On Thu, Oct 14, 2021, Stephan Schott wrote:
>I had this issue some time ago, so I gave it a try now to see if I can
>replicate it. I also had trouble with pip, and it seems that the conda
>update (line 152 of UseMiniconda.cmake) causes pip to be removed (at least
>under Linux). Then, as pip is never explicitly requested, it is never
>installed again, and makes pip to be missing. Now, the issue there is that
>if cmake fails to update the shebang of amber.pip, it apparently also fails
>to update the shebang of other python scripts. Can someone else try to
>replicate this independently? I think forcing pip to be installed could be
>a quick solution, but maybe someone has an idea why the conda update causes
>the uninstall on the first place.
I still can't seem to reproduce this here. On my ubuntu 20.04 system, there
is no $AMBERHOME/miniconda/bin/pip, so the symlink to $AMBERHOME/bin is
never made, and the FixShebang script doesn't try to do anything with the
(missing) $AMBERHOME/bin/amber.pip file. But the other amber.xxx python
scripts all seem to be correctly fixed.
The behavior in the paragraph above is what the code looks like it should be
doing, and what was intended. One thing to try would be to uncomment line
36 in cmake/FixCondaShebang/FixCondaShebang.cmake. This will indicate what
order all these fixes are done in, which might somehow not be be that same
on all machines. Or, if amber.pip is present in the list, try to figure
out why that is. One could also add a check on the existence of ${FILE}
inside the loop starting at line 42, so that configuretime_file_replace()
won't try to work on a non-existent file, and hence won't fail.
Since you are able to get things to fail, you might be able to track down the
problem.
But my further question is this: does any workflow really make use of
$AMBERHOME/bin/{amber.conda,amber.ipython,amber.jupyter}? I have a feeling
that the answer might be "no", and we could just skip this whole process
entirely.
...thanks...dac
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Thu Oct 14 2021 - 09:00:02 PDT
