 from an admin shell should set up some needed registry entries.

I reinstalled Acrobat Reader and something I did (running NVDA?) triggered it to ask if I wanted to set up accessibility. I went through the setup and now the python script is working for me, at least enough that it is finding information in the PDF. NVDA still isn't working, and unfortunately a reboot didn't help. I tried the registry fix tool and another reboot -- still no luck. Neither Acrobat nor Foxit will read the math

• an AF version of math: test-af.pdf for Foxit
• a structure element version of math for Acrobat: t1-se.pdf

HOWEVER: there is some progress. NVDA is no longer failing with the QueryService call:

Traceback (most recent call last):
  File "queueHandler.pyc", line 67, in flushQueue
  File "speech\sayAll.pyc", line 339, in nextLine
  File "speech\types.pyc", line 55, in _flattenNestedSequences
  File "speech\types.pyc", line 55, in <genexpr>
  File "speech\speech.pyc", line 1653, in getTextInfoSpeech
  File "speech\speech.pyc", line 1454, in _extendSpeechSequence_addMathForTextInfo
  File "virtualBuffers\__init__.pyc", line 515, in getMathMl
  File "baseObject.pyc", line 59, in __get__
  File "baseObject.pyc", line 167, in _getPropertyViaCache
  File "NVDAObjects\IAccessible\adobeAcrobat.pyc", line 200, in _get_mathMl
  File "comtypes\_post_coinit\_cominterface_meta_patcher.pyc", line 19, in __getattr__
AttributeError: GetAttribute

This is with 2025.3 RC1. Line 200 in _get_mathMl is the MS Office hack:

		math = self.pdDomNode.GetAttribute("MSFT_MathML", "MSFT_Office")

I'll try wrapping this inside of a "try" and see if that solves the problem.

[NSoiffer]

A took a while to get a build going on my machine, but indeed changing that line to:

		try:
			math = self.pdDomNode.GetAttribute("MSFT_MathML", "MSFT_Office")
			if math:
				return math
		except (AttributeError, COMError):
			log.debugWarning('_get_mathMl: self.pdDomNode.GetAttribute("MSFT_MathML", "MSFT_Office failed"')

solves the problem... at least for AF files in Foxit. For Acrobat, I'm still getting the QueryService failure. In the python program I have, QueryService works for both Foxit and Acrobat.

Adding the above guard doesn't explain why it is a problem on one machine but not on another though.

@seanbudd: what branch should I submit the fix to? I suspect it is too late for 2025.3, although this should be 100% safe because all that it is doing is wrapping the existing code inside a try/except. It is probably good practice to do this for any COM call.

[seanbudd]

@NSoiffer - please submit it to master

[seanbudd]

@NSoiffer - is this something you still plan on addressing?