I am usually reluctant when upgrading, it's always painful. But I decided to make the jump from R2010b to R2013a on XP 32bits. What a slap... I upgraded only 24h ago. I spent the whole night and day passing a bunch of non-regression test. Is it my PC or the QA ? Regression is the word.
First, I have the feeling to work in slow motion : startup, IDE. It takes a looong time to start, on the same machine with no network connection, no antivirus, no nothing, compared to R2010b (I still have R2010b installed). Once calculating, it's OK, it's just the initialization and the IDE.
Talking of IDE, the look and feel 'a la Eclipse', as well as the graphical toolbar are - to me - unnecessary, and even counterproductive at first glance. Roll, little mouse, roll. Google does it (apparently) simple, that's part of the success.
So far, I tried to play with the huge collection of data produced by previous versions. First surprise, with the R2013a import button, because of the import popup window I can no longer open mat-files with custom extension from the IDE - I need to do it with a command line (I found no way to force a mat-file format for a .abc file, or I missed something ?). Unefficient, it was simply working fine up to R2010b.
Talking of import data/save workspace buttons in the IDE, sometimes they are not responding, it seems the functionality is dead in the toolbar (no sign of life, no java exception messages in the command line panel). And if (when) it works, sometimes a popup in the new import window say something like : "the following variables were imported" etc..., and nothing appears in the workspace... Until a good old .mat file is imported with the good old import wizard : then the content of the good old .mat file plus everything that has been silently discarded previously appears in the workspace. It smells funny. Is there an external dependency on something like a java version?
My code use matlab engines (written before DCT...) managed by MATLAB app through custom JAVA agents to perform single node cluster computing. The new thing is, because of some (*acceptable*) API changes in matlab functions (fileparts etc...), error were raised in mfiles run by engines . In those cases, engines are closed, and sometimes the handle of the faulty m-file seems not to be properly closed by the engine (?), so I can't fix the m-file (forbidden save) unless I reboot the station. not fun at all. Still, everything was OK in 2010b.
I didn't try other functionalities, but to me it will be hard to rely on such unstable environment to work with confidence.
I'm afraid I will find with this release other little but pretty annoying obstacles in the future without finding real improvements in the way I use MATLAB.
I'll try an earlier version tomorrow, the one before the introduction of the new interface, or I will keep R2010b.
But, regarding my future needs, I'm worried about the way your product is turning... and I will wait a while before buying DCT for my coming GPU plans.
ps : graphic rendering (and anti-aliasing!) is still unchanged, apparently. I exchange 100 IDEs like this one against one smooth rendering of 3D objects without sharp pixels !