pyat: Matlab test failures about at HOT 6 CLOSED

Of the 12 errors that we get 10 can be resolved by reverting commit f8e44f4.
In atmat/atphysics/ParameterSummaryFunctions/atlinopt.m:

-  JUMPS = diff([0;P]) < -1.e-3;
+  JUMPS = [0; diff(P)] < -1.e-3;

from at.

I am using Matlab 2016a or Matlab 2018a on MacOS 10.14. I created a new working directory, I tried running on several lattices, with various refpoints, I could not reproduce you problem… Finally I tried Matlab 2016a on debian8, and now I got it ! This looks very strange, I'm looking at that right now !

from at.

It turns out that matlab behaves differently with empty matrices. On macOS it can add empty matrices with different sizes ( 0X1, 0x0, 1x0 for instance) while it fails on linux. I pushed a fix on #82. Can you test it?

from at.

Interesting, I would not have expected such a difference between the same Matlab version on different platforms. #82 fixes all 12 of the failed tests, and everything behaves as expected now.

from at.

Problem solved with #82

from at.

Just to note that pyat is now helping to catch potential Matlab AT bugs. I think this is a good thing!

from at.