populate git: refuse work trees whose .git names places elsewhere

A .git directory names places elsewhere in more ways than being a
symlink or a file: core.worktree, symlinks to a configuration, to
objects, or to refs elsewhere, a commondir file naming another
repository, and linked work trees.  Git's commands follow them, so
fetching, resetting, setting the remote's URL, and cleaning change what
is elsewhere: with core.worktree, git clean removes everything in the
work tree it names.  And Git runs commands the configuration names, like
core.fsmonitor, and hooks.

Populate only uses a work tree whose .git directory has no symlinks, no
commondir file, no linked work trees, and only configuration that git
clone and populate write, where submodules' repositories may have their
work tree in the work tree.  Otherwise populating fails with an error
telling what's wrong, unless the new option reclone replaces the work
tree by a fresh clone.  Git runs without hooks and without fsmonitor in
any case.

The work tree also gets reset to the commit if HEAD is there already: a
lock file rewritten in place, or created and staged by Nix, stays
otherwise once the source stops drawing inputs.

A safety scenario refuses work trees with core.worktree, a symlinked
configuration, a commondir file, and core.fsmonitor, replaces them with
reclone, and checks that a hook doesn't run.  A git scenario stops
drawing inputs and checks that the work tree is clean.

1 files changed, 93 insertions, 2 deletions

diff --git a/pkgs/populate/default.nix b/pkgs/populate/default.nix
index c707b7c..7130570 100644
--- a/pkgs/populate/default.nix
+++ b/pkgs/populate/default.nix
@@ -142,6 +142,68 @@ let
     fi
   '';
 
+  # Shell code printing why populate can't use the .git directory of the
+  # Git work tree at path, or nothing.  A .git directory that git clone and
+  # populate made has no symlinks, no commondir file, which names another
+  # repository, no linked work trees, and only the configuration they write:
+  # the repository's format, the remote, branches, and submodules, whose
+  # repositories have their work tree in the work tree.
+  gitProblems = path: let
+    gitDir = "${path}/.git";
+    # Keys of the configuration git clone writes, as patterns of case.
+    keys = [
+      "core.repositoryformatversion" "core.filemode" "core.bare"
+      "core.logallrefupdates" "core.symlinks" "core.ignorecase"
+      "core.precomposeunicode"
+      "extensions.objectformat" "extensions.refstorage"
+      "remote.origin.url" "remote.origin.fetch"
+      "branch.*.remote" "branch.*.merge"
+      "submodule.active" "submodule.*.url" "submodule.*.active"
+    ];
+  in /* sh */ ''(
+    links=$(find ${quote gitDir} -type l | head -n 3 | paste -s -d ' ' -)
+    if test -n "$links"; then
+      printf '%s contains symlinks: %s' ${quote gitDir} "$links"
+      exit
+    fi
+    common=$(find ${quote gitDir} -name commondir | head -n 1)
+    if test -n "$common"; then
+      printf '%s names another repository' "$common"
+      exit
+    fi
+    if test -e ${quote gitDir}/worktrees; then
+      printf '%s has linked work trees' ${quote gitDir}
+      exit
+    fi
+    find ${quote gitDir} -name config -type f | while IFS= read -r config; do
+      command git config --file "$config" --list --name-only |
+      while IFS= read -r key; do
+        case $key in
+          ${concatStringsSep "|" keys}) ;;
+          core.worktree)
+            # Submodules' repositories have their work tree in the work tree.
+            worktree=$(command git config --file "$config" core.worktree)
+            dir=$(dirname "$config")
+            case $dir/$worktree in
+              ${quote gitDir}/modules/*)
+                resolved=$(cd "$dir" && cd "$worktree" 2>/dev/null && pwd -P) || :
+                case $resolved in
+                  "$(cd ${quote path} && pwd -P)"/*) continue ;;
+                esac
+                ;;
+            esac
+            printf '%s sets core.worktree to %s' "$config" "$worktree"
+            exit 1
+            ;;
+          *)
+            printf '%s sets %s, which neither git clone nor populate does' "$config" "$key"
+            exit 1
+            ;;
+        esac
+      done || exit 1
+    done || :
+  )'';
+
   check = { force, target }: let
     sentinelFile = "${target.path}/.populate";
   in runShell target /* sh */ ''
@@ -1000,6 +1062,12 @@ let
 
   pop.git = target: source: runShell target /* sh */ ''
     set -efu
+    # Hooks and commands a repository's configuration names run with Git's
+    # commands, so populate's Git doesn't run any.  The configuration gets
+    # checked as well, but Git also runs a hook it has itself.
+    git() {
+      command git -c core.hooksPath=/dev/null -c core.fsmonitor=false "$@"
+    }
     # Remove target path if it doesn't look like a git worktree.
     # This can happen e.g. when it had a different type earlier.  A work
     # tree that a symlink points to, or whose .git is a symlink or a file
@@ -1010,6 +1078,27 @@ let
         ! test -d ${quote target.path}/.git; then
       rm -fR ${quote target.path}
     fi
+    # A .git directory can name places elsewhere too, which Git's commands
+    # change, or commands, which they run.  Populate only uses a work tree
+    # whose .git directory has nothing like that, and else replaces it if
+    # reclone is set, or fails, telling why.
+    if test -e ${quote target.path}; then
+      # Not $((, which would be arithmetic.
+      if problems=$( ${gitProblems target.path} ); then :; else
+        exit 1
+      fi
+      if test -n "$problems"; then
+        ${if source.reclone then /* sh */ ''
+          >&2 printf 'replacing the Git work tree %s, because %s\n' \
+              ${quote target.path} "$problems"
+          rm -fR ${quote target.path}
+        '' else /* sh */ ''
+          >&2 printf 'error: populate doesn'"'"'t use the Git work tree %s, because %s.  Fix that, remove the work tree, or set the source'"'"'s reclone to true to replace it by a fresh clone.\n' \
+              ${quote target.path} "$problems"
+          exit 1
+        ''}
+      fi
+    fi
     if ! test -e ${quote target.path}; then
       ${if source.shallow then /* sh */ ''
         git init ${quote target.path}
@@ -1043,9 +1132,11 @@ let
           ''}
         fi
       ''}
-      git reset --hard "$hash" >&2
-      git submodule update --init --recursive
     fi
+    # Also if HEAD is there already: rewriting a lock file in place, e.g.,
+    # changes the work tree, and stages the lock file if Nix created it.
+    git reset --hard "$hash" >&2
+    git submodule update --init --recursive
 
     git clean -dfx \
         ${concatMapStringsSep " "