There was a thread about _mfile_encoding_.  I didn't follow the whole thing but I think the gist was that it was something that would come from the host system.

There is at least one case where we know for sure: if the texinfo in the file specifies `@documentencoding UTF-8`.

Thus we should override `__mfile_encoding__` on a file-by-file basis if there is an explicit `@documentencoding` tag.

In Symbolic, the texinfo is full of UTF-8 unicode chars.  This is starting to work on Windows, as is the unicode-rendered math.  Funnily enough, one thing that doesn't work is the "1×2 matrix" in:

>> [sin(a/2) pi]
ans = (sym 1×2 matrix)

  ⎡   ⎛a⎞   ⎤
  ⎢sin⎜─⎟  π⎥
  ⎣   ⎝2⎠   ⎦

(On Windows, one gets something like "$\tilde{A}--$".
This is because this string is directly written in the @sym/display.m file:

 function [s1 s2] = sym_describe(x, unicode_dec)
   if (unicode_dec)
-    timesstr = '×';

And is thus interpreted by whatever lies _mfile_encoding_ chooses today/on this OS/with this moon phase.  But this is crazy because we just pulled utf-8 texinfo from the top of the very same file...

I plan to workaround this by not having any non-ascii chars in my code:
https://github.com/cbm755/octsympy/issues/941
Although this still leaves `!%` BISTs.