Don't crash reading the config of a subrepo with no remote tree

[jackmarsh]

Building a Rust plugin against a repo with ~100 crate subrepos, we hit an intermittent segfault under remote execution whenever Please read a subrepo's config:

panic: runtime error: invalid memory address or nil pointer dereference
[signal SIGSEGV: segmentation violation code=0x1 addr=0x18]
please/src/remote/fs.New({...}, 0x0, {...})     fs.go:67
please/src/remote.(*Client).SubrepoFS(...)      remote.go:330
please/src/core.(*Subrepo).FS.func1()           subrepo.go:58
please/src/core.readSubrepoConfig(...)          subrepo.go:119

Note the literal 0x0 argument. SubrepoFS looks the subrepo's output tree up by label and passes whatever it finds straight to remotefs.New:

tree := c.subrepoTrees[target.Label]
return remotefs.New(c.remoteFSClient, tree, root)

subrepoTrees is only ever populated in one place (utils.go, when a target's outputs are set from a remote build), so a subrepo that was built locally, served from the cache, or not built yet has no entry, and tree is nil. New then does append(tree.Children, tree.Root) and dies. readSubrepoConfig calls Subrepo.FS() unconditionally — it has to, in order to look for .plzconfig — so any repo whose subrepo targets can come from cache rather than a fresh remote build can hit this.

That matches what we saw: it was sensitive to cache state and flipped between otherwise identical runs.

This makes a nil tree (or one with no root) produce an empty filesystem, so the config lookup reports "does not exist" and parsing continues — which is already the handling for a subrepo that has no .plzconfig. The added test panics at the same addr=0x18 without the change and passes with it.

plz test //src/remote/... is green.

[peterebden]

Suggested change

	tree = &pb.Tree{Root: &pb.Directory{}, Children: tree.Children}
	tree.Root = &pb.Directory{}

Think this is slightly clearer about what it wants to do

[jackmarsh]

Applied, thanks - reads better.

[peterebden]

Thanks!