Strange behavior of diff()

Hi all,

I'm using 2012b and I'm trying to find the difference between successive elements of a vector of uint64 numbers. Here's some of them, but I get very puzzling results form diff(). For instance:

a =       [4648759972
           4648759972
           4651621722
           4652352784
           4652733003
           4653926753]

Then diff(a) gives the wrong result:

               [5166781
                5166781
                2305031
                1573969
                1193750]

However diff(a(1:2)) gives the correct result: 0. Also diff(a(2:3)) gives the correct result: 2861750

I've been searching for an explanation but I did not find a similar problem ever reported, and I don't understand what is happening. Can anybody explain to me what is going on here?

Thanks a lot!

Paul

[dpb]

>> a = uint64([4648759972 4648759972 4651621722 4652352784 4652733003 4653926753]);
>> diff(a)
ans =
         0     2861750    731062      380219     1193750
>> ver
---------------------------------------------------------------------
MATLAB Version: 8.0.0.783 (R2012b)
...
>>

Try clear diff first, then retry...

Oh, and which diff before clear just for grins might be informative.

[Stephen23]

>> a = uint64([4648759972 4648759972 4651621722 4652352784 4652733003 4653926753])
a =
           4648759972           4648759972           4651621722           4652352784           4652733003             4653926753
>> diff(a)
ans =
                    0              2861750               731062               380219              1193750

[Paul]

Hi all, thanks for your quick responses.

which diff yields the standard Matlab path in /datafun/. Clearing diff had no effect. Adding the uint64 indicator for my variable, as Stephen suggested, sadly gives me the same results as before... Obviously I can code a quick and dirty workaround so I'm not in a tight spot but, it is very puzzling isn't it?

[edit] Ok, so, if I change the class of my variable into double, it works no problem, so diff(double(a)) works correctly.

I've also figured out what is exactly going wrong in the case of diff(uint64(a)). Instead of doing [X(2)-X(1) X(3)-X(2) .. X(N+1)-X(N)], it does [X(end)-X(end-1) X(end)-X(end-2) ... X(end)-X(1)]

[dpb]

As illustrated above I don't see that problem with R2012b here using either numeric class; double or uint64.

Submit the sample code that produces the error to TMW as a bug report at www.mathworks.com

[John BG]

tried

a = [4648759972 4648759972 4651621722 4652352784 4652733003 4653926753]

b=diff(a)

a_uint=uint64(a)

b_uint=diff(a)

all look fine in R2015a

Once had odd results with R2012a, reinstalled and it was fine. Have you tried to reinstall?

regards

John

[e-mail]

[Jan]

What does

which diff -all

reply for uint64?

[Walter Roberson]

And to ask more directly,

which diff(uint64(0))

that will ensure you get the correct function shown for whatever uint64 is resolving to for you.

[Paul]

Thanks Walter; an external toolbox rerouted the call to an external diff function with bad behavior. Deleting that function solved all problems. Sorry for the late reply.

Answer 1

[Paul]

Problem solved; external diff function caused bad behavior; deleting this external function solved the problem.