Trouble with patchcmd and double #

Question

I need to patch the following macro from gloss-greek.ldf (polyglossia):

\def\greek@numbers{%
   \let\latin@alph\@alph%
   \let\latin@Alph\@Alph%
   \if@greek@numerals
      \def\greek@alph##1{\protect\greeknumber{##1}}%
      \def\greek@Alph##1{\protect\Greeknumber{##1}}%
      \let\@alph\greek@alph%
      \let\@Alph\greek@Alph%
   \fi}

to the following

\def\greek@numbers{%
   \let\latin@alph\@alph%
   \let\latin@Alph\@Alph%
   \if@greek@numerals
      \def\greek@alph##1{\protect\greeknumber{\the##1}}%
      \def\greek@Alph##1{\protect\Greeknumber{\the##1}}%
      \let\@alph\greek@alph%
      \let\@Alph\greek@Alph%
   \fi}

(to add the two \thes).

However I get problems related to different catcodes of #.

I've tried approaches like the ones mentioned in Patching arguments inside a macro but I have failed so far.

Working example:

\documentclass{memoir}

\usepackage{amsmath}
\usepackage{fontspec}
\usepackage{xpatch}

\usepackage{polyglossia}
\setdefaultlanguage{greek}
\setotherlanguage{english}
\newfontfamily\greekfont{Times New Roman}

\usepackage{xpatch}
\tracingpatches
\makeatletter
\newcommand{\patchme}{%
\xpatchcmd{\greek@numbers}{\greeknumber{##1}}{\greeknumber{\the##1}}{}{} 
\xpatchcmd{\greek@numbers}{\Greeknumber{##1}}{\Greeknumber{\the##1}}{}{}
}
\makeatother


\begin{document}
\patchme
\begin{subequations}
\begin{align}
a &= b \label{eq:eq1}\\
c &= d \label{eq:eq2}
\end{align}
\end{subequations}

This is \eqref{eq:eq2}? Is it really?

\end{document}

If the patch were applied the reference would appear correct. Also the log reads:

[debug] tracing \patchcmd on input line 28
[debug] analyzing '\greek@numbers'
[debug] ++ control sequence is defined
[debug] ++ control sequence is a macro
[debug] -- nested patching command and parameters in patch
[debug] -> the patching command seems to be nested in the
[debug] argument to some other command
[debug] -> the patch text seems to contain # characters
[debug] -> either avoid nesting or use # characters with
[debug] category code 12 in the patch text
[debug] -> simply doubling the # characters will not work

Answer 1

Patching commands with ## with etoolbox (which xpatch uses internally) is not possible, as far as I can see.

I'm not sure why you need \the in that place, because the output is the same with or without it. Anyway, I'd prefer \number.

It is possible to do the patch with regexpatch. The patches can be done all at once, noting we want to replace every occurrence of {##1} with {\number##1}, so we can use \xpatchcmd*.

\documentclass{memoir}

\usepackage{amsmath}
\usepackage{fontspec}
\usepackage{regexpatch}

\usepackage{polyglossia}
\setdefaultlanguage{greek}
\setotherlanguage{english}
\newfontfamily\greekfont{Times New Roman}

\makeatletter
\xpatchcmd*{\greek@numbers}
  {{##1}}
  {{\number##1}}
  {\typeout{Hurray!}}{\typeout{Rats!}}
\makeatother

\begin{document}

\begin{subequations}
\begin{align}
a &= b \label{eq:eq1}\\
c &= d \label{eq:eq2}
\end{align}
\end{subequations}

This is \eqref{eq:eq2}? Is it really?

\end{document}

Here's the relevant part of the terminal output:

(/usr/local/texlive/2015/texmf-dist/tex/latex/polyglossia/gloss-english.ldf)
Hurray!
(./doublehash.aux) (/usr/local/texlive/2015/texmf-dist/tex/latex/tipa/t3cmr.fd)
[1] (./doublehash.aux) )
Output written on doublehash.pdf (1 page).

Answer 2

If the problem is about how to fix the MWE, a fix can be had by avoiding the use of xpatch altogether, and merely respecifying, in total, the desired definition of \greek@numbers in the preamble.

\documentclass{memoir}

\usepackage{amsmath}
\usepackage{fontspec}

\usepackage{polyglossia}
\setdefaultlanguage{greek}
\setotherlanguage{english}
\newfontfamily\greekfont{Times New Roman}

\makeatletter
\def\greek@numbers{%
   \let\latin@alph\@alph%
   \let\latin@Alph\@Alph%
   \if@greek@numerals
      \def\greek@alph##1{\protect\greeknumber{\the##1}}%
      \def\greek@Alph##1{\protect\Greeknumber{\the##1}}%
      \let\@alph\greek@alph%
      \let\@Alph\greek@Alph%
   \fi}
\makeatother


\begin{document}
\begin{subequations}
\begin{align}
a &= b \label{eq:eq1}\\
c &= d \label{eq:eq2}
\end{align}
\end{subequations}

This is \eqref{eq:eq2}? Is it really?

\end{document}

If however, the real purpose of the OP's question is to understand the nature of why \xpatchcmd doesn't like ##1 in its arguments, then this answer sheds no light.

A simpler MWE to demonstrate the problem of ##1 in an \xpatchcmd macro is as follows. If the patch worked, the result would be Q(C), not B(C).

\documentclass{article}
\usepackage{xpatch}
\def\A{\def\B##1{B({##1})}}
\xpatchcmd{\A}{B({##1})}{Q({##1})}{\xdef\result{successful}}{\xdef\result{unsuccessful}} 
\A
\begin{document}
\B{C}.  Patch was \result.
\end{document}