git_cache 0.1.1 → 0.1.2

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: d3b7101169665dc9115fad2454928336b6ef83cca3b1db38a8ae6acef81062b1
-  data.tar.gz: f58df6572f0e3511ef30e4e7f1b0843e54ee59b157edb77d84f268b8a4bc0135
+  metadata.gz: 1c5d6cf239e6dd1cd7ba89cc4721ddc05dc5a1c1ff2d0c6b7fded9b06bac5eae
+  data.tar.gz: b435524513c83ce7fdca4aed22351cb29fda9da8435f6a7af94c591b87791f5b
 SHA512:
-  metadata.gz: 0be87372ec841eb2288330508ce7bd5aaa61cb7f1332b78ff4866f63c071534204309dfb2bc4417bb47f093901dc22063cf1ba5c84d9f0f0391addf660fa0b91
-  data.tar.gz: 45fa8853c29cc760cd59f3054a39d43bf13d72b5fc9f2afcb4cdf5460a3b30b96a212754d6c878d08d418986cf2858a1108c1c68e5b191b3a437ed76e0d775b7
+  metadata.gz: 7d84f21b22047b18c1d1734d31b84454555d3a5abc8ae7d63ea8b6a9a443393fc0c8cd2bfaf2c8935cd514e3771d999b873ac360a21a107501a7515d44dc7935
+  data.tar.gz: 135e7c7031ffa4bf11b2d3b3ace3179c0b4bdaa4b3d67d1e5e1d43b47add706194db510b0bb504a501f873b200f31128d8a4c591f69c9f00f26dd7514f63790d

data/CHANGELOG.md

@@ -1,5 +1,9 @@
 # Release History
 
+### v0.1.2 / 2026-08-20
+
+* FIXED: Prevent git auto maintenance from racing with cache removals
+
 ### v0.1.1 / 2026-05-05
 
 * BREAKING CHANGE: Make GitCache.sources_writable? private for now

data/lib/git_cache/error.rb

@@ -9,11 +9,11 @@ class GitCache
     # Create a GitCache::Error.
     #
     # @param message [String] The error message
-    # @param result [::ExecService::Result] The result of a git
-    #     command execution, or `nil` if this error was not due to a git
-    #     command error.
+    # @param result [::ExecService::Result,nil] The result of a git
+    #     command execution. Omit or pass `nil` if this error was not due to
+    #     a git command error.
     #
-    def initialize(message, result)
+    def initialize(message, result = nil)
       super(message)
       @exec_result = result
     end

data/lib/git_cache/version.rb

@@ -5,5 +5,5 @@ class GitCache
   # Version of the git_cache gem
   # @return [String]
   #
-  VERSION = "0.1.1"
+  VERSION = "0.1.2"
 end

data/lib/git_cache.rb

@@ -21,6 +21,7 @@ class GitCache
     require "digest"
     require "fileutils"
     require "json"
+    require "securerandom"
     require "exec_service"
     @cache_dir = ::File.expand_path(cache_dir || default_cache_dir)
     @exec = ::ExecService.new(out: :capture, err: :capture)
@@ -140,8 +141,7 @@ class GitCache
     Array(remotes).map do |remote|
       dir = repo_base_dir_for(remote)
       if ::File.directory?(dir)
-        ::FileUtils.chmod_R("u+w", dir, force: true)
-        ::FileUtils.rm_rf(dir)
+        remove_dir(dir)
         remote
       end
     end.compact.sort
@@ -203,9 +203,7 @@ class GitCache
       end
       results.map(&:sha).uniq.each do |sha|
         unless repo_lock.source_exists?(sha)
-          sha_dir = ::File.join(dir, sha)
-          ::FileUtils.chmod_R("u+w", sha_dir, force: true)
-          ::FileUtils.rm_rf(sha_dir)
+          remove_dir(::File.join(dir, sha))
         end
       end
     end
@@ -217,7 +215,19 @@ class GitCache
   FORMAT_VERSION = "v1"
   REPO_DIR_NAME = "repo"
   LOCK_FILE_NAME = "repo.lock"
-  private_constant :REPO_DIR_NAME, :LOCK_FILE_NAME, :FORMAT_VERSION
+  TRASH_DIR_PREFIX = ".trash-"
+
+  # Config applied to every git invocation. Auto maintenance would otherwise
+  # spawn a detached `git maintenance run --auto` process after each fetch,
+  # which keeps writing into the cache repo after the fetch has returned, and
+  # thus races with our own traversals and removals of that directory. These
+  # are managed cache repos that we recreate at will, so background repacking
+  # buys them nothing. Requires git 2.30 or later; older gits ignore config
+  # keys they do not recognize.
+  GIT_CONFIG_ARGS = ["-c", "maintenance.auto=false"].freeze
+
+  private_constant :REPO_DIR_NAME, :LOCK_FILE_NAME, :FORMAT_VERSION,
+                   :TRASH_DIR_PREFIX, :GIT_CONFIG_ARGS
 
   def repo_base_dir_for(remote)
     ::File.join(@cache_dir, ::GitCache.remote_dir_name(remote))
@@ -229,13 +239,72 @@ class GitCache
   end
 
   def git(dir, cmd, error_message: nil)
-    result = @exec.exec(["git"] + cmd, chdir: dir)
+    result = @exec.exec(["git"] + GIT_CONFIG_ARGS + cmd, chdir: dir)
     if !result.success? && error_message
       raise ::GitCache::Error.new(error_message, result)
     end
     result
   end
 
+  # Removes a directory from the cache.
+  #
+  # The directory is first renamed out of the way, which is atomic and thus
+  # unaffected by any concurrent writes happening inside it, so the cache
+  # entry is gone as far as any client is concerned as soon as this returns.
+  # Only then is the renamed copy deleted, best effort. Anything left behind
+  # is invisible to clients and is swept up by later removals.
+  #
+  # Raises {GitCache::Error} if the directory could not be removed at all.
+  #
+  def remove_dir(dir)
+    return unless ::File.exist?(dir)
+    parent = ::File.dirname(dir)
+    begin
+      ::File.rename(dir, ::File.join(parent, "#{TRASH_DIR_PREFIX}#{::SecureRandom.hex(8)}"))
+    rescue ::SystemCallError
+      # Some filesystems (notably on Windows) refuse to rename a directory
+      # that has open handles under it. Fall back to deleting in place.
+      unless remove_dir_in_place(dir)
+        raise ::GitCache::Error, "Unable to remove directory: #{dir}"
+      end
+    end
+    sweep_trash(parent)
+  end
+
+  # Deletes a directory in place, retrying because a concurrent writer can
+  # defeat a single pass. Returns whether the directory is gone.
+  #
+  def remove_dir_in_place(dir, attempts: 3)
+    attempts.times do
+      chmod_recursive("u+w", dir)
+      ::FileUtils.rm_rf(dir)
+      return true unless ::File.exist?(dir)
+    end
+    false
+  end
+
+  # Recursive chmod that tolerates entries disappearing while it runs. The
+  # force: option of FileUtils.chmod_R covers only the chmod of each entry,
+  # not the traversal that finds them, so a concurrent writer removing a
+  # directory mid-walk raises out of chmod_R despite it.
+  #
+  def chmod_recursive(mode, dir)
+    ::FileUtils.chmod_R(mode, dir, force: true)
+  rescue ::SystemCallError
+    nil
+  end
+
+  # Deletes any leftover trash directories in the given directory. Failures
+  # are ignored; the leftovers are harmless and we can try again next time.
+  #
+  def sweep_trash(dir)
+    ::Dir.children(dir).each do |child|
+      remove_dir_in_place(::File.join(dir, child)) if child.start_with?(TRASH_DIR_PREFIX)
+    end
+  rescue ::SystemCallError
+    nil
+  end
+
   def ensure_repo_base_dir(remote)
     dir = repo_base_dir_for(remote)
     ::FileUtils.mkdir_p(dir)
@@ -265,8 +334,7 @@ class GitCache
     ::FileUtils.mkdir_p(repo_dir)
     result = git(repo_dir, ["remote", "get-url", "origin"])
     unless result.success? && result.captured_out.strip == remote
-      ::FileUtils.chmod_R("u+w", repo_dir, force: true)
-      ::FileUtils.rm_rf(repo_dir)
+      remove_dir(repo_dir)
       ::FileUtils.mkdir_p(repo_dir)
       git(repo_dir, ["init"],
           error_message: "Unable to initialize git repository")
@@ -304,11 +372,11 @@ class GitCache
       if repo_lock.source_exists?(sha, path)
         ::GitCache.safe_join(source_path, path)
       else
-        ::FileUtils.chmod_R("u+w", source_path, force: true)
+        chmod_recursive("u+w", source_path)
         begin
           copy_from_repo(repo_path, source_path, sha, path)
         ensure
-          ::FileUtils.chmod_R("a-w", source_path, force: true) unless ::GitCache.sources_writable?
+          chmod_recursive("a-w", source_path) unless ::GitCache.sources_writable?
         end
       end
     repo_lock.access_source!(sha, path)

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: git_cache
 version: !ruby/object:Gem::Version
-  version: 0.1.1
+  version: 0.1.2
 platform: ruby
 authors:
 - Daniel Azuma
@@ -61,8 +61,8 @@ licenses:
 - MIT
 metadata:
   bug_tracker_uri: https://github.com/dazuma/git_cache/issues
-  changelog_uri: https://rubydoc.info/gems/git_cache/0.1.1/file/CHANGELOG.md
-  documentation_uri: https://rubydoc.info/gems/git_cache/0.1.1
+  changelog_uri: https://rubydoc.info/gems/git_cache/0.1.2/file/CHANGELOG.md
+  documentation_uri: https://rubydoc.info/gems/git_cache/0.1.2
   homepage_uri: https://github.com/dazuma/git_cache
 rdoc_options: []
 require_paths:
@@ -78,7 +78,7 @@ required_rubygems_version: !ruby/object:Gem::Requirement
     - !ruby/object:Gem::Version
       version: '0'
 requirements: []
-rubygems_version: 4.0.6
+rubygems_version: 4.0.16
 specification_version: 4
 summary: A local file system cache of data from git repositories.
 test_files: []