On Wed, Feb 19, 2014 at 9:33 PM, Greg Von Kuster <greg@bx.psu.edu> wrote:
Hi Peter,
Dave B and I just discovered that issue that causes your tests to fail.
The problem lies with our current implementation supporting the <move_directory_files> action tag, which is used in your recipe. This tag ultimately uses Python's os.listdir via shutil.move.
It turns out that certain Python versions produce different results from os.listdir - the order is different. This is a problem with MIRA because during the instlllation it is moving a directory of files where the directory includes symlinks to other files in the directory. When the symlinks are moved after the files to which they link, problems occur. Again, this behavior is intermittent depending on the Pyhton version ( and perhaps the os ).
In any case, we have a fix for out move_directory_files which Dave is now committing. So your tests should pass with tonight's run - Let's hope!
Thanks Peter!
Well done Greg & Dave, That explanation makes perfect sense to me - but I can see it took some digging to solve it! There is actually one MIRA binary (mira) and several aliases (e.g. mirabait) which are really symlinks back to mira. Magic. My next query would be: why wasn't the failing action <move_directory_files> being caught as a failed install? Thanks, Peter