New comment by tornaria on void-packages repository

https://github.com/void-linux/void-packages/pull/36209#issuecomment-1072771825

Comment:
> May I suggest that we split off a package like
> 
> ```
> sagemath-devel_package() { # or whatever you want to call it
>  depends = "sagemath"
>  depends += "move here all the -devel that is in sagemath depends, the compilers etc"
>  desc += "development stuff, needed to pass all doctests, compile cython code etc."
> ```
> 
> Do you see any drawbacks?

What you suggest seems to ill-defined or at least very difficult to determine. What are some example packages that are now listed as runtime dependencies but you would move to `sagemath-devel`? What is the gain (say in installed-size) vs. what is to loose (as in what would stop working). If you are worried about download or installed size, we could have some `sagemath-minimal` which leaves out some functionality, e.g. some of the large databases. But the difference in size has to be worth it.

On a different note I was thinking to have a meta-package `sagemath-builddeps` which depends on all packages that are needed to `build` sagemath from the official source release (i.e. using all possible packages from system). This would *not* depend on `sagemath`, so it's a user choice to install from source.

I'm also working on a `sagemath-beta` package that will install current beta of sagemath -- the catch is that I want this to be installable side-by-side with the main `sagemath` package so we can test new releases as they come out.

In fact, the current PR (update of lrcalc) is motivated by this. In order to have `sagemath` and `sagemath-beta` installed at the same time we need both to use the same dependencies. Since sagemath-9.6.beta5 works only with lrcalc-2.1, we need to have sagemath-9.5 work with lrcalc-2.1 as well.

There are annoying doctest failures triggered by updates to python packages -- I thought I had all covered for the updates to ipython 8.0 and scipy 1.8 but it seems ipython 8.1 has another change. This is the failure we are seeing in CI, it should be easy to fix and push upstream if it's not already fixed in 9.6.beta5.