weakdeps don't load from fallback environment

[timholy]

Extension modules (weak dependencies) only seem to load if both packages are in the currently-active environment. The docs say

An extension will only be loaded if the extension dependencies are loaded from the same environment or environments higher in the environment stack than the package itself.

but I am not certain that applies here, since both packages are in the same (default) environment. Either the docs might benefit from clarification, or there may be a bug.

Demo:

julia> using PkgCacheInspector, MethodAnalysis

julia> methods(methodinstances, (PkgCacheInfo,))
# 1 method for generic function "methodinstances" from MethodAnalysis:
 [1] methodinstances(info::PkgCacheInfo)
     @ MethodAnalysisExt ~/.julia/packages/PkgCacheInspector/s2sOo/ext/MethodAnalysisExt.jl:5

That works (both are in my default environment).

Now open a fresh session:

(@v1.10) pkg> generate A
  Generating  project A:
    A/Project.toml
    A/src/A.jl

(@v1.10) pkg> activate ./A
  Activating project at `/tmp/pkgs/A`

(A) pkg> st
Project A v0.1.0
Status `/tmp/pkgs/A/Project.toml` (empty project)

julia> using PkgCacheInspector, MethodAnalysis

julia> methods(methodinstances, (PkgCacheInfo,))
# 1 method for generic function "methodinstances" from MethodAnalysis:
 [1] methodinstances(top)
     @ ~/.julia/packages/MethodAnalysis/oGkyP/src/MethodAnalysis.jl:171

Note that this is the generic method, indicating that the one specialized for PkgCacheInfo didn't load.

In case this has already been fixed, worth saying my master is a few days old:

julia> versioninfo()
Julia Version 1.10.0-DEV.368
Commit 65e6919079 (2023-01-14 12:26 UTC)
Platform Info:
  OS: Linux (x86_64-linux-gnu)
  CPU: 16 × 12th Gen Intel(R) Core(TM) i7-1260P
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-14.0.6 (ORCJIT, alderlake)
  Threads: 1 on 16 virtual cores

[KristofferC]

julia/base/loading.jl

Line 1115 in b8ca7cf

break # For now, only insert triggers for packages in the first load_path.

.

I think that is an old thing that was there before I had a more clear picture of how extensions should work with environment stacks. Should be fine to just remove it. I will do that and add a test.