Do not rehost third party docs of non-SciML packages #239
Conversation
|
I do like the work towards not embedding then, but I would actually prefer if there would then maybe be a page “related projects” or something like that, where these are listed instead. Maybe even in 2 groups
|
| @@ -82,9 +51,6 @@ docsmodules = [ | |||
| "SurrogatesBase", | |||
| "CommonSolve", | |||
| ], | |||
| "Third-Party Interfaces" => ["ArrayInterface", "StaticArrayInterface", "AbstractFFTs", | |||
There was a problem hiding this comment.
why ArrayInterface and StaticArrayInterface? Those are only built here.
| "Symbolics" => "https://github.com/JuliaSymbolics/Symbolics.jl", | ||
| "SymbolicUtils" => "https://github.com/JuliaSymbolics/SymbolicUtils.jl", | ||
| "TermInterface" => "https://github.com/JuliaSymbolics/TermInterface.jl", |
There was a problem hiding this comment.
This will break the docs build.
The docs are updated every 2 hours. Should we change that to 1? |
They exist. You can for example find the similar information in https://docs.sciml.ai/Overview/stable/highlevels/partial_differential_equation_solvers/. Given that you have been advocating for this and didn't notice that they were added 2 years ago, that's pretty good evidence that just links in pages is not discoverable. Which again is exactly what the interviews had already determined. |
|
We can merge something here, but this PR is very overreaching and seems to just be emotional. Why break the Symbolics, TermInterface, SymbolicUtils, ArrayInterface, and StaticArraysInterface documentation as part of this? SparseDiffTools, FiniteDiff. This needs to be done with a finer toothed comb as right now this is breaking to packages which are setup to have this as their only docs. |
Sorry I mainly checked for actually ManifoldDiffEq, which is not listed on that page, so I did not find that page. |
|
Ah, not sure how I missed that one (though there is a small typo as well, Manopt does not have a capital O). Oh even more: Why is Manopt listed then there but not also completely reposted / listed in the menu? So there seems to be a certain choice it seems. Well. Long story short, I would really prefer if these docs would not rehost documentation. |
Documentation of e.g. Symbolics, TermInterface, and ArrayInterface are all hosted outside of SciML: https://symbolics.juliasymbolics.org/dev/ https://juliasymbolics.github.io/TermInterface.jl/dev/ https://juliaarrays.github.io/ArrayInterface.jl/dev/ |
|
Oh that's an issue, we can fix that |
|
Superseded by #240 |
Fixes #151.
The PR removes the rehosted documentation of non-SciML packages. Currently, the SciML rehosts a huge number of external docs. If the SciML docs want to refer to them, they should link to these but not copy them, even more so since I assume typically this happens without the consent of these third party packages. The external packages already host their docs but it seems in some cases Google already ranks the SciML docs higher (even though the SciML copy is generally outdated since updates of the third party docs are not automatically pushed to the SciML docs). In some cases (e.g. Lux) that do not use the standard Documenter layout, the rehosted copy is also completely broken: https://docs.sciml.ai/Lux/stable/
IMO the fact that MultiDocumenter has no support for linking to other docs in the top bar is no reason to keep rehosting the third party docs. The current behaviour is just plain wrong, it violates the preference and possibly even license of external packages and is done without consent.