checkout: take into account path chars ignored on HFS+

Summary: Apparently the HFS+ file system ignores some characters, so we must also ignore them when validating paths.

Reviewed By: sggutier

Differential Revision: D54037548

fbshipit-source-id: 2a323aa45557cf206439a9d5711e911180ea42c2

eden/scm/lib/vfs/src/pathauditor.rs

@@ -5,6 +5,7 @@
  * GNU General Public License version 2.
  */
 
+use std::borrow::Cow;
 use std::fs::symlink_metadata;
 use std::path::Path;
 use std::path::PathBuf;
@@ -43,6 +44,13 @@ static INVALID_COMPONENTS: Lazy<Vec<&'static str>> = Lazy::new(|| {
         .collect()
 });
 
+// From encoding.py: These unicode characters are ignored by HFS+ (Apple Technote 1150,
+// "Unicode Subtleties"), so we need to ignore them in some places for sanity.
+const IGNORED_HFS_CHARS: [char; 16] = [
+    '\u{200c}', '\u{200d}', '\u{200e}', '\u{200f}', '\u{202a}', '\u{202b}', '\u{202c}', '\u{202d}',
+    '\u{202e}', '\u{206a}', '\u{206b}', '\u{206c}', '\u{206d}', '\u{206e}', '\u{206f}', '\u{feff}',
+];
+
 #[derive(thiserror::Error, Debug)]
 pub enum AuditError {
     #[error("Can't read/write file through ancestor symlink \"{0}\"")]
@@ -135,14 +143,19 @@ fn valid_windows_component(component: &str) -> bool {
 /// It also checks that no trailing dots are part of the component and checks that shortnames
 /// on Windows are valid.
 fn audit_invalid_components(path: &str) -> Result<(), AuditError> {
-    let path = if cfg!(not(windows)) {
-        path.to_owned()
+    let path: Cow<str> = if cfg!(not(windows)) {
+        Cow::Borrowed(path)
     } else {
-        path.to_lowercase()
+        Cow::Owned(path.to_lowercase())
     };
     for s in path.split(SEPARATORS) {
-        if s.is_empty() || INVALID_COMPONENTS.contains(&s) || !valid_windows_component(s) {
-            return Err(AuditError::InvalidComponent(s.to_owned()));
+        let s = if s.contains(IGNORED_HFS_CHARS) {
+            Cow::Owned(s.replace(IGNORED_HFS_CHARS, ""))
+        } else {
+            Cow::Borrowed(s)
+        };
+        if s.is_empty() || INVALID_COMPONENTS.contains(&&*s) || !valid_windows_component(&s) {
+            return Err(AuditError::InvalidComponent(s.into_owned()));
         }
     }
     Ok(())

eden/scm/tests/test-commit.t

@@ -3,6 +3,7 @@
   $ eagerepo
 This is needed to avoid filelog() revset in "log", which isn't compatible w/ eagerepo.
   $ setconfig experimental.pathhistory=true
+  $ setconfig checkout.use-rust=true
 
 commit date test
 
@@ -668,10 +669,15 @@ verify pathauditor blocks evil filepaths
   [255]
 #else
   $ hg co --clean tip
-  abort: path contains illegal component: .h\xe2\x80\x8cg/hgrc (esc)
+  abort: Can't write 'RepoPath(".h\u{200c}g/hgrc")' after handling error "Can't write into .h‌g/hgrc
+  
+  Caused by:
+      0: Invalid component in ".h‌g/hgrc"
+      1: Invalid path component ".hg"": Invalid path component ".hg"
   [255]
 #endif
 
+#if windows
   $ cd $TESTTMP/audit2
   $ cat > evil-commit.py <<EOF
   > from __future__ import absolute_import
@@ -709,6 +715,7 @@ verify pathauditor blocks evil filepaths
   $ hg co --clean tip
   abort: path contains illegal component: HG8B6C~2/hgrc
   [255]
+#endif
 
 # test that an unmodified commit template message aborts