zeitwerk 2.6.1 → 2.6.7

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: cf4acabb8b402560402158aced0f30ae6bae5120dc65d719f4fb15c1385a2454
-  data.tar.gz: 0a297687455cf8c3924a64c5e36223c35b0f69712c807ee9bc70748f8c42ff37
+  metadata.gz: e6c7bf12c3a33195c57541b840ca6d7dbc21b19e0d7b91d8933ad37aa88b325d
+  data.tar.gz: 4b8dd86417dd633b78c02eada9ed6956e8e708397c988237aaddecb9644ba469
 SHA512:
-  metadata.gz: 1d6666a098ae430f452edcd8b5e458541f20b20c4ee5903c4d18fe19cdce5afe81d6311a1bff5f9412b18ae57274e9b13132f8d59d9189b0ccd6cf7711337d6e
-  data.tar.gz: a131561faa463e489f7e9c4898bc8e212b2f6814ba3ce634b7ef62e25467f38b88e4c63106e8d53ee933681e1f0dc196244d16f13c299d2ca0771ee4031655a1
+  metadata.gz: 4f0bc60d9914a9e815aed501a030f7e9bde7927f750f3836f96ac86a52c0699f4ca69456bd381845428f6b09adccaa085443f39bb0fdaa651437742e21d10af2
+  data.tar.gz: 3ba55f7bb24725d156cc4043697bfcd6092e2893f1aa71296965c82ae9ed66b79c30f853b535693b6b884d7d3d26cf1b122545e395ba11c92f65552e14116248

data/README.md

@@ -3,7 +3,7 @@
 
 
 [![Gem Version](https://img.shields.io/gem/v/zeitwerk.svg?style=for-the-badge)](https://rubygems.org/gems/zeitwerk)
-[![Build Status](https://img.shields.io/github/workflow/status/fxn/zeitwerk/CI?event=push&style=for-the-badge)](https://github.com/fxn/zeitwerk/actions?query=event%3Apush)
+[![Build Status](https://img.shields.io/github/actions/workflow/status/fxn/zeitwerk/ci.yml?branch=main&event=push&style=for-the-badge)](https://github.com/fxn/zeitwerk/actions/workflows/ci.yml?query=branch%3main)
 
 <!-- TOC -->
 
@@ -27,8 +27,14 @@
   - [Autoloading](#autoloading)
   - [Eager loading](#eager-loading)
     - [Eager load exclusions](#eager-load-exclusions)
+    - [Eager load directories](#eager-load-directories)
+    - [Eager load namespaces](#eager-load-namespaces)
+    - [Eager load namespaces shared by several loaders](#eager-load-namespaces-shared-by-several-loaders)
     - [Global eager load](#global-eager-load)
+  - [Loading individual files](#loading-individual-files)
   - [Reloading](#reloading)
+    - [Configuration and usage](#configuration-and-usage)
+    - [Thread-safety](#thread-safety)
   - [Inflection](#inflection)
     - [Zeitwerk::Inflector](#zeitwerkinflector)
     - [Zeitwerk::GemInflector](#zeitwerkgeminflector)
@@ -44,6 +50,7 @@
     - [Use case: Files that do not follow the conventions](#use-case-files-that-do-not-follow-the-conventions)
     - [Use case: The adapter pattern](#use-case-the-adapter-pattern)
     - [Use case: Test files mixed with implementation files](#use-case-test-files-mixed-with-implementation-files)
+  - [Shadowed files](#shadowed-files)
   - [Edge cases](#edge-cases)
   - [Beware of circular dependencies](#beware-of-circular-dependencies)
   - [Reopening third-party namespaces](#reopening-third-party-namespaces)
@@ -51,9 +58,6 @@
   - [Encodings](#encodings)
   - [Rules of thumb](#rules-of-thumb)
   - [Debuggers](#debuggers)
-    - [debug.rb](#debugrb)
-    - [Byebug](#byebug)
-    - [Break](#break)
 - [Pronunciation](#pronunciation)
 - [Supported Ruby versions](#supported-ruby-versions)
 - [Testing](#testing)
@@ -169,7 +173,7 @@ class HttpCrawler
 end
 ```
 
-The first example needs a custom [inflection](https://github.com/fxn/zeitwerk#inflection) rule:
+The first example needs a custom [inflection](#inflection) rule:
 
 ```ruby
 loader.inflector.inflect("max_retries" => "MAX_RETRIES")
@@ -206,7 +210,7 @@ serializers/user_serializer.rb -> UserSerializer
 <a id="markdown-custom-root-namespaces" name="custom-root-namespaces"></a>
 #### Custom root namespaces
 
-While `Object` is by far the most common root namespace, you can associate a different one to a particular root directory. The method `push_dir` accepts a class or module object in the optional `namespace` keyword argument.
+While `Object` is by far the most common root namespace, you can associate a different one to a particular root directory. The method `push_dir` accepts a non-anonymous class or module object in the optional `namespace` keyword argument.
 
 For example, given:
 
@@ -384,6 +388,12 @@ module MyGem
 end
 ```
 
+Due to technical reasons, the entry point of the gem has to be loaded with `Kernel#require`, which is the standard way to load a gem. Loading that file with `Kernel#load` or `Kernel#require_relative` won't generally work.
+
+`Zeitwerk::Loader.for_gem` is idempotent when invoked from the same file, to support gems that want to reload (unlikely).
+
+If the entry point of your gem lives in a subdirectory of `lib` because it is reopening a namespace defined somewhere else, please use the generic API to setup the loader, and make sure you check the section [_Reopening third-party namespaces_](#reopening-third-party-namespaces) down below.
+
 Loaders returned by `Zeitwerk::Loader.for_gem` issue warnings if `lib` has extra Ruby files or directories.
 
 For example, if the gem has Rails generators under `lib/generators`, by convention that directory defines a `Generators` Ruby module. If `generators` is just a container for non-autoloadable code and templates, not acting as a project namespace, you need to setup things accordingly.
@@ -400,10 +410,6 @@ Otherwise, there's a flag to say the extra stuff is OK:
 Zeitwerk::Loader.for_gem(warn_on_extra_files: false)
 ```
 
-This method is idempotent when invoked from the same file, to support gems that want to reload (unlikely).
-
-If the entry point of your gem lives in a subdirectory of `lib` because it is reopening a namespace defined somewhere else, please use the generic API to setup the loader, and make sure you check the section [_Reopening third-party namespaces_](https://github.com/fxn/zeitwerk#reopening-third-party-namespaces) down below.
-
 <a id="markdown-autoloading" name="autoloading"></a>
 ### Autoloading
 
@@ -438,10 +444,12 @@ loader.eager_load
 
 That skips [ignored files and directories](#ignoring-parts-of-the-project).
 
-In gems, the method needs to be invoked after the main namespace has been defined, as shown in [Synopsis](https://github.com/fxn/zeitwerk#synopsis).
+In gems, the method needs to be invoked after the main namespace has been defined, as shown in [Synopsis](#synopsis).
 
 Eager loading is synchronized and idempotent.
 
+Attempting to eager load without previously calling `setup` raises `Zeitwerk::SetupRequired`.
+
 <a id="markdown-eager-load-exclusions" name="eager-load-exclusions"></a>
 #### Eager load exclusions
 
@@ -464,6 +472,81 @@ Which may be handy if the project eager loads in the test suite to [ensure proje
 
 The `force` flag does not affect ignored files and directories, those are still ignored.
 
+<a id="markdown-eager-load-directories" name="eager-load-directories"></a>
+#### Eager load directories
+
+The method `Zeitwerk::Loader#eager_load_dir` eager loads a given directory, recursively:
+
+```ruby
+loader.eager_load_dir("#{__dir__}/custom_web_app/routes")
+```
+
+This is useful when the loader is not eager loading the entire project, but you still need some subtree to be loaded for things to function properly.
+
+Both strings and `Pathname` objects are supported as arguments. If the argument is not a directory managed by the receiver, the method raises `Zeitwerk::Error`.
+
+[Eager load exclusions](#eager-load-exclusions), [ignored files and directories](#ignoring-parts-of-the-project), and [shadowed files](#shadowed-files) are not eager loaded.
+
+`Zeitwerk::Loader#eager_load_dir` is idempotent, but compatible with reloading. If you eager load a directory and then reload, eager loading that directory will load its (current) contents again.
+
+The method checks if a regular eager load was already executed, in which case it returns fast.
+
+Nested root directories which are descendants of the argument are skipped. Those subtrees are considered to be conceptually apart.
+
+Attempting to eager load a directory without previously calling `setup` raises `Zeitwerk::SetupRequired`.
+
+<a id="markdown-eager-load-namespaces" name="eager-load-namespaces"></a>
+#### Eager load namespaces
+
+The method `Zeitwerk::Loader#eager_load_namespace` eager loads a given namespace, recursively:
+
+```ruby
+loader.eager_load_namespace(MyApp::Routes)
+```
+
+This is useful when the loader is not eager loading the entire project, but you still need some namespace to be loaded for things to function properly.
+
+The argument has to be a class or module object and the method raises `Zeitwerk::Error` otherwise.
+
+If the namespace is spread over multiple directories in the receiver's source tree, they are all eager loaded. For example, if you have a structure like
+
+```
+root_dir1/my_app/routes
+root_dir2/my_app/routes
+root_dir3/my_app/routes
+```
+
+where `root_directory{1,2,3}` are root directories, eager loading `MyApp::Routes` will eager load the contents of the three corresponding directories.
+
+There might exist external source trees implementing part of the namespace. This happens routinely, because top-level constants are stored in the globally shared `Object`. It happens also when deliberately [reopening third-party namespaces](#reopening-third-party-namespaces). Such external code is not eager loaded, the implementation is carefully scoped to what the receiver manages to avoid side-effects elsewhere.
+
+This method is flexible about what it accepts. Its semantics have to be interpreted as: "_If_ you manage this namespace, or part of this namespace, please eager load what you got". In particular, if the receiver does not manage the namespace, it will simply do nothing, this is not an error condition.
+
+[Eager load exclusions](#eager-load-exclusions), [ignored files and directories](#ignoring-parts-of-the-project), and [shadowed files](#shadowed-files) are not eager loaded.
+
+`Zeitwerk::Loader#eager_load_namespace` is idempotent, but compatible with reloading. If you eager load a namespace and then reload, eager loading that namespace will load its (current) descendants again.
+
+The method checks if a regular eager load was already executed, in which case it returns fast.
+
+If root directories are assigned to custom namespaces, the method behaves as you'd expect, according to the namespacing relationship between the custom namespace and the argument.
+
+Attempting to eager load a namespace without previously calling `setup` raises `Zeitwerk::SetupRequired`.
+
+<a id="markdown-eager-load-namespaces-shared-by-several-loaders" name="eager-load-namespaces-shared-by-several-loaders"></a>
+#### Eager load namespaces shared by several loaders
+
+The method `Zeitwerk::Loader.eager_load_namespace` broadcasts `eager_load_namespace` to all loaders.
+
+```ruby
+Zeitwerk::Loader.eager_load_namespace(MyFramework::Routes)
+```
+
+This may be handy, for example, if a framework supports plugins and a shared namespace needs to be eager loaded for the framework to function properly.
+
+Please, note that loaders only eager load namespaces they manage, as documented above. Therefore, this method does not allow you to eager load namespaces not managed by Zeitwerk loaders.
+
+This method does not require that all registered loaders have `setup` already invoked, since that is out of your control. If there's any in that state, it is simply skipped.
+
 <a id="markdown-global-eager-load" name="global-eager-load"></a>
 #### Global eager load
 
@@ -479,9 +562,31 @@ Note that thanks to idempotence `Zeitwerk::Loader.eager_load_all` won't eager lo
 
 This method does not accept the `force` flag, since in general it wouldn't be a good idea to force eager loading in 3rd party code.
 
+This method does not require that all registered loaders have `setup` already invoked, since that is out of your control. If there's any in that state, it is simply skipped.
+
+<a id="markdown-loading-individual-files" name="loading-individual-files"></a>
+### Loading individual files
+
+The method `Zeitwerk::Loader#load_file` loads an individual Ruby file:
+
+```ruby
+loader.load_file("#{__dir__}/custom_web_app/routes.rb")
+```
+
+This is useful when the loader is not eager loading the entire project, but you still need an individual file to be loaded for things to function properly.
+
+Both strings and `Pathname` objects are supported as arguments. The method raises `Zeitwerk::Error` if the argument is not a Ruby file, is [ignored](#ignoring-parts-of-the-project), is [shadowed](#shadowed-files), or is not managed by the receiver.
+
+`Zeitwerk::Loader#load_file` is idempotent, but compatible with reloading. If you load a file and then reload, a new call will load its (current) contents again.
+
+If you want to eager load a directory, `Zeitwerk::Loader#eager_load_dir` is more efficient than invoking `Zeitwerk::Loader#load_file` on its files.
+
 <a id="markdown-reloading" name="reloading"></a>
 ### Reloading
 
+<a id="markdown-configuration-and-usage" name="configuration-and-usage"></a>
+#### Configuration and usage
+
 Zeitwerk is able to reload code, but you need to enable this feature:
 
 ```ruby
@@ -495,7 +600,7 @@ loader.reload
 
 There is no way to undo this, either you want to reload or you don't.
 
-Enabling reloading after setup raises `Zeitwerk::Error`. Attempting to reload without having it enabled raises `Zeitwerk::ReloadingDisabledError`.
+Enabling reloading after setup raises `Zeitwerk::Error`. Attempting to reload without having it enabled raises `Zeitwerk::ReloadingDisabledError`. Attempting to reload without previously calling `setup` raises `Zeitwerk::SetupRequired`.
 
 Generally speaking, reloading is useful while developing running services like web applications. Gems that implement regular libraries, so to speak, or services running in testing or production environments, won't normally have a use case for reloading. If reloading is not enabled, Zeitwerk is able to use less memory.
 
@@ -503,12 +608,34 @@ Reloading removes the currently loaded classes and modules and resets the loader
 
 It is important to highlight that this is an instance method. Don't worry about project dependencies managed by Zeitwerk, their loaders are independent.
 
-Reloading is not thread-safe:
+<a id="markdown-thread-safety" name="thread-safety"></a>
+#### Thread-safety
+
+In order to reload safely, no other thread can be autoloading or reloading concurrently. Client code is responsible for this coordination.
+
+For example, a web framework that serves each request in its own thread and has reloading enabled could create a read-write lock on boot like this:
+
+```ruby
+require "concurrent/atomic/read_write_lock"
+
+MyFramework::RELOAD_RW_LOCK = Concurrent::ReadWriteLock.new
+```
+
+You acquire the lock for reading for serving each individual request:
+
+```ruby
+MyFramework::RELOAD_RW_LOCK.with_read_lock do
+  serve(request)
+end
+```
 
-* You should not reload while another thread is reloading.
-* You should not autoload while another thread is reloading.
+Then, when a reload is triggered, just acquire the lock for writing in order to execute the method call safely:
 
-In order to reload in a thread-safe manner, frameworks need to implement some coordination. For example, a web framework that serves each request with its own thread may have a globally accessible read/write lock: When a request comes in, the framework acquires the lock for reading at the beginning, and releases it at the end. On the other hand, the code in the framework responsible for the call to `Zeitwerk::Loader#reload` needs to acquire the lock for writing.
+```ruby
+MyFramework::RELOAD_RW_LOCK.with_write_lock do
+  loader.reload
+end
+```
 
 On reloading, client code has to update anything that would otherwise be storing a stale object. For example, if the routing layer of a web framework stores reloadable controller class objects or instances in internal structures, on reload it has to refresh them somehow, possibly reevaluating routes.
 
@@ -823,6 +950,8 @@ However, sometimes it might still be convenient to tell Zeitwerk to completely i
 
 You can ignore file names, directory names, and glob patterns. Glob patterns are expanded when they are added and again on each reload.
 
+There is an edge case related to nested root directories. Conceptually, root directories are independent source trees. If you ignore a parent of a nested root directory, the nested root directory is not affected. You need to ignore it explictly if you want it ignored too.
+
 Let's see some use cases.
 
 <a id="markdown-use-case-files-that-do-not-follow-the-conventions" name="use-case-files-that-do-not-follow-the-conventions"></a>
@@ -905,10 +1034,39 @@ loader.ignore(tests)
 loader.setup
 ```
 
+<a id="markdown-shadowed-files" name="shadowed-files"></a>
+### Shadowed files
+
+In Ruby, if you have several files called `foo.rb` in different directories of `$LOAD_PATH` and execute
+
+```ruby
+require "foo"
+```
+
+the first one found gets loaded, and the rest are ignored.
+
+Zeitwerk behaves in a similar way. If `foo.rb` is present in several root directories (at the same namespace level), the constant `Foo` is autoloaded from the first one, and the rest of the files are not evaluated. If logging is enabled, you'll see something like
+
+```
+file #{file} is ignored because #{previous_occurrence} has precedence
+```
+
+(This message is not public interface and may change, you cannot rely on that exact wording.)
+
+Even if there's only one `foo.rb`, if the constant `Foo` is already defined when Zeitwerk finds `foo.rb`, then the file is ignored too. This could happen if `Foo` was defined by a dependency, for example. If logging is enabled, you'll see something like
+
+```
+file #{file} is ignored because #{constant_path} is already defined
+```
+
+(This message is not public interface and may change, you cannot rely on that exact wording.)
+
+Shadowing only applies to Ruby files, namespace definition can be spread over multiple directories. And you can also reopen third-party namespaces if done [orderly](#reopening-third-party-namespaces).
+
 <a id="markdown-edge-cases" name="edge-cases"></a>
 ### Edge cases
 
-A class or module that acts as a namespace:
+[Explicit namespaces](#explicit-namespaces) like `Trip` here:
 
 ```ruby
 # trip.rb
@@ -922,14 +1080,15 @@ module Trip::Geolocation
 end
 ```
 
-has to be defined with the `class` or `module` keywords, as in the example above.
+have to be defined with the `class`/`module` keywords, as in the example above.
 
 For technical reasons, raw constant assignment is not supported:
 
 ```ruby
 # trip.rb
-Trip = Class.new { ... }  # NOT SUPPORTED
-Trip = Struct.new { ... } # NOT SUPPORTED
+Trip = Class { ...}        # NOT SUPPORTED
+Trip = Struct.new { ... }  # NOT SUPPORTED
+Trip = Data.define { ... } # NOT SUPPORTED
 ```
 
 This only affects explicit namespaces, those idioms work well for any other ordinary class or module.
@@ -984,7 +1143,7 @@ require "active_job"
 require "active_job/queue_adapters"
 
 require "zeitwerk"
-# By passign the flag, we acknowledge the extra directory lib/active_job
+# By passing the flag, we acknowledge the extra directory lib/active_job
 # has to be managed by the loader and no warning has to be issued for it.
 loader = Zeitwerk::Loader.for_gem(warn_on_extra_files: false)
 loader.setup
@@ -1012,6 +1171,8 @@ loader.push_dir(Pathname.new("/bar"), namespace: Bar)
 loader.dirs(namespaces: true) # => { "/foo" => Object, "/bar" => Bar }
 ```
 
+By default, ignored root directories are filtered out. If you want them included, please pass `ignored: true`.
+
 These collections are read-only. Please add to them with `Zeitwerk::Loader#push_dir`.
 
 <a id="markdown-encodings" name="encodings"></a>
@@ -1044,22 +1205,9 @@ The test suite passes on Windows with codepage `Windows-1252` if all the involve
 <a id="markdown-debuggers" name="debuggers"></a>
 ### Debuggers
 
-<a id="markdown-debugrb" name="debugrb"></a>
-#### debug.rb
-
-The new [debug.rb](https://github.com/ruby/debug) gem and Zeitwerk are mostly compatible. This is the new debugger that is going to ship with Ruby 3.1.
-
-There's one exception, though: Due to a technical limitation of tracepoints, explicit namespaces are not autoloaded while expressions are evaluated in the REPL. See [ruby/debug#408](https://github.com/ruby/debug/issues/408).
-
-<a id="markdown-byebug" name="byebug"></a>
-#### Byebug
-
-Zeitwerk and [Byebug](https://github.com/deivid-rodriguez/byebug) have a similar edge incompatibility.
-
-<a id="markdown-break" name="break"></a>
-#### Break
+Zeitwerk works fine with [debug.rb](https://github.com/ruby/debug) and [Break](https://github.com/gsamokovarov/break).
 
-Zeitwerk works fine with [@gsamokovarov](https://github.com/gsamokovarov)'s [Break](https://github.com/gsamokovarov/break) debugger.
+[Byebug](https://github.com/deivid-rodriguez/byebug) is compatible except for an edge case explained in [deivid-rodriguez/byebug#564](https://github.com/deivid-rodriguez/byebug/issues/564).
 
 <a id="markdown-pronunciation" name="pronunciation"></a>
 ## Pronunciation

data/lib/zeitwerk/error.rb

@@ -12,4 +12,10 @@ module Zeitwerk
 
   class NameError < ::NameError
   end
+
+  class SetupRequired < Error
+    def initialize
+      super("please, finish your configuration and call Zeitwerk::Loader#setup once all is ready")
+    end
+  end
 end

data/lib/zeitwerk/explicit_namespace.rb

@@ -11,28 +11,28 @@ module Zeitwerk
   module ExplicitNamespace # :nodoc: all
     class << self
       include RealModName
+      extend Internal
 
       # Maps constant paths that correspond to explicit namespaces according to
       # the file system, to the loader responsible for them.
       #
-      # @private
       # @sig Hash[String, Zeitwerk::Loader]
       attr_reader :cpaths
+      private :cpaths
 
-      # @private
       # @sig Mutex
       attr_reader :mutex
+      private :mutex
 
-      # @private
       # @sig TracePoint
       attr_reader :tracer
+      private :tracer
 
       # Asserts `cpath` corresponds to an explicit namespace for which `loader`
       # is responsible.
       #
-      # @private
       # @sig (String, Zeitwerk::Loader) -> void
-      def register(cpath, loader)
+      internal def register(cpath, loader)
         mutex.synchronize do
           cpaths[cpath] = loader
           # We check enabled? because, looking at the C source code, enabling an
@@ -41,24 +41,28 @@ module Zeitwerk
         end
       end
 
-      # @private
       # @sig (Zeitwerk::Loader) -> void
-      def unregister_loader(loader)
+      internal def unregister_loader(loader)
         cpaths.delete_if { |_cpath, l| l == loader }
         disable_tracer_if_unneeded
       end
 
-      private
+      # This is an internal method only used by the test suite.
+      #
+      # @sig (String) -> bool
+      internal def registered?(cpath)
+        cpaths.key?(cpath)
+      end
 
       # @sig () -> void
-      def disable_tracer_if_unneeded
+      private def disable_tracer_if_unneeded
         mutex.synchronize do
           tracer.disable if cpaths.empty?
         end
       end
 
       # @sig (TracePoint) -> void
-      def tracepoint_class_callback(event)
+      private def tracepoint_class_callback(event)
         # If the class is a singleton class, we won't do anything with it so we
         # can bail out immediately. This is several orders of magnitude faster
         # than accessing its name.

data/lib/zeitwerk/gem_loader.rb

@@ -43,7 +43,7 @@ module Zeitwerk
         next if abspath == expected_namespace_dir
 
         basename_without_ext = basename.delete_suffix(".rb")
-        cname = inflector.camelize(basename_without_ext, abspath)
+        cname = inflector.camelize(basename_without_ext, abspath).to_sym
         ftype = dir?(abspath) ? "directory" : "file"
 
         warn(<<~EOS)

data/lib/zeitwerk/internal.rb

@@ -0,0 +1,12 @@
+# frozen_string_literal: true
+
+# This is a private module.
+module Zeitwerk::Internal
+  def internal(method_name)
+    private method_name
+
+    mangled = "__#{method_name}"
+    alias_method mangled, method_name
+    public mangled
+  end
+end

data/lib/zeitwerk/kernel.rb

@@ -19,6 +19,9 @@ module Kernel
   # included in Object, and changes in ancestors don't get propagated into
   # already existing ancestor chains on Ruby < 3.0.
   alias_method :zeitwerk_original_require, :require
+  class << self
+    alias_method :zeitwerk_original_require, :require
+  end
 
   # @sig (String) -> true | false
   def require(path)

data/lib/zeitwerk/loader/callbacks.rb

@@ -11,18 +11,18 @@ module Zeitwerk::Loader::Callbacks
     cref  = autoloads.delete(file)
     cpath = cpath(*cref)
 
-    # If reloading is enabled, we need to put this constant for unloading
-    # regardless of what cdef? says. In Ruby < 3.1 the internal state is not
-    # fully cleared. Module#constants still includes it, and you need to
-    # remove_const. See https://github.com/ruby/ruby/pull/4715.
-    to_unload[cpath] = [file, cref] if reloading_enabled?
     Zeitwerk::Registry.unregister_autoload(file)
 
     if cdef?(*cref)
       log("constant #{cpath} loaded from file #{file}") if logger
+      to_unload[cpath] = [file, cref] if reloading_enabled?
       run_on_load_callbacks(cpath, cget(*cref), file) unless on_load_callbacks.empty?
     else
-      raise Zeitwerk::NameError.new("expected file #{file} to define constant #{cpath}, but didn't", cref.last)
+      msg = "expected file #{file} to define constant #{cpath}, but didn't"
+      log(msg) if logger
+      crem(*cref)
+      to_unload[cpath] = [file, cref] if reloading_enabled?
+      raise Zeitwerk::NameError.new(msg, cref.last)
     end
   end
 
@@ -42,7 +42,7 @@ module Zeitwerk::Loader::Callbacks
     # Without the mutex and subsequent delete call, t2 would reset the module.
     # That not only would reassign the constant (undesirable per se) but, worse,
     # the module object created by t2 wouldn't have any of the autoloads for its
-    # children, since t1 would have correctly deleted its lazy_subdirs entry.
+    # children, since t1 would have correctly deleted its namespace_dirs entry.
     mutex2.synchronize do
       if cref = autoloads.delete(dir)
         autovivified_module = cref[0].const_set(cref[1], Module.new)
@@ -71,9 +71,9 @@ module Zeitwerk::Loader::Callbacks
   # @private
   # @sig (Module) -> void
   def on_namespace_loaded(namespace)
-    if subdirs = lazy_subdirs.delete(real_mod_name(namespace))
-      subdirs.each do |subdir|
-        set_autoloads_in_dir(subdir, namespace)
+    if dirs = namespace_dirs.delete(real_mod_name(namespace))
+      dirs.each do |dir|
+        set_autoloads_in_dir(dir, namespace)
       end
     end
   end