Bug in matlab's builtin sort routine

Matlab's 2012b has introduced all sorts of bugs in its sort routine. Consider the following

syms d1_X1_f1 d1_X1_f2 d1_X2_f1 d1_X2_f2 d1_b1_f1 real
A = sort([d1_X1_f1, d1_X1_f2, d1_X2_f1, d1_X2_f2, d1_b1_f1 ])
A'

In R2011a, this is correctly sorted as

   d1_X1_f1
   d1_X1_f2
   d1_X2_f1
   d1_X2_f2
   d1_b1_f1

But in R2012b, we get

 d1_X1_f1
 d1_X1_f2
 d1_X2_f1
 d1_b1_f1
 d1_X2_f2

There is at least one other bug that I've discovered, which I've been able to work around, but this one is really nasty. Does anybody have out there a sort algorithm that works properly? Since sort is a builtin, I can't just pull in the file from my 2011a distribution.

Any advice would be appreciated!

[Andreas Goser]

Did you report the bugs to MathWorks? It is important you do that even if you have a workaround yourself. I too often dealt with users being angry "why didn't you fix this bug in the last years!?!?" and they never reported the bug. Believe it ot not, but the majority of bugs is just found by one or two users - and that with far more than a million users!

[Leo Simon]

I did report this, and a related one due to symbolic sorting. Hopefully they'll do something about it.

[Andreas Goser]

Hope is not a strategy ;-) I researched and the bugs are fixed. I will contact you by email for additional information.

Answer 1

[Walter Roberson]

If you use

T = [d1_X1_f1, d1_X1_f2, d1_X2_f1, d1_X2_f2, d1_b1_f1 ];
which sort(T)

I suspect you will find that a symbolic toolbox sort.m is used (but that file might call into MuPAD)

Answer 2

[Leo Simon]

I guess I've found my own workaround, which is incredibly unsatisfactory, but does seem to work. Convert the symbolic array to a cell array of strings, then sort, then convert back to syms. Uggh, I feel ill just typing this....

[Andreas Goser]

Worked with the customer offline. If another customer runs into this and calls Technical Support, refer to "G906530"