Open SylvainCorlay opened 1 year ago
What would be a good test to see if that setting is working?
My understanding is that this will take requirements from build
and host
and make them run
requirements.
The goal is for people linking with Julia's C API (having julia as a host dependency) to automatically inherit it as a run dependency pinning down the requirement to what will be binary compatible with what was used at build time.
A regular run_exports
section in the Julia feedstock adding julia as a run_export of itself would do that.
(Strong run exports are typically for compilers and is not relevant here).
@ngam do you have any thoughts here? Otherwise, I'll draft the pull request.
That’s a good idea.
On a side note (especially relevant later when we have more of a julia ecosystem): we should add a julia global pin with x.x as well
I can also help with the PR. There are many examples in conda-forge we could follow
(Strong run exports are typically for compilers and is not relevant here).
Technically, Julia is a compiler. It becomes more like a traditional compiler in version 1.9 in that it is creating shared libraries, .so, files per package. It may be to the degree that we may want to compile those shared libraries for the user and distribute them.
Comment:
The julia package should probably have a
run_exports
item for itself, pinningx.x
.