lib/model: Scanning unknown items is OK as long as the parent is known (fixes #2915)

lib/model/model.go

@@ -2086,31 +2086,39 @@ func stringSliceWithout(ss []string, s string) []string {
 	return ss
 }
 
-// unifySubs takes a list of files or directories and trims them down to
+// The exists function is expected to return true for all known paths
-// themselves or the closest parent that exists() returns true for, while
+// (excluding "" and ".")
-// removing duplicates and subdirectories. That is, if we have foo/ in the
-// list, we don't also need foo/bar/ because that's already covered.
 func unifySubs(dirs []string, exists func(dir string) bool) []string {
-	var subs []string
+	subs := trimUntilParentKnown(dirs, exists)
+	sort.Strings(subs)
+	return simplifySortedPaths(subs)
+}
 
-	// Trim each item to itself or its closest known parent
+func trimUntilParentKnown(dirs []string, exists func(dir string) bool) []string {
+	var subs []string
 	for _, sub := range dirs {
 		for sub != "" && sub != ".stfolder" && sub != ".stignore" {
-			if exists(sub) {
+			sub = filepath.Clean(sub)
+			parent := filepath.Dir(sub)
+			if parent == "." || exists(parent) {
 				break
 			}
-			sub = filepath.Dir(sub)
+			sub = parent
 			if sub == "." || sub == string(filepath.Separator) {
 				// Shortcut. We are going to scan the full folder, so we can
 				// just return an empty list of subs at this point.
 				return nil
 			}
 		}
+		if sub == "" {
+			return nil
+		}
 		subs = append(subs, sub)
 	}
+	return subs
+}
 
-	// Remove any paths that are already covered by their parent
+func simplifySortedPaths(subs []string) []string {
-	sort.Strings(subs)
 	var cleaned []string
 next:
 	for _, sub := range subs {

lib/model/model_test.go

@@ -1236,50 +1236,56 @@ func TestUnifySubs(t *testing.T) {
 		out    []string // expected output
 	}{
 		{
-			// trailing slashes are cleaned, known paths are just passed on
+			// 0. trailing slashes are cleaned, known paths are just passed on
 			[]string{"foo/", "bar//"},
 			[]string{"foo", "bar"},
 			[]string{"bar", "foo"}, // the output is sorted
 		},
 		{
-			// "foo/bar" gets trimmed as it's covered by foo
+			// 1. "foo/bar" gets trimmed as it's covered by foo
 			[]string{"foo", "bar/", "foo/bar/"},
 			[]string{"foo", "bar"},
 			[]string{"bar", "foo"},
 		},
 		{
-			// "bar" gets trimmed to "" as it's unknown,
+			// 2. "" gets simplified to the empty list; ie scan all
-			// "" gets simplified to the empty list
+			[]string{"foo", ""},
-			[]string{"foo", "bar", "foo/bar"},
 			[]string{"foo"},
 			nil,
 		},
 		{
-			// two independent known paths, both are kept
+			// 3. "foo/bar" is unknown, but it's kept
+			// because its parent is known
+			[]string{"foo/bar"},
+			[]string{"foo"},
+			[]string{"foo/bar"},
+		},
+		{
+			// 4. two independent known paths, both are kept
 			// "usr/lib" is not a prefix of "usr/libexec"
 			[]string{"usr/lib", "usr/libexec"},
-			[]string{"usr/lib", "usr/libexec"},
+			[]string{"usr", "usr/lib", "usr/libexec"},
 			[]string{"usr/lib", "usr/libexec"},
 		},
 		{
-			// "usr/lib" is a prefix of "usr/lib/exec"
+			// 5. "usr/lib" is a prefix of "usr/lib/exec"
 			[]string{"usr/lib", "usr/lib/exec"},
-			[]string{"usr/lib", "usr/libexec"},
+			[]string{"usr", "usr/lib", "usr/libexec"},
 			[]string{"usr/lib"},
 		},
 		{
-			// .stignore and .stfolder are special and are passed on
+			// 6. .stignore and .stfolder are special and are passed on
 			// verbatim even though they are unknown
 			[]string{".stfolder", ".stignore"},
 			[]string{},
 			[]string{".stfolder", ".stignore"},
 		},
 		{
-			// but the presense of something else unknown forces an actual
+			// 7. but the presense of something else unknown forces an actual
 			// scan
 			[]string{".stfolder", ".stignore", "foo/bar"},
 			[]string{},
-			nil,
+			[]string{".stfolder", ".stignore", "foo"},
 		},
 	}