Fix linker failure on windows

[GuillaumeGomez]

Fixes #38933.

r? @ollie27

[nagisa]

Instead of not setting PATH altogether, we should just ignore the components that make construction of the PATH to fail.

[GuillaumeGomez]

It would give an invalid PATH. Not sure this is what we want... At least, the current change doesn't set the variable at all if any path is not valid.

[nagisa]

It would give an invalid PATH.

What? Of course it would become a valid PATH once you filter out all the parts that can’t be put into PATH. Maybe it would not be complete, but certainly valid.

What your current patch does, on the other hand, is making whatever function this code had not come into effect whenever there’s any path from any source that cannot be put into PATH.

[GuillaumeGomez]

I think I missed something then. In here, we just join 2 paths, if one is invalid, we should only keep the other one, right? I'll try to update in this way then.

[GuillaumeGomez]

@nagisa: Better?

[nagisa]

Remembering that $PATH and sess.host_filesearch(PathKind::All).get_dylib_search_paths() both are collection of pathes (i.e. vec![p1, p2, p3, ...] and vec![dsp1, dsp2, dsp3, ...]) and we want to merge them into a single $PATH… pathes from $PATH cannot be invalid as they are parsed from the same $PATH, but the dsp1.. could be. Even then you still want to join them as they have some purpose – not merging any of the dsp* in when only dsp42 is invalid seems lazy.

So here’s some implementation strategies I had in mind:

// validate pathes by joining them one by one:
env::join_paths(new_paths.filter(|p| env::join_paths(iter::once(path)).is_ok()));

or

// write an iterator such that
struct Iter<E: Clone> { last_yielded: E, other: ... }

loop {
let i = Iter { ... };
let r = env::join_paths(i);
if r.is_ok() { break; }
// i.last_yielded is the invalid path that cannot be included into $PATH, remove and retry.
}

does it make what I meant any clearer?

[GuillaumeGomez]

Indeed. I'll just take your first example then.

[GuillaumeGomez]

Updated.

[GuillaumeGomez]

Any news @nagisa?

[nagisa]

This approach LGTM to me.

[GuillaumeGomez]

@bors: r=nagisa

[bors]

📌 Commit 6f58b7f has been approved by nagisa

[bors]

⌛ Testing commit 6f58b7f with merge 4ce7acc...

[bors]

☀️ Test successful - status-appveyor, status-travis
Approved by: nagisa
Pushing 4ce7acc to master...