Re: No Explicit Python Version Dependency In Package Definitions?

[Ricardo Wurmus]

Hi Phil,

> So I have a system say with python3.8 and python-scipy, and I decide I
> want to then install python-pandas (for example).  Will it not then
> build this for python 3.9 (due to the new definition), if the version of
> python has incremented between the installs of python-scipy and
> python-pandas?  Would I then have to manually reinstall python-scipy to
> have it under 3.9 (as well as 3.8) (or do a guix upgrade)?

Yes.  It’s important to upgrade your Python packages together to avoid
weirdness.  I suggest using a manifest.

> Last question, if today I wanted to create a profile that installed
> python-scipy against the python 3.9 package definition (which already
> exists in Guix it's just not the default).  Do I have to change the
> python-3 definition as per above, or is there another way of saying "use
> python3.9 just for this profile".
>
> I suspect I could use a manifest to install 3.9 rather than the default,
> but won't any packages still depend on 3.8 unless I switch the python-3
> definition?

Inside a manifest you can use the Guix package transformation API to
replace all instances of one Python variant with another — recursively
throughout the whole package graph.  You’ll have to compile all those
packages locally, though, because these package variants haven’t been
built by the build farm.  Luckily Guix does this for you.  It just takes
time (and electricity).

-- 
Ricardo