NVDA pronounces the phrase "temp name" when matrix is raised to a power

[mathmandan]

Consider the following example (MathML code):

<math display="block" xmlns="http://www.w3.org/1998/Math/MathML">
 <msup>
 <mrow>
 <mo fence="true" lspace="0" rspace="0" symmetric="true">(</mo>
 <mspace width="-4.981pt"/>
 <mrow>
 <mspace width="4.981pt"/>
 <mtable>
 <mtr>
 <mtd>
 <mn>1</mn>
 </mtd>
 <mtd>
 <mn>2</mn>
 </mtd>
 </mtr>
 <mtr>
 <mtd>
 <mn>3</mn>
 </mtd>
 <mtd>
 <mn>4</mn>
 </mtd>
 </mtr>
 </mtable>
 <mspace width="4.981pt"/>
 </mrow>
 <mspace width="-4.981pt"/>
 <mo fence="true" lspace="0" rspace="0" symmetric="true">)</mo>
 </mrow>
 <mn>2</mn>
 </msup>
</math>

Pasting this code into the "Math Input Area" at https://daisy.github.io/MathCATDemo/ yields the following output:

TEMP NAME of  the 2 by 2 matrix 
<break time='800ms'/>row 1 <break time='400ms'/> 
<mark name='M6qsyqqx-7'/> 1 <break time='200ms'/> <mark name='M6qsyqqx-9'/> 2 <break time='400ms'/>
row 2 <break time='400ms'/> 
<mark name='M6qsyqqx-12'/> 3 <break time='200ms'/> <mark name='M6qsyqqx-14'/> 4 <break time='800ms'/> 
end matrix <mark name='M6qsyqqx-16'/> 
squared

Or more simply:

TEMP NAME of the 2 by 2 matrix
row 1: 1, 2
row 2: 3, 4
end matrix
squared

I am not sure why the phrase "TEMP NAME of" is being included? Otherwise this reading seems very good.

FYI, the MathML shown above was generated using LuaLaTeX, and this issue was previously mentioned at https://tex.stackexchange.com/q/766321/ . The original LaTeX source code:

\DocumentMetadata{%uncompress, 
  language = en-US, % 
  pdfversion = 2.0, %
  pdfstandard = {ua-2, A-4f}, %
  tagging = on, %
  tagging-setup = {math/setup=mathml-SE, extra-modules={verbatim-mo}} %
}

\documentclass{article}

%\usepackage{mathtools} % optional

\begin{document}

\[
{%
  \begin{pmatrix}
    1 & 2 \\
    3 & 4
  \end{pmatrix}%
}^{2} 
\]

\end{document}

[wittjeff]

I can reproduce this with cargo test on main, and the mspace/fence attributes from LuaLaTeX turn out to be a red herring — a plain <msup><mrow><mo>(</mo><mtable>…</mtable><mo>)</mo></mrow><mn>2</mn></msup> reads the same way:

TEMP NAME of the 2 by 2 matrix; row 1; 1, 2; row 2; 3, 4; squared

Canonicalization is fine in both cases (the spacing folds away to a ( mtable ) mrow). The intent tree is what goes wrong:

power( TEMP_NAME( matrix(…) ), 2 )

TEMP_NAME is the internal transport wrapper that replace_nodes_tree puts around a bare x: replacement. The mrow matrix intent rule (Rules/Intent/general.yaml) replaces with x: "*[2]" rather than an intent:, so the mrow's result is TEMP_NAME(matrix); the power rule's x: "*[1]" wraps that once more, and lift_children in src/speech.rs only strips one level. The leftover TEMP_NAME element then hits the generic function-intent speech rule, which speaks its name: "TEMP NAME of …".

Fix: make lift_children flatten nested wrappers (a wrapper whose children are all elements is spliced recursively; a leaf wrapper holding text is still only unwrapped as a direct child, so that path is unchanged). With that, both the plain and the LuaLaTeX MathML read

the 2 by 2 matrix; row 1; 1, 2; row 2; 3, 4; squared

The full suite passes (cargo test: 1982 + 5106 tests, 0 failures) and the patch adds two regression tests to tests/Languages/en/mtable.rs. I suspect this is also what the commented-out Greek corpus cases mentioning "TEMP NAME του … πίνακας" were hitting (matrix inverse / matrix powers).

Fix + regression tests are in #763.