Reason why someone would activate the option "OverrideDoubleBasicArrays" is to work with the desired precision (superior to that offered by type "double").

On the other hand, in "classic" Matlab, no one uses the builtin function "double" with explicit intentionality to reduce the precision for data (most often on the contrary).

In order to shoot not themselves in the foot I suggest altering the method "double" like this:

if ~(mp.OverrideDoubleBasicArrays && ismp(x))
r = mpimpl(402,x); else r = x; end

the intuition tells me that in the very near future this tool will be very useful; It seems that Pavel prepares a lot of updates (very pleasing, I think)

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
function advanpix_update()
% check for a newer version than the locally installed,
% download the new version and then launches the installer


v = sscanf (version, '%d.%d.%d') ; % /matlabcentral/fileexchange/17285-getversion
v = 10.^(0:-1:-(length(v)-1))*v ;
if ~(v>=8.4), disp(' ... this implementation requires Matlab R2014b or newer.')
return; end

try
text = webread('http://www.advanpix.com/download/','Timeout',5);
% get version available for download
newver = strtrim(strtok(regexprep(text,'.+Download Trial.+Ver.',''),'<'));
% get local installed version
info = evalc('mp.Info');
text = regexp(info,'Version.*','ignorecase','dotexceptnewline','match');
curver = regexprep(strtrim(regexprep(char(text),'[^0-9.\s]','')),'\s+','.');
if ~strcmp(newver,curver)
instdir = winqueryreg('HKEY_CURRENT_USER','Software\Advanpix','InstallDir');
if exist(fullfile(instdir,'download'),'dir')~=7
mkdir(fullfile(instdir,'download'));
end
Url = 'http://www.advanpix.com/wp-content/plugins/download-monitor/download.php?id=1';
exe_name = ['AdvanpixMCT-',newver,'-Setup.exe'];
disp('Try downloading the newest version of the toolbox. Wait, please ...');
% Check for supported Microsoft .NET Framework
if NET.isNETSupported
disp('Supported .NET Framework not found')
% create .Net object to handle the file download
NET.addAssembly('System.Net');
client = System.Net.WebClient;
installer = fullfile(instdir,'download',exe_name);
client.DownloadFile(Url,fullfile(instdir,'download',installer));
else
options = weboptions('Timeout',30,'ContentType','binary');
olddir = cd(fullfile(instdir,'download'));
try installer = websave(exe_name,Url,options);
catch, cd(olddir); end
cd(olddir);
end
% run the installer
system(installer);
else
disp('The latest version of the toolbox is already installed.');
end
catch ME
rethrow(ME)
end
end

I have a generalized eigenvalue problem Ax = lambda*Bx, where A and B are real multiprecision matrices. I used eig(A,B) and expected that non-real eigenvalues came out in perfectly complex conjugate pairs (because I want to group them in pairs), but this is not the case. I'm not sure this should be considered a bug, or rather a question about whether there is a way to enforce that EIG returns perfectly conjugate pairs when A and B are real mp matrices?

Here is a simple example (tested in MATLAB 2015A and MCT version 3.9.4.10443):

H = [ 2.285927555469414e-03 -9.683344516608315e-03 0 0 ;

current status:

>> a = mp(mprandn(2,1))
b = [0, 1]
a =
0.4334207802637199612271739466794224
-0.2701469591393686343297786869883695
b =
0 1

>> cast(a,'double')
ans =
0.43342
-0.27015
>> cast(a,'like',b)
ans =
0.43342
-0.27015
>> cast(b,'mp')
Error using cast
Unsupported data type for conversion: 'mp'.

>> cast(b,'like',a)
Error using cast
Unsupported data type for conversion: 'mp'.

>> cast(a,'ss')
Error using cast
Unsupported data type for conversion: 'ss'.

with a CAST method like below ...

function r = cast(A,newclass,varargin) %#ok<INUSL>
narginchk(2,3);
if strcmp(newclass,'like')
narginchk(3,3); newclass = class(varargin{1});
end
try r = eval([newclass,'(A)']);
catch, r = eval([newclass,'(double(A))']); end
end

... the new status:

>> cast(b,'mp')
Error using cast
Unsupported data type for conversion: 'mp'.

>> cast(b,'like',a)
ans =
0 1

>> cast(a,'ss')
ans =

d =
u1
y1 0.4334
y2 -0.2701

Static gain.

Of course, my version is incomplete, but enough for me :)

By the way, on what grounds 'mp' inherits not 'double' ?

from Version History:

...

[US,TS, Success] = ordschur(...) % three-outputs, with status as last one.

Note:

ORDQZ has the same peculiarity - see GCARE or GDARE code:

[HH,JJ,~,z(perm,:),Success] = ordqz(HH,JJ,q,z,'lhp') % five outputs

I am using an old version (more than 2 years ago) and can not compute the eigenvalues/vectors of asymmetric matrix. I do not know whether it can be computed in the new version. If not, please develop one. Appreciate it!