Accessing MATLAB's unicode strings from C

Question

How can I access the underlying unicode data of MATLAB strings through the MATLAB Engine or MEX C interfaces?

Here's an example. Let's put unicode characters in a UTF-8 encoded file test.txt, then read it as

fid=fopen('test.txt','r','l','UTF-8');
s=fscanf(fid, '%s')

in MATLAB.

Now if I first do feature('DefaultCharacterSet', 'UTF-8'), then from C engEvalString(ep, "s"), then as output I get back the text from the file as UTF-8. This proves that MATLAB stores it as unicode internally. However if I do mxArrayToString(engGetVariable(ep, "s")), I get what unicode2native(s, 'Latin-1') would give me in MATLAB: all non-Latin-1 characters replaced by character code 26. What I need is getting access to the underlying unicode data as a C string in any unicode format (UTF-8, UTF-16, etc.), and preserving the non-Latin-1 characters. Is this possible?

My platform is OS X, MATLAB R2012b.

Addendum: The documentation explicitly states that "[mxArrayToString()] supports multibyte encoded characters", yet it still gives me only a Latin-1 approximation to the original data.

Answer 1

First, let me share a few references I found online:

• According to mxChar description,

  MATLAB stores characters as 2-byte Unicode characters on machines with multi-byte character sets

  Still the term MBCS is somewhat ambiguous to me, I think they meant UTF-16 in this context (although I'm not sure about surrogate pairs, which probably makes it UCS-2 instead).

  UPDATE: MathWorks changed the wording to:

  MATLAB uses 16-bit unsigned integer character encoding for Unicode characters.

• The mxArrayToString page states that it does handle multibyte encoded characters (unlinke mxGetString which only handles single-byte encoding schemes). Unfortunately, no example on how to do this.

• Finally, here is a thread on the MATLAB newsgroup which mentions a couple of undocumented function that are related to this (you can find those yourself by loading the libmx.dll library into a tool like Dependency Walker on Windows).

---

Here's a small experiment I did in MEX:

my_func.c

#include "mex.h"

void mexFunction(int nlhs, mxArray *plhs[], int nrhs, const mxArray *prhs[])
{
    char str_ascii[] = {0x41, 0x6D, 0x72, 0x6F, 0x00};   // {'A','m','r','o',0}
    char str_utf8[] = {
        0x41,                   // U+0041
        0xC3, 0x80,             // U+00C0
        0xE6, 0xB0, 0xB4,       // U+6C34
        0x00
    };
    char str_utf16_le[] = {
        0x41, 0x00,             // U+0041
        0xC0, 0x00,             // U+00C0
        0x34, 0x6C,             // U+6C34
        0x00, 0x00
    };

    plhs[0] = mxCreateString(str_ascii);
    plhs[1] = mxCreateString_UTF8(str_utf8);        // undocumented!
    plhs[2] = mxCreateString_UTF16(str_utf16_le);   // undocumented!
}

I create three strings in C code encoded with ASCII, UTF-8, and UTF-16LE respectively. I then pass them to MATLAB using the mxCreateString MEX function (and other undocumented versions of it).

I got the byte sequences by consulting Fileformat.info website: A (U+0041), À (U+00C0), and 水 (U+6C34).

Let's test the above function inside MATLAB:

%# call the MEX function
[str_ascii, str_utf8, str_utf16_le] = my_func()

%# MATLAB exposes the two strings in a decoded form (Unicode code points)
double(str_utf8)       %# decimal form: [65, 192, 27700]
assert(isequal(str_utf8, str_utf16_le))

%# convert them to bytes (in HEX)
b1 = unicode2native(str_utf8, 'UTF-8')
b2 = unicode2native(str_utf16_le, 'UTF-16')
cellstr(dec2hex(b1))'  %# {'41','C3','80','E6','B0','B4'}
cellstr(dec2hex(b2))'  %# {'FF','FE','41','00','C0','00','34','6C'}
                       %# (note that first two bytes are BOM markers)

%# show string
view_unicode_string(str_utf8)

I am making use of the embedded Java capability to view the strings:

function view_unicode_string(str)
    %# create Swing JLabel
    jlabel = javaObjectEDT('javax.swing.JLabel', str);
    font = java.awt.Font('Arial Unicode MS', java.awt.Font.PLAIN, 72);
    jlabel.setFont(font);
    jlabel.setHorizontalAlignment(javax.swing.SwingConstants.CENTER);

    %# place Java component inside a MATLAB figure
    hfig = figure('Menubar','none');
    [~,jlabelHG] = javacomponent(jlabel, [], hfig);
    set(jlabelHG, 'Units','normalized', 'Position',[0 0 1 1])
end

---

Now let's work in the reverse direction (accepting a string from MATLAB into C):

my_func_reverse.c

#include "mex.h"

void print_hex(const unsigned char* s, size_t len)
{
    size_t i;
    for(i=0; i<len; ++i) {
        mexPrintf("0x%02X ", s[i] & 0xFF);
    }
    mexPrintf("0x00\n");
}

void mexFunction(int nlhs, mxArray *plhs[], int nrhs, const mxArray *prhs[])
{
    char *str;
    if (nrhs<1 || !mxIsChar(prhs[0])) {
        mexErrMsgIdAndTxt("mex:error", "Expecting a string");
    }
    str = mxArrayToString_UTF8(prhs[0]); // get UTF-8 encoded string from Unicode
    print_hex(str, strlen(str));         // print bytes
    plhs[0] = mxCreateString_UTF8(str);  // create Unicode string from UTF-8
    mxFree(str);
}

And we test this from inside MATLAB:

>> s = char(hex2dec(['0041';'00C0';'6C34'])');   %# "\u0041\u00C0\u6C34"
>> ss = my_func_reverse(s);
0x41 0xC3 0x80 0xE6 0xB0 0xB4 0x00               %# UTF-8 encoding
>> assert(isequal(s,ss))

---

Finally I should say that if for some reason you are still having problems, the easiest thing would be to convert the non-ASCII strings to uint8 datatype before passing this from MATLAB to your engine program.

So inside the MATLAB process do:

%# read contents of a UTF-8 file
fid = fopen('test.txt', 'rb', 'native', 'UTF-8');
str = fread(fid, '*char')';
fclose(fid);
str_bytes = unicode2native(str,'UTF-8');  %# convert to bytes

%# or simply read the file contents as bytes to begin with
%fid = fopen('test.txt', 'rb');
%str_bytes = fread(fid, '*uint8')';
%fclose(fid);

and access the variable using the Engine API as:

mxArray *arr = engGetVariable(ep, "str_bytes");
uint8_T *bytes = (uint8_T*) mxGetData(arr);
// now you decode this utf-8 string on your end ...

---

All tests were done on WinXP running R2012b with the default charset:

>> feature('DefaultCharacterSet')
ans =
windows-1252

Hope this helps..

---

EDIT:

In MATLAB R2014a, many undocumented C functions were removed from libmx library (including the ones used above), and replaced with equivalent C++ functions exposed under the namespace matrix::detail::noninlined::mx_array_api.

It should be easy to adjust the examples above (as explained here) to run on the latest R2014a version.