use gix_fs::current_dir(precompose_unicode).

That way, paths will be precomposed when we work with them.

Author: Byron

Cargo.lock

@@ -17,6 +17,7 @@ gix-sec = { version = "^0.10.3", path = "../gix-sec" }
 gix-path = { version = "^0.10.3", path = "../gix-path" }
 gix-ref = { version = "^0.40.1", path = "../gix-ref" }
 gix-hash = { version = "^0.14.1", path = "../gix-hash" }
+gix-fs = { version = "^0.9.1", path = "../gix-fs" }
 
 bstr = { version = "1.3.0", default-features = false, features = ["std", "unicode"] }
 thiserror = "1.0.26"

gix-discover/Cargo.toml

@@ -39,7 +39,15 @@ pub(crate) mod function {
         // us the parent directory. (`Path::parent` just strips off the last
         // path component, which means it will not do what you expect when
         // working with paths paths that contain '..'.)
-        let cwd = current_dir.map_or_else(|| std::env::current_dir().map(Cow::Owned), |cwd| Ok(Cow::Borrowed(cwd)))?;
+        let cwd = current_dir.map_or_else(
+            || {
+                // The paths we return are relevant to the repository, but at this time it's impossible to know
+                // what `core.precomposeUnicode` is going to be. Hence the one using these paths will have to
+                // transform the paths as needed, because we can't. `false` means to leave the obtained path as is.
+                gix_fs::current_dir(false).map(Cow::Owned)
+            },
+            |cwd| Ok(Cow::Borrowed(cwd)),
+        )?;
         #[cfg(windows)]
         let directory = dunce::simplified(directory);
         let dir = gix_path::normalize(directory.into(), cwd.as_ref()).ok_or_else(|| Error::InvalidInput {

gix-discover/src/upwards/mod.rs

@@ -63,6 +63,10 @@ pub struct Options<'a> {
     /// that this is merely an optimization for those who discover a lot of repositories in the same process.
     ///
     /// If unset, the current working directory will be obtained automatically.
+    /// Note that the path here might or might not contained decomposed unicode, which may end up in a path
+    /// relevant us, like the git-dir or the worktree-dir. However, when opening the repository, it will
+    /// change decomposed unicode to precomposed unicode based on the value of `core.precomposeUnicode`, and we
+    /// don't have to deal with that value here just yet.
     pub current_dir: Option<&'a std::path::Path>,
 }
 

gix-discover/src/upwards/types.rs

@@ -25,6 +25,7 @@ gix-path = { version = "^0.10.3", path = "../gix-path" }
 gix-quote = { version = "^0.4.10", path = "../gix-quote" }
 gix-object = { version = "^0.40.1", path = "../gix-object" }
 gix-pack = { version = "^0.46.1", path = "../gix-pack", default-features = false }
+gix-fs = { version = "^0.9.1", path = "../gix-fs" }
 serde = { version = "1.0.114", optional = true, default-features = false, features = ["derive"]}
 
 tempfile = "3.1.0"

gix-odb/Cargo.toml

@@ -17,7 +17,7 @@ pub struct Options {
     /// If false, no multi-pack indices will be used. If true, they will be used if their hash matches `object_hash`.
     pub use_multi_pack_index: bool,
     /// The current directory of the process at the time of instantiation.
-    /// If unset, it will be retrieved using `std::env::current_dir()`.
+    /// If unset, it will be retrieved using `gix_fs::current_dir(false)`.
     pub current_dir: Option<std::path::PathBuf>,
 }
 
@@ -80,7 +80,13 @@ impl Store {
         }: Options,
     ) -> std::io::Result<Self> {
         let _span = gix_features::trace::detail!("gix_odb::Store::at()");
-        let current_dir = current_dir.map_or_else(std::env::current_dir, Ok)?;
+        let current_dir = current_dir.map_or_else(
+            || {
+                // It's only used for real-pathing alternate paths and there it just needs to be consistent (enough).
+                gix_fs::current_dir(false)
+            },
+            Ok,
+        )?;
         if !objects_dir.is_dir() {
             return Err(std::io::Error::new(
                 std::io::ErrorKind::Other, // TODO: use NotADirectory when stabilized

gix-odb/src/store_impls/dynamic/init.rs

@@ -246,7 +246,7 @@ pub fn to_windows_separators<'a>(path: impl Into<Cow<'a, BStr>>) -> Cow<'a, BStr
 /// instead.
 ///
 /// Note that we might access the `current_dir` if we run out of path components to pop off, which is expected to be absolute
-/// as typical return value of `std::env::current_dir()`.
+/// as typical return value of `std::env::current_dir()` or `gix_fs::current_dir(…)` when `core.precomposeUnicode` is known.
 /// As a `current_dir` like `/c` can be exhausted by paths like `../../r`, `None` will be returned to indicate the inability
 /// to produce a logically consistent path.
 pub fn normalize<'a>(path: Cow<'a, Path>, current_dir: &Path) -> Option<Cow<'a, Path>> {

gix-path/src/convert.rs

@@ -30,6 +30,8 @@ pub(crate) mod function {
     /// Do not fail for non-existing components, but assume these are as is.
     ///
     /// If `path` is relative, the current working directory be used to make it absolute.
+    /// Note that the returned path will be verbatim, and repositories with `core.precomposeUnicode`
+    /// set will probably want to precompose the paths unicode.
     pub fn realpath(path: impl AsRef<Path>) -> Result<PathBuf, Error> {
         let path = path.as_ref();
         let cwd = path

gix-path/src/realpath.rs

@@ -1,6 +1,8 @@
 #![allow(clippy::result_large_err)]
+
 use super::{util, Error};
-use crate::config::tree::{Core, Extensions};
+use crate::config::cache::util::ApplyLeniency;
+use crate::config::tree::{Core, Extensions, Key};
 
 /// A utility to deal with the cyclic dependency between the ref store and the configuration. The ref-store needs the
 /// object hash kind, and the configuration needs the current branch name to resolve conditional includes with `onbranch`.
@@ -12,6 +14,7 @@ pub(crate) struct StageOne {
     pub lossy: Option<bool>,
     pub object_hash: gix_hash::Kind,
     pub reflog: Option<gix_ref::store::WriteReflog>,
+    pub precompose_unicode: bool,
 }
 
 /// Initialization
@@ -69,6 +72,13 @@ impl StageOne {
             )?;
             config.append(worktree_config);
         };
+        let precompose_unicode = config
+            .boolean("core", None, Core::PRECOMPOSE_UNICODE.name())
+            .map(|v| Core::PRECOMPOSE_UNICODE.enrich_error(v))
+            .transpose()
+            .with_leniency(lenient)
+            .map_err(Error::ConfigBoolean)?
+            .unwrap_or_default();
 
         let reflog = util::query_refupdates(&config, lenient)?;
         Ok(StageOne {
@@ -78,6 +88,7 @@ impl StageOne {
             lossy,
             object_hash,
             reflog,
+            precompose_unicode,
         })
     }
 }

gix/src/config/cache/incubate.rs

@@ -27,6 +27,7 @@ impl Cache {
             is_bare,
             object_hash,
             reflog: _,
+            precompose_unicode: _,
         }: StageOne,
         git_dir: &std::path::Path,
         branch_name: Option<&gix_ref::FullNameRef>,

gix/src/config/cache/init.rs

@@ -205,9 +205,9 @@ pub fn into(
         write_file(tpl, PathCursor(&mut dot_git).at(filename))?;
     }
 
-    {
+    let caps = {
         let mut config = gix_config::File::default();
-        {
+        let caps = {
             let caps = fs_capabilities.unwrap_or_else(|| gix_fs::Capabilities::probe(&dot_git));
             let mut core = config.new_section("core", None).expect("valid section name");
 
@@ -218,14 +218,16 @@ pub fn into(
             core.push(key("symlinks"), Some(bool(caps.symlink).into()));
             core.push(key("ignorecase"), Some(bool(caps.ignore_case).into()));
             core.push(key("precomposeunicode"), Some(bool(caps.precompose_unicode).into()));
-        }
+            caps
+        };
         let mut cursor = PathCursor(&mut dot_git);
         let config_path = cursor.at("config");
         std::fs::write(config_path, config.to_bstring()).map_err(|err| Error::IoWrite {
             source: err,
             path: config_path.to_owned(),
         })?;
-    }
+        caps
+    };
 
     Ok(gix_discover::repository::Path::from_dot_git_dir(
         dot_git,
@@ -234,7 +236,7 @@ pub fn into(
         } else {
             gix_discover::repository::Kind::WorkTree { linked_git_dir: None }
         },
-        &std::env::current_dir()?,
+        &gix_fs::current_dir(caps.precompose_unicode)?,
     )
     .expect("by now the `dot_git` dir is valid as we have accessed it"))
 }

gix/src/create.rs

@@ -43,7 +43,8 @@ impl ThreadSafeRepository {
         let (git_dir, worktree_dir) = path.into_repository_and_work_tree_directories();
         let mut options = trust_map.into_value_by_level(trust);
         options.git_dir_trust = trust.into();
-        options.current_dir = Some(std::env::current_dir().map_err(upwards::Error::CurrentDir)?);
+        // Note that we will adjust the `current_dir` later so it matches the value of `core.precomposeUnicode`.
+        options.current_dir = Some(gix_fs::current_dir(false).map_err(upwards::Error::CurrentDir)?);
         Self::open_from_paths(git_dir, worktree_dir, options).map_err(Into::into)
     }
 

gix/src/discover.rs

@@ -68,7 +68,8 @@ impl ThreadSafeRepository {
         let path = crate::create::into(directory.as_ref(), kind, create_options)?;
         let (git_dir, worktree_dir) = path.into_repository_and_work_tree_directories();
         open_options.git_dir_trust = Some(gix_sec::Trust::Full);
-        open_options.current_dir = std::env::current_dir()?.into();
+        // The repo will use `core.precomposeUnicode` to adjust the value as needed.
+        open_options.current_dir = gix_fs::current_dir(false)?.into();
         let repo = ThreadSafeRepository::open_from_paths(git_dir, worktree_dir, open_options)?;
 
         let branch_name = repo

gix/src/init.rs

@@ -5,7 +5,6 @@ use gix_features::threading::OwnShared;
 use gix_macros::momo;
 
 use super::{Error, Options};
-use crate::config::cache::util::ApplyLeniency;
 use crate::{
     config,
     config::{
@@ -88,7 +87,9 @@ impl ThreadSafeRepository {
                 }
             }
         };
-        let cwd = std::env::current_dir()?;
+
+        // The be altered later based on `core.precomposeUnicode`.
+        let cwd = gix_fs::current_dir(false)?;
         let (git_dir, worktree_dir) = gix_discover::repository::Path::from_dot_git_dir(path, kind, &cwd)
             .expect("we have sanitized path with is_git()")
             .into_repository_and_work_tree_directories();
@@ -137,7 +138,8 @@ impl ThreadSafeRepository {
             }
         };
 
-        let cwd = std::env::current_dir()?;
+        // The be altered later based on `core.precomposeUnicode`.
+        let cwd = gix_fs::current_dir(false)?;
         let (git_dir, worktree_dir) = gix_discover::repository::Path::from_dot_git_dir(path, path_kind, &cwd)
             .expect("we have sanitized path with is_git()")
             .into_repository_and_work_tree_directories();
@@ -150,9 +152,9 @@ impl ThreadSafeRepository {
     }
 
     pub(crate) fn open_from_paths(
-        git_dir: PathBuf,
+        mut git_dir: PathBuf,
         mut worktree_dir: Option<PathBuf>,
-        options: Options,
+        mut options: Options,
     ) -> Result<Self, Error> {
         let _span = gix_trace::detail!("open_from_paths()");
         let Options {
@@ -171,47 +173,58 @@ impl ThreadSafeRepository {
                 },
             ref api_config_overrides,
             ref cli_config_overrides,
-            ref current_dir,
+            ref mut current_dir,
         } = options;
-        let current_dir = current_dir.as_deref().expect("BUG: current_dir must be set by caller");
         let git_dir_trust = git_dir_trust.expect("trust must be determined by now");
 
-        // TODO: assure we handle the worktree-dir properly as we can have config per worktree with an extension.
-        //       This would be something read in later as have to first check for extensions. Also this means
-        //       that each worktree, even if accessible through this instance, has to come in its own Repository instance
-        //       as it may have its own configuration. That's fine actually.
-        let common_dir = gix_discover::path::from_plain_file(git_dir.join("commondir").as_ref())
+        let mut common_dir = gix_discover::path::from_plain_file(git_dir.join("commondir").as_ref())
             .transpose()?
             .map(|cd| git_dir.join(cd));
-        let common_dir_ref = common_dir.as_deref().unwrap_or(&git_dir);
-
         let repo_config = config::cache::StageOne::new(
-            common_dir_ref,
+            common_dir.as_deref().unwrap_or(&git_dir),
             git_dir.as_ref(),
             git_dir_trust,
             lossy_config,
             lenient_config,
         )?;
+
+        if repo_config.precompose_unicode {
+            git_dir = gix_utils::str::precompose_path(git_dir.into()).into_owned();
+            if let Some(common_dir) = common_dir.as_mut() {
+                if let Cow::Owned(precomposed) = gix_utils::str::precompose_path((&*common_dir).into()) {
+                    *common_dir = precomposed;
+                }
+            }
+            if let Some(worktree_dir) = worktree_dir.as_mut() {
+                if let Cow::Owned(precomposed) = gix_utils::str::precompose_path((&*worktree_dir).into()) {
+                    *worktree_dir = precomposed;
+                }
+            }
+        }
+        let common_dir_ref = common_dir.as_deref().unwrap_or(&git_dir);
+
+        let current_dir = {
+            let current_dir_ref = current_dir.as_mut().expect("BUG: current_dir must be set by caller");
+            if repo_config.precompose_unicode {
+                if let Cow::Owned(precomposed) = gix_utils::str::precompose_path((&*current_dir_ref).into()) {
+                    *current_dir_ref = precomposed;
+                }
+            }
+            current_dir_ref.as_path()
+        };
+
         let mut refs = {
             let reflog = repo_config.reflog.unwrap_or(gix_ref::store::WriteReflog::Disable);
             let object_hash = repo_config.object_hash;
-            let precompose_unicode = repo_config
-                .git_dir_config
-                .boolean("core", None, Core::PRECOMPOSE_UNICODE.name())
-                .map(|v| Core::PRECOMPOSE_UNICODE.enrich_error(v))
-                .transpose()
-                .with_leniency(lenient_config)
-                .map_err(|err| Error::Config(err.into()))?
-                .unwrap_or_default();
             match &common_dir {
                 Some(common_dir) => crate::RefStore::for_linked_worktree(
                     git_dir.to_owned(),
                     common_dir.into(),
                     reflog,
                     object_hash,
-                    precompose_unicode,
+                    repo_config.precompose_unicode,
                 ),
-                None => crate::RefStore::at(git_dir.to_owned(), reflog, object_hash, precompose_unicode),
+                None => crate::RefStore::at(git_dir.to_owned(), reflog, object_hash, repo_config.precompose_unicode),
             }
         };
         let head = refs.find("HEAD").ok();

gix/src/open/repository.rs

@@ -123,7 +123,9 @@ impl<'repo> Remote<'repo> {
                 let (git_dir, _work_dir) = gix_discover::repository::Path::from_dot_git_dir(
                     dir.clone().into_owned(),
                     kind,
-                    &std::env::current_dir()?,
+                    // precomposed unicode doesn't matter here as long as the produced path is accessible,
+                    // which is a given either way.
+                    &gix_fs::current_dir(false)?,
                 )
                 .ok_or_else(|| Error::InvalidRemoteRepositoryPath {
                     directory: dir.into_owned(),

gix/src/remote/connect.rs

@@ -1,7 +1,56 @@
+use std::borrow::Cow;
 use std::error::Error;
 
 use crate::util::named_subrepo_opts;
 
+#[test]
+fn on_root_with_decomposed_unicode() -> crate::Result {
+    let tmp = gix_testtools::tempfile::TempDir::new()?;
+
+    let decomposed = "a\u{308}";
+
+    let root = tmp.path().join(decomposed);
+    std::fs::create_dir(&root)?;
+
+    let repo = gix::init(root)?;
+    let precompose_unicode = repo
+        .config_snapshot()
+        .boolean("core.precomposeUnicode")
+        .expect("created by init based on fs-capabilities");
+
+    assert!(repo.git_dir().is_dir());
+    let work_dir = repo.work_dir().expect("non-bare");
+    assert!(work_dir.is_dir());
+
+    if precompose_unicode {
+        assert!(
+            matches!(
+                gix::utils::str::precompose_path(repo.git_dir().into()),
+                Cow::Borrowed(_),
+            ),
+            "there is no change, as the path is already precomposed"
+        );
+        assert!(matches!(
+            gix::utils::str::precompose_path(work_dir.into()),
+            Cow::Borrowed(_),
+        ));
+    } else {
+        assert!(
+            matches!(
+                gix::utils::str::precompose_path(repo.git_dir().into()),
+                Cow::Owned(_),
+            ),
+            "this has an effect as the path isn't precomposed, a necessity on filesystems that don't fold decomposition"
+        );
+        assert!(matches!(
+            gix::utils::str::precompose_path(work_dir.into()),
+            Cow::Owned(_),
+        ));
+    }
+
+    Ok(())
+}
+
 #[test]
 fn bare_repo_with_index() -> crate::Result {
     let repo = named_subrepo_opts(