diff options
Diffstat (limited to 'guides')
-rw-r--r-- | guides/source/5_2_release_notes.md | 2 | ||||
-rw-r--r-- | guides/source/6_0_release_notes.md | 10 | ||||
-rw-r--r-- | guides/source/action_view_overview.md | 28 | ||||
-rw-r--r-- | guides/source/active_record_multiple_databases.md | 2 | ||||
-rw-r--r-- | guides/source/active_storage_overview.md | 6 | ||||
-rw-r--r-- | guides/source/active_support_core_extensions.md | 13 | ||||
-rw-r--r-- | guides/source/autoloading_and_reloading_constants.md | 1376 | ||||
-rw-r--r-- | guides/source/autoloading_and_reloading_constants_classic_mode.md | 1351 | ||||
-rw-r--r-- | guides/source/command_line.md | 2 | ||||
-rw-r--r-- | guides/source/configuring.md | 11 | ||||
-rw-r--r-- | guides/source/contributing_to_ruby_on_rails.md | 2 | ||||
-rw-r--r-- | guides/source/documents.yaml | 8 | ||||
-rw-r--r-- | guides/source/engines.md | 2 | ||||
-rw-r--r-- | guides/source/layouts_and_rendering.md | 46 | ||||
-rw-r--r-- | guides/source/rails_application_templates.md | 4 | ||||
-rw-r--r-- | guides/source/upgrading_ruby_on_rails.md | 96 |
16 files changed, 1649 insertions, 1310 deletions
diff --git a/guides/source/5_2_release_notes.md b/guides/source/5_2_release_notes.md index ac247bc3f9..7aac07dbbe 100644 --- a/guides/source/5_2_release_notes.md +++ b/guides/source/5_2_release_notes.md @@ -326,7 +326,7 @@ Please refer to the [Changelog][action-view] for detailed changes. select divider `option`. ([Pull Request](https://github.com/rails/rails/pull/31088)) -* Change `form_with` to generates ids by default. +* Change `form_with` to generate ids by default. ([Commit](https://github.com/rails/rails/commit/260d6f112a0ffdbe03e6f5051504cb441c1e94cd)) * Add `preload_link_tag` helper. diff --git a/guides/source/6_0_release_notes.md b/guides/source/6_0_release_notes.md index c152076628..7c5478a03d 100644 --- a/guides/source/6_0_release_notes.md +++ b/guides/source/6_0_release_notes.md @@ -672,6 +672,12 @@ Please refer to the [Changelog][active-storage] for detailed changes. is saved instead of immediately. ([Pull Request](https://github.com/rails/rails/pull/33303)) +* Optionally replace existing files instead of adding to them when assigning to + a collection of attachments (as in `@user.update!(images: [ … ])`). Use + `config.active_storage.replace_on_assign_to_many` to control this behavior. + ([Pull Request](https://github.com/rails/rails/pull/33303), + [Pull Request](https://github.com/rails/rails/pull/36716)) + * Add the ability to reflect on defined attachments using the existing Active Record reflection mechanism. ([Pull Request](https://github.com/rails/rails/pull/33018)) @@ -688,10 +694,6 @@ Please refer to the [Changelog][active-storage] for detailed changes. `mini_magick` directly. ([Pull Request](https://github.com/rails/rails/pull/32471)) -* Replace existing images instead of adding to them when updating an - attached model via `update` or `update!` with, say, `@user.update!(images: [ … ])`. - ([Pull Request](https://github.com/rails/rails/pull/33303)) - Active Model ------------ diff --git a/guides/source/action_view_overview.md b/guides/source/action_view_overview.md index a1b69edd22..dda3ae0863 100644 --- a/guides/source/action_view_overview.md +++ b/guides/source/action_view_overview.md @@ -1025,6 +1025,34 @@ If `@article.author_id` is 1, this would return: <label for="article_author_id_3">M. Clark</label> ``` +Recovering some option passed (e.g. programatically checking an object from collection): + +```ruby +collection_radio_buttons(:article, :author_id, Author.all, :id, :name_with_initial, {checked: Author.last}) +``` + +In this case, the last object from the collection will be checked: + +```html +<input id="article_author_id_1" name="article[author_id]" type="radio" value="1" /> +<label for="article_author_id_1">D. Heinemeier Hansson</label> +<input id="article_author_id_2" name="article[author_id]" type="radio" value="2" /> +<label for="article_author_id_2">D. Thomas</label> +<input id="article_author_id_3" name="article[author_id]" type="radio" value="3" checked="checked" /> +<label for="article_author_id_3">M. Clark</label> +``` + +To access the passed options programatically (e.g. adding a custom class if checked): + +**Sample html.erb** + +```html+erb +<%= collection_radio_buttons(:article, :author_id, Author.all, :id, :name_with_initial, {checked: Author.last, required: true} do |rb| %> + <%= rb.label(class: "#{'my-custom-class' if rb.value == Author.last.id}") { rb.radio_button + rb.text } %> +<% end %> +``` + + #### collection_check_boxes Returns `check_box` tags for the collection of existing return values of `method` for `object`'s class. diff --git a/guides/source/active_record_multiple_databases.md b/guides/source/active_record_multiple_databases.md index d7d9bd1ca9..07be21a254 100644 --- a/guides/source/active_record_multiple_databases.md +++ b/guides/source/active_record_multiple_databases.md @@ -180,7 +180,7 @@ to primary based on the HTTP verb and whether there was a recent write. If the application is receiving a POST, PUT, DELETE, or PATCH request the application will automatically write to the primary. For the specified time after the write the application -will read from the replica. For a GET or HEAD request the application will read from the +will read from the primary. For a GET or HEAD request the application will read from the replica unless there was a recent write. To activate the automatic connection switching middleware, add or uncomment the following diff --git a/guides/source/active_storage_overview.md b/guides/source/active_storage_overview.md index 932a5dc2e9..54f8f5c2b5 100644 --- a/guides/source/active_storage_overview.md +++ b/guides/source/active_storage_overview.md @@ -43,6 +43,8 @@ tables. Use `rails db:migrate` to run the migration. WARNING: `active_storage_attachments` is a polymorphic join table that stores your model's class name. If your model's class name changes, you will need to run a migration on this table to update the underlying `record_type` to your model's new class name. +WARNING: If you are using UUIDs instead of integers as the primary key on your models you will need to change the column type of `record_id` for the `active_storage_attachments` table in the generated migration accordingly. + Declare Active Storage services in `config/storage.yml`. For each service your application uses, provide a name and the requisite configuration. The example below declares three services named `local`, `test`, and `amazon`: @@ -398,6 +400,10 @@ helper allows you to set the disposition. rails_blob_path(user.avatar, disposition: "attachment") ``` +WARNING: To prevent XSS attacks, ActiveStorage forces the Content-Disposition header +to "attachment" for some kind of files. To change this behaviour see the +available configuration opions in [Configuring Rails Applications](configuring.html#configuring-active-storage). + If you need to create a link from outside of controller/view context (Background jobs, Cronjobs, etc.), you can access the rails_blob_path like this: diff --git a/guides/source/active_support_core_extensions.md b/guides/source/active_support_core_extensions.md index 8cb49ca6ae..f36cacfe8d 100644 --- a/guides/source/active_support_core_extensions.md +++ b/guides/source/active_support_core_extensions.md @@ -155,15 +155,6 @@ Complex(1).duplicable? # => true 1.method(:+).duplicable? # => false ``` -`duplicable?` matches the current Ruby version's `dup` behavior, -so results will vary according the version of Ruby you're using. -In Ruby 2.4, for example, Complex and Rational are not duplicable: - -```ruby -Rational(1).duplicable? # => false -Complex(1).duplicable? # => false -``` - WARNING: Any class can disallow duplication by removing `dup` and `clone` or raising exceptions from them. Thus only `rescue` can tell whether a given arbitrary object is duplicable. `duplicable?` depends on the hard-coded list above, but it is much faster than `rescue`. Use it only if you know the hard-coded list is enough in your use case. NOTE: Defined in `active_support/core_ext/object/duplicable.rb`. @@ -2358,10 +2349,6 @@ There's also a related idiom that uses the splat operator: [*object] ``` -which in Ruby 1.8 returns `[nil]` for `nil`, and calls to `Array(object)` otherwise. (Please if you know the exact behavior in 1.9 contact fxn.) - -Thus, in this case the behavior is different for `nil`, and the differences with `Kernel#Array` explained above apply to the rest of `object`s. - NOTE: Defined in `active_support/core_ext/array/wrap.rb`. ### Duplicating diff --git a/guides/source/autoloading_and_reloading_constants.md b/guides/source/autoloading_and_reloading_constants.md index 7dfc39e192..8cd2d353de 100644 --- a/guides/source/autoloading_and_reloading_constants.md +++ b/guides/source/autoloading_and_reloading_constants.md @@ -1,18 +1,18 @@ **DO NOT READ THIS FILE ON GITHUB, GUIDES ARE PUBLISHED ON https://guides.rubyonrails.org.** -Autoloading and Reloading Constants -=================================== +Autoloading and Reloading Constants (Zeitwerk Mode) +====================================================== -This guide documents how constant autoloading and reloading works. +This guide documents how autoloading and reloading works in `zeitwerk` mode. After reading this guide, you will know: -* Key aspects of Ruby constants -* What are the `autoload_paths` and how does eager loading work in production? -* How constant autoloading works -* What is `require_dependency` -* How constant reloading works -* Solutions to common autoloading gotchas +* Autoloading modes +* Related Rails configuration +* Project structure +* Autoloading, reloading, and eager loading +* Single Table Inheritance +* And more -------------------------------------------------------------------------------- @@ -20,13 +20,15 @@ After reading this guide, you will know: Introduction ------------ -Ruby on Rails allows applications to be written as if their code was preloaded. +INFO. This guide documents autoloading in `zeitwerk` mode, which is new in Rails 6. If you'd like to read about `classic` mode instead, please check [Autoloading and Reloading Constants (Classic Mode)](autoloading_and_reloading_constants_classic_mode.html). -In a normal Ruby program classes need to load their dependencies: +In a normal Ruby program, dependencies need to be loaded by hand. For example, the following controller uses classes `ApplicationController` and `Post`, and normally you'd need to put `require` calls for them: ```ruby -require 'application_controller' -require 'post' +# DO NOT DO THIS. +require "application_controller" +require "post" +# DO NOT DO THIS. class PostsController < ApplicationController def index @@ -35,16 +37,7 @@ class PostsController < ApplicationController end ``` -Our Rubyist instinct quickly sees some redundancy in there: If classes were -defined in files matching their name, couldn't their loading be automated -somehow? We could save scanning the file for dependencies, which is brittle. - -Moreover, `Kernel#require` loads files once, but development is much more smooth -if code gets refreshed when it changes without restarting the server. It would -be nice to be able to use `Kernel#load` in development, and `Kernel#require` in -production. - -Indeed, those features are provided by Ruby on Rails, where we just write +This is not the case in Rails applications, where application classes and modules are just available everywhere: ```ruby class PostsController < ApplicationController @@ -54,1330 +47,255 @@ class PostsController < ApplicationController end ``` -This guide documents how that works. - - -Constants Refresher -------------------- - -While constants are trivial in most programming languages, they are a rich -topic in Ruby. - -It is beyond the scope of this guide to document Ruby constants, but we are -nevertheless going to highlight a few key topics. Truly grasping the following -sections is instrumental to understanding constant autoloading and reloading. - -### Nesting - -Class and module definitions can be nested to create namespaces: - -```ruby -module XML - class SAXParser - # (1) - end -end -``` - -The *nesting* at any given place is the collection of enclosing nested class and -module objects outwards. The nesting at any given place can be inspected with -`Module.nesting`. For example, in the previous example, the nesting at -(1) is - -```ruby -[XML::SAXParser, XML] -``` - -It is important to understand that the nesting is composed of class and module -*objects*, it has nothing to do with the constants used to access them, and is -also unrelated to their names. - -For instance, while this definition is similar to the previous one: - -```ruby -class XML::SAXParser - # (2) -end -``` - -the nesting in (2) is different: - -```ruby -[XML::SAXParser] -``` - -`XML` does not belong to it. - -We can see in this example that the name of a class or module that belongs to a -certain nesting does not necessarily correlate with the namespaces at the spot. - -Even more, they are totally independent, take for instance - -```ruby -module X - module Y - end -end - -module A - module B - end -end - -module X::Y - module A::B - # (3) - end -end -``` - -The nesting in (3) consists of two module objects: - -```ruby -[A::B, X::Y] -``` - -So, it not only doesn't end in `A`, which does not even belong to the nesting, -but it also contains `X::Y`, which is independent from `A::B`. - -The nesting is an internal stack maintained by the interpreter, and it gets -modified according to these rules: - -* The class object following a `class` keyword gets pushed when its body is -executed, and popped after it. - -* The module object following a `module` keyword gets pushed when its body is -executed, and popped after it. - -* A singleton class opened with `class << object` gets pushed, and popped later. - -* When `instance_eval` is called using a string argument, -the singleton class of the receiver is pushed to the nesting of the eval'ed -code. When `class_eval` or `module_eval` is called using a string argument, -the receiver is pushed to the nesting of the eval'ed code. - -* The nesting at the top-level of code interpreted by `Kernel#load` is empty -unless the `load` call receives a true value as second argument, in which case -a newly created anonymous module is pushed by Ruby. - -It is interesting to observe that blocks do not modify the stack. In particular -the blocks that may be passed to `Class.new` and `Module.new` do not get the -class or module being defined pushed to their nesting. That's one of the -differences between defining classes and modules in one way or another. - -### Class and Module Definitions are Constant Assignments - -Let's suppose the following snippet creates a class (rather than reopening it): - -```ruby -class C -end -``` - -Ruby creates a constant `C` in `Object` and stores in that constant a class -object. The name of the class instance is "C", a string, named after the -constant. +Idiomatic Rails applications only issue `require` calls to load stuff from their `lib` directory, the Ruby standard library, Ruby gems, etc. That is, anything that does not belong to their autoload paths, explained below. -That is, -```ruby -class Project < ApplicationRecord -end -``` +Enabling Zeitwerk Mode +---------------------- -performs a constant assignment equivalent to +The autoloading `zeitwerk` mode is enabled by default in Rails 6 applications running on CRuby: ```ruby -Project = Class.new(ApplicationRecord) +# config/application.rb +config.load_defaults "6.x" # enables zeitwerk mode in CRuby ``` -including setting the name of the class as a side-effect: +In `zeitwerk` mode, Rails uses [Zeitwerk](https://github.com/fxn/zeitwerk) internally to autoload, reload, and eager load. Rails instantiates and configures a dedicated Zeitwerk instance that manages the project. -```ruby -Project.name # => "Project" -``` +INFO. You do not configure Zeitwerk manually in a Rails application. Rather, you configure the application using the portable configuration points explained in this guide, and Rails translates that to Zeitwerk on your behalf. -Constant assignment has a special rule to make that happen: if the object -being assigned is an anonymous class or module, Ruby sets the object's name to -the name of the constant. +Project Structure +----------------- -INFO. From then on, what happens to the constant and the instance does not -matter. For example, the constant could be deleted, the class object could be -assigned to a different constant, be stored in no constant anymore, etc. Once -the name is set, it doesn't change. +In a Rails application file names have to match the constants they define, with directories acting as namespaces. -Similarly, module creation using the `module` keyword as in +For example, the file `app/helpers/users_helper.rb` should define `UsersHelper` and the file `app/controllers/admin/payments_controller.rb` should define `Admin::PaymentsController`. -```ruby -module Admin -end -``` - -performs a constant assignment equivalent to +Rails configures Zeitwerk to inflect file names with `String#camelize`. For example, it expects that `app/controllers/users_controller.rb` defines the constant `UsersController` because ```ruby -Admin = Module.new +"users_controller".camelize # => UsersController ``` -including setting the name as a side-effect: +If you need to customize any of these inflections, for example to add an acronym, please have a look at `config/initializers/inflections.rb`. -```ruby -Admin.name # => "Admin" -``` +Please, check the [Zeitwerk documentation](https://github.com/fxn/zeitwerk#file-structure) for further details. -WARNING. The execution context of a block passed to `Class.new` or `Module.new` -is not entirely equivalent to the one of the body of the definitions using the -`class` and `module` keywords. But both idioms result in the same constant -assignment. - -Thus, an informal expression like "the `String` class" technically means the -class object stored in the constant called "String". That constant, in turn, -belongs to the class object stored in the constant called "Object". - -`String` is an ordinary constant, and everything related to them such as -resolution algorithms applies to it. - -Likewise, in the controller - -```ruby -class PostsController < ApplicationController - def index - @posts = Post.all - end -end -``` - -`Post` is not syntax for a class. Rather, `Post` is a regular Ruby constant. If -all is good, the constant is evaluated to an object that responds to `all`. +Autoload paths +-------------- -That is why we talk about *constant* autoloading, Rails has the ability to -load constants on the fly. +We call _autoload paths_ to the list of application directories whose contents are to be autoloaded. For example, `app/models`. Such directories represent the root namespace: `Object`. -### Constants are Stored in Modules +INFO. Autoload paths are called _root directories_ in Zeitwerk documentation, but we'll stay with "autoload path" in this guide. -Constants belong to modules in a very literal sense. Classes and modules have -a constant table; think of it as a hash table. +Within an autoload path, file names must match the constants they define as documented [here](https://github.com/fxn/zeitwerk#file-structure). -Let's analyze an example to really understand what that means. While common -abuses of language like "the `String` class" are convenient, the exposition is -going to be precise here for didactic purposes. +By default, the autoload paths of an application consist of all the subdirectories of `app` that exist when the application boots ---except for `aasets`, `javascripts`, `views`,--- plus the autoload paths of engines it might depend on. -Let's consider the following module definition: +For example, if `UsersHelper` is implemented in `app/helpers/users_helper.rb`, the module is autoloadable, you do not need (and should not write) a `require` call for it: -```ruby -module Colors - RED = '0xff0000' -end ``` - -First, when the `module` keyword is processed, the interpreter creates a new -entry in the constant table of the class object stored in the `Object` constant. -Said entry associates the name "Colors" to a newly created module object. -Furthermore, the interpreter sets the name of the new module object to be the -string "Colors". - -Later, when the body of the module definition is interpreted, a new entry is -created in the constant table of the module object stored in the `Colors` -constant. That entry maps the name "RED" to the string "0xff0000". - -In particular, `Colors::RED` is totally unrelated to any other `RED` constant -that may live in any other class or module object. If there were any, they -would have separate entries in their respective constant tables. - -Pay special attention in the previous paragraphs to the distinction between -class and module objects, constant names, and value objects associated to them -in constant tables. - -### Resolution Algorithms - -#### Resolution Algorithm for Relative Constants - -At any given place in the code, let's define *cref* to be the first element of -the nesting if it is not empty, or `Object` otherwise. - -Without getting too much into the details, the resolution algorithm for relative -constant references goes like this: - -1. If the nesting is not empty the constant is looked up in its elements and in -order. The ancestors of those elements are ignored. - -2. If not found, then the algorithm walks up the ancestor chain of the cref. - -3. If not found and the cref is a module, the constant is looked up in `Object`. - -4. If not found, `const_missing` is invoked on the cref. The default -implementation of `const_missing` raises `NameError`, but it can be overridden. - -Rails autoloading **does not emulate this algorithm**, but its starting point is -the name of the constant to be autoloaded, and the cref. See more in [Relative -References](#autoloading-algorithms-relative-references). - -#### Resolution Algorithm for Qualified Constants - -Qualified constants look like this: - -```ruby -Billing::Invoice +$ bin/rails runner 'p UsersHelper' +UsersHelper ``` -`Billing::Invoice` is composed of two constants: `Billing` is relative and is -resolved using the algorithm of the previous section. - -INFO. Leading colons would make the first segment absolute rather than -relative: `::Billing::Invoice`. That would force `Billing` to be looked up -only as a top-level constant. - -`Invoice` on the other hand is qualified by `Billing` and we are going to see -its resolution next. Let's define *parent* to be that qualifying class or module -object, that is, `Billing` in the example above. The algorithm for qualified -constants goes like this: - -1. The constant is looked up in the parent and its ancestors. In Ruby >= 2.5, -`Object` is skipped if present among the ancestors. `Kernel` and `BasicObject` -are still checked though. +Autoload paths automatically pick any custom directories under `app`. For example, if your application has `app/presenters`, or `app/services`, etc., they are added to autoload paths. -2. If the lookup fails, `const_missing` is invoked in the parent. The default -implementation of `const_missing` raises `NameError`, but it can be overridden. +The array of autoload paths can be extended by mutating `config.autoload_paths`, in `config/application.rb`, but nowadays this is discouraged. -INFO. In Ruby < 2.5 `String::Hash` evaluates to `Hash` and the interpreter -issues a warning: "toplevel constant Hash referenced by String::Hash". Starting -with 2.5, `String::Hash` raises `NameError` because `Object` is skipped. +WARNING. Please, do not mutate `ActiveSupport::Dependencies.autoload_paths`, the public interface to change autoload paths is `config.autoload_paths`. -As you see, this algorithm is simpler than the one for relative constants. In -particular, the nesting plays no role here, and modules are not special-cased, -if neither they nor their ancestors have the constants, `Object` is **not** -checked. -Rails autoloading **does not emulate this algorithm**, but its starting point is -the name of the constant to be autoloaded, and the parent. See more in -[Qualified References](#autoloading-algorithms-qualified-references). - - -Vocabulary +$LOAD_PATH ---------- -### Parent Namespaces - -Given a string with a constant path we define its *parent namespace* to be the -string that results from removing its rightmost segment. - -For example, the parent namespace of the string "A::B::C" is the string "A::B", -the parent namespace of "A::B" is "A", and the parent namespace of "A" is "". - -The interpretation of a parent namespace when thinking about classes and modules -is tricky though. Let's consider a module M named "A::B": - -* The parent namespace, "A", may not reflect nesting at a given spot. - -* The constant `A` may no longer exist, some code could have removed it from -`Object`. - -* If `A` exists, the class or module that was originally in `A` may not be there -anymore. For example, if after a constant removal there was another constant -assignment there would generally be a different object in there. - -* In such case, it could even happen that the reassigned `A` held a new class or -module called also "A"! - -* In the previous scenarios M would no longer be reachable through `A::B` but -the module object itself could still be alive somewhere and its name would -still be "A::B". - -The idea of a parent namespace is at the core of the autoloading algorithms -and helps explain and understand their motivation intuitively, but as you see -that metaphor leaks easily. Given an edge case to reason about, take always into -account that by "parent namespace" the guide means exactly that specific string -derivation. - -### Loading Mechanism - -Rails autoloads files with `Kernel#load` when `config.cache_classes` is false, -the default in development mode, and with `Kernel#require` otherwise, the -default in production mode. - -`Kernel#load` allows Rails to execute files more than once if [constant -reloading](#constant-reloading) is enabled. - -This guide uses the word "load" freely to mean a given file is interpreted, but -the actual mechanism can be `Kernel#load` or `Kernel#require` depending on that -flag. - - -Autoloading Availability ------------------------- - -Rails is always able to autoload provided its environment is in place. For -example the `runner` command autoloads: - -``` -$ rails runner 'p User.column_names' -["id", "email", "created_at", "updated_at"] -``` - -The console autoloads, the test suite autoloads, and of course the application -autoloads. - -By default, Rails eager loads the application files when it boots in production -mode, so most of the autoloading going on in development does not happen. But -autoloading may still be triggered during eager loading. - -For example, given - -```ruby -class BeachHouse < House -end -``` - -if `House` is still unknown when `app/models/beach_house.rb` is being eager -loaded, Rails autoloads it. - - -autoload_paths and eager_load_paths ------------------------------------ - -As you probably know, when `require` gets a relative file name: +Autoload paths are added to `$LOAD_PATH` by default. However, Zeitwerk uses absolute file names internally, and your application should not issue `require` calls for autoloadable files, so those directories are actually not needed there. You can opt-out with this flag: ```ruby -require 'erb' +config.add_autoload_paths_to_load_path = false ``` -Ruby looks for the file in the directories listed in `$LOAD_PATH`. That is, Ruby -iterates over all its directories and for each one of them checks whether they -have a file called "erb.rb", or "erb.so", or "erb.o", or "erb.dll". If it finds -any of them, the interpreter loads it and ends the search. Otherwise, it tries -again in the next directory of the list. If the list gets exhausted, `LoadError` -is raised. - -We are going to cover how constant autoloading works in more detail later, but -the idea is that when a constant like `Post` is hit and missing, if there's a -`post.rb` file for example in `app/models` Rails is going to find it, evaluate -it, and have `Post` defined as a side-effect. - -All right, Rails has a collection of directories similar to `$LOAD_PATH` in which -to look up `post.rb`. That collection is called `autoload_paths` and by -default it contains: - -* All subdirectories of `app` in the application and engines present at boot - time. For example, `app/controllers`. They do not need to be the default - ones, any custom directories like `app/workers` belong automatically to - `autoload_paths`. +That may speed legit `require` calls a bit, since there are less lookups. Also, if your application uses [Bootsnap](https://github.com/Shopify/bootsnap), that saves the library from building unnecessary indexes, and saves the RAM they would need. -* Any existing second level directories called `app/*/concerns` in the - application and engines. -* The directory `test/mailers/previews`. +Reloading +--------- -`eager_load_paths` is initially the `app` paths above +Rails automatically reloads classes and modules if application files change. -How files are autoloaded depends on `eager_load` and `cache_classes` config settings which typically vary in development, production, and test modes: +More precisely, if the web server is running and application files have been modified, Rails unloads all autoloaded constants just before the next request is processed. That way, application classes or modules used during that request are going to be autoloaded, thus picking up their current implementation in the file system. - * In **development**, you want quicker startup with incremental loading of application code. So `eager_load` should be set to `false`, and Rails will autoload files as needed (see [Autoloading Algorithms](#autoloading-algorithms) below) -- and then reload them when they change (see [Constant Reloading](#constant-reloading) below). - * In **production**, however, you want consistency and thread-safety and can live with a longer boot time. So `eager_load` is set to `true`, and then during boot (before the app is ready to receive requests) Rails loads all files in the `eager_load_paths` and then turns off auto loading (NB: autoloading may be needed during eager loading). Not autoloading after boot is a `good thing`, as autoloading can cause the app to be have thread-safety problems. - * In **test**, for speed of execution (of individual tests) `eager_load` is `false`, so Rails follows development behaviour. +Reloading can be enabled or disabled. The setting that controls this behavior is `config.cache_classes`, which is false by default in `development` mode (reloading enabled), and true by default in `production` mode (reloading disabled). -What is described above are the defaults with a newly generated Rails app. There are multiple ways this can be configured differently (see [Configuring Rails Applications](configuring.html#rails-general-configuration). -). But using `autoload_paths` on its own in the past (before Rails 5) developers might configure `autoload_paths` to add in extra locations (e.g. `lib` which used to be an autoload path list years ago, but no longer is). However this is now discouraged for most purposes, as it is likely to lead to production-only errors. It is possible to add new locations to both `config.eager_load_paths` and `config.autoload_paths` but use at your own risk. +Rails detects files have changed using an evented file monitor (default), or walking the autoload paths, depending on `config.file_watcher`. -See also [Autoloading in the Test Environment](#autoloading-in-the-test-environment). +In a Rails console there is no file watcher active regardless of the value of `config.cache_classes`. This is so because, normally, it would be confusing to have code reloaded in the middle of a console session, the same way you generally want an individual request to be served by a consistent, non-changing set of application classes and modules. -`config.autoload_paths` is not changeable from environment-specific configuration files. - -The value of `autoload_paths` can be inspected. In a just-generated application -it is (edited): +However, you can force a reload in the console executing `reload!`: ``` -$ rails r 'puts ActiveSupport::Dependencies.autoload_paths' -.../app/assets -.../app/channels -.../app/controllers -.../app/controllers/concerns -.../app/helpers -.../app/jobs -.../app/mailers -.../app/models -.../app/models/concerns -.../activestorage/app/assets -.../activestorage/app/controllers -.../activestorage/app/javascript -.../activestorage/app/jobs -.../activestorage/app/models -.../actioncable/app/assets -.../actionview/app/assets -.../test/mailers/previews +$ bin/rails c +Loading development environment (Rails 6.0.0) +irb(main):001:0> User.object_id +=> 70136277390120 +irb(main):002:0> reload! +Reloading... +=> true +irb(main):003:0> User.object_id +=> 70136284426020 ``` -INFO. `autoload_paths` is computed and cached during the initialization process. -The application needs to be restarted to reflect any changes in the directory -structure. +as you can see, the class object stored in the `User` constant is different after reloading. +### Reloading and Stale Objects -Autoloading Algorithms ----------------------- +It is very important to understand that Ruby does not have a way to truly reload classes and modules in memory, and have that reflected everywhere they are already used. Technically, "unloading" the `User` class means removing the `User` constant via `Object.send(:remove_const, "User")`. -### Relative References +Therefore, if you store a reloadable class or module object in a place that is not reloaded, that value is going to become stale. -A relative constant reference may appear in several places, for example, in +For example, if an initializer stores and caches a certain class object ```ruby -class PostsController < ApplicationController - def index - @posts = Post.all - end -end +# config/initializers/configure_payment_gateway.rb +# DO NOT DO THIS. +$PAYMENT_GATEWAY = Rails.env.production? ? RealGateway : MockedGateway +# DO NOT DO THIS. ``` -all three constant references are relative. - -#### Constants after the `class` and `module` Keywords - -Ruby performs a lookup for the constant that follows a `class` or `module` -keyword because it needs to know if the class or module is going to be created -or reopened. - -If the constant is not defined at that point it is not considered to be a -missing constant, autoloading is **not** triggered. - -So, in the previous example, if `PostsController` is not defined when the file -is interpreted Rails autoloading is not going to be triggered, Ruby will just -define the controller. - -#### Top-Level Constants - -On the contrary, if `ApplicationController` is unknown, the constant is -considered missing and an autoload is going to be attempted by Rails. - -In order to load `ApplicationController`, Rails iterates over `autoload_paths`. -First it checks if `app/assets/application_controller.rb` exists. If it does not, -which is normally the case, it continues and finds -`app/controllers/application_controller.rb`. +and `MockedGateway` gets reloaded, `$PAYMENT_GATEWAY` still stores the class object `MockedGateway` evaluated to when the initializer ran. Reloading does not change the class object stored in `$PAYMENT_GATEWAY`. -If the file defines the constant `ApplicationController` all is fine, otherwise -`LoadError` is raised: +Similarly, in the Rails console, if you have a user instance and reload: ``` -unable to autoload constant ApplicationController, expected -<full path to application_controller.rb> to define it (LoadError) +> user = User.new +> reload! ``` -INFO. Rails does not require the value of autoloaded constants to be a class or -module object. For example, if the file `app/models/max_clients.rb` defines -`MAX_CLIENTS = 100` autoloading `MAX_CLIENTS` works just fine. - -#### Namespaces +the `user` object is instance of a stale class object. Ruby gives you a new class if you evaluate `User` again, but does not update the class `user` is instance of. -Autoloading `ApplicationController` looks directly under the directories of -`autoload_paths` because the nesting in that spot is empty. The situation of -`Post` is different, the nesting in that line is `[PostsController]` and support -for namespaces comes into play. - -The basic idea is that given +Another use case of this gotcha is subclassing reloadable classes in a place that is not reloaded: ```ruby -module Admin - class BaseController < ApplicationController - @@all_roles = Role.all - end +# lib/vip_user.rb +class VipUser < User end ``` -to autoload `Role` we are going to check if it is defined in the current or -parent namespaces, one at a time. So, conceptually we want to try to autoload -any of - -``` -Admin::BaseController::Role -Admin::Role -Role -``` - -in that order. That's the idea. To do so, Rails looks in `autoload_paths` -respectively for file names like these: - -``` -admin/base_controller/role.rb -admin/role.rb -role.rb -``` - -modulus some additional directory lookups we are going to cover soon. - -INFO. `'Constant::Name'.underscore` gives the relative path without extension of -the file name where `Constant::Name` is expected to be defined. - -Let's see how Rails autoloads the `Post` constant in the `PostsController` -above assuming the application has a `Post` model defined in -`app/models/post.rb`. - -First it checks for `posts_controller/post.rb` in `autoload_paths`: - -``` -app/assets/posts_controller/post.rb -app/controllers/posts_controller/post.rb -app/helpers/posts_controller/post.rb -... -test/mailers/previews/posts_controller/post.rb -``` - -Since the lookup is exhausted without success, a similar search for a directory -is performed, we are going to see why in the [next section](#automatic-modules): +if `User` is reloaded, since `VipUser` is not, the superclass of `VipUser` is the original stale class object. -``` -app/assets/posts_controller/post -app/controllers/posts_controller/post -app/helpers/posts_controller/post -... -test/mailers/previews/posts_controller/post -``` +Bottom line: **do not cache reloadable classes or modules**. -If all those attempts fail, then Rails starts the lookup again in the parent -namespace. In this case only the top-level remains: -``` -app/assets/post.rb -app/controllers/post.rb -app/helpers/post.rb -app/mailers/post.rb -app/models/post.rb -``` +Eager Loading +------------- -A matching file is found in `app/models/post.rb`. The lookup stops there and the -file is loaded. If the file actually defines `Post` all is fine, otherwise -`LoadError` is raised. +In production-like environments it is generally better to load all the application code when the application boots. Eager loading puts everything in memory ready to serve requests right away, and it is also [CoW](https://en.wikipedia.org/wiki/Copy-on-write)-friendly. -### Qualified References +Eager loading is controlled by the flag `config.eager_load`, which is enabled by default in `production` mode. -When a qualified constant is missing Rails does not look for it in the parent -namespaces. But there is a caveat: when a constant is missing, Rails is -unable to tell if the trigger was a relative reference or a qualified one. +The order in which files are eager loaded is undefined. -For example, consider +if the `Zeitwerk` constant is defined, Rails invokes `Zeitwerk::Loader.eager_load_all` regardless of the application autoloading mode. That ensures dependencies managed by Zeitwerk are eager loaded. -```ruby -module Admin - User -end -``` -and - -```ruby -Admin::User -``` - -If `User` is missing, in either case all Rails knows is that a constant called -"User" was missing in a module called "Admin". +Single Table Inheritance +------------------------ -If there is a top-level `User` Ruby would resolve it in the former example, but -wouldn't in the latter. In general, Rails does not emulate the Ruby constant -resolution algorithms, but in this case it tries using the following heuristic: +Single Table Inheritance is a feature that doesn't play well with lazy loading. Reason is, its API generally needs to be able to enumerate the STI hierarchy to work correctly, whereas lazy loading defers loading classes until they are referenced. You can't enumerate what you haven't referenced yet. -> If none of the parent namespaces of the class or module has the missing -> constant then Rails assumes the reference is relative. Otherwise qualified. +In a sense, applications need to eager load STI hierarchies regardless of the loading mode. -For example, if this code triggers autoloading +Of course, if the application eager loads on boot, that is already accomplished. When it does not, it is in practice enough to instantiate the existing types in the database, which in development or test modes is usually fine. One way to do that is to throw this module into the `lib` directory: ```ruby -Admin::User -``` - -and the `User` constant is already present in `Object`, it is not possible that -the situation is - -```ruby -module Admin - User -end -``` - -because otherwise Ruby would have resolved `User` and no autoloading would have -been triggered in the first place. Thus, Rails assumes a qualified reference and -considers the file `admin/user.rb` and directory `admin/user` to be the only -valid options. - -In practice, this works quite well as long as the nesting matches all parent -namespaces respectively and the constants that make the rule apply are known at -that time. - -However, autoloading happens on demand. If by chance the top-level `User` was -not yet loaded, then Rails assumes a relative reference by contract. - -Naming conflicts of this kind are rare in practice, but if one occurs, -`require_dependency` provides a solution by ensuring that the constant needed -to trigger the heuristic is defined in the conflicting place. - -### Automatic Modules - -When a module acts as a namespace, Rails does not require the application to -define a file for it, a directory matching the namespace is enough. +module StiPreload + unless Rails.application.config.eager_load + extend ActiveSupport::Concern -Suppose an application has a back office whose controllers are stored in -`app/controllers/admin`. If the `Admin` module is not yet loaded when -`Admin::UsersController` is hit, Rails needs first to autoload the constant -`Admin`. - -If `autoload_paths` has a file called `admin.rb` Rails is going to load that -one, but if there's no such file and a directory called `admin` is found, Rails -creates an empty module and assigns it to the `Admin` constant on the fly. - -### Generic Procedure - -Relative references are reported to be missing in the cref where they were hit, -and qualified references are reported to be missing in their parent (see -[Resolution Algorithm for Relative -Constants](#resolution-algorithm-for-relative-constants) at the beginning of -this guide for the definition of *cref*, and [Resolution Algorithm for Qualified -Constants](#resolution-algorithm-for-qualified-constants) for the definition of -*parent*). - -The procedure to autoload constant `C` in an arbitrary situation is as follows: - -``` -if the class or module in which C is missing is Object - let ns = '' -else - let M = the class or module in which C is missing - - if M is anonymous - let ns = '' - else - let ns = M.name - end -end - -loop do - # Look for a regular file. - for dir in autoload_paths - if the file "#{dir}/#{ns.underscore}/c.rb" exists - load/require "#{dir}/#{ns.underscore}/c.rb" - - if C is now defined - return - else - raise LoadError - end + included do + cattr_accessor :preloaded, instance_accessor: false end - end - # Look for an automatic module. - for dir in autoload_paths - if the directory "#{dir}/#{ns.underscore}/c" exists - if ns is an empty string - let C = Module.new in Object and return - else - let C = Module.new in ns.constantize and return + class_methods do + def descendants + preload_sti unless preloaded + super end - end - end - - if ns is empty - # We reached the top-level without finding the constant. - raise NameError - else - if C exists in any of the parent namespaces - # Qualified constants heuristic. - raise NameError - else - # Try again in the parent namespace. - let ns = the parent namespace of ns and retry - end - end -end -``` - - -require_dependency ------------------- - -Constant autoloading is triggered on demand and therefore code that uses a -certain constant may have it already defined or may trigger an autoload. That -depends on the execution path and it may vary between runs. - -There are times, however, in which you want to make sure a certain constant is -known when the execution reaches some code. `require_dependency` provides a way -to load a file using the current [loading mechanism](#loading-mechanism), and -keeping track of constants defined in that file as if they were autoloaded to -have them reloaded as needed. - -`require_dependency` is rarely needed, but see a couple of use cases in -[Autoloading and STI](#autoloading-and-sti) and [When Constants aren't -Triggered](#when-constants-aren-t-missed). - -WARNING. Unlike autoloading, `require_dependency` does not expect the file to -define any particular constant. Exploiting this behavior would be a bad practice -though, file and constant paths should match. - - -Constant Reloading ------------------- - -When `config.cache_classes` is false Rails is able to reload autoloaded -constants. -For example, if you're in a console session and edit some file behind the -scenes, the code can be reloaded with the `reload!` command: - -``` -> reload! -``` - -When the application runs, code is reloaded when something relevant to this -logic changes. In order to do that, Rails monitors a number of things: - -* `config/routes.rb`. - -* Locales. - -* Ruby files under `autoload_paths`. - -* `db/schema.rb` and `db/structure.sql`. - -If anything in there changes, there is a middleware that detects it and reloads -the code. - -Autoloading keeps track of autoloaded constants. Reloading is implemented by -removing them all from their respective classes and modules using -`Module#remove_const`. That way, when the code goes on, those constants are -going to be unknown again, and files reloaded on demand. - -INFO. This is an all-or-nothing operation, Rails does not attempt to reload only -what changed since dependencies between classes makes that really tricky. -Instead, everything is wiped. - - -Module#autoload isn't Involved ------------------------------- - -`Module#autoload` provides a lazy way to load constants that is fully integrated -with the Ruby constant lookup algorithms, dynamic constant API, etc. It is quite -transparent. - -Rails internals make extensive use of it to defer as much work as possible from -the boot process. But constant autoloading in Rails is **not** implemented with -`Module#autoload`. - -One possible implementation based on `Module#autoload` would be to walk the -application tree and issue `autoload` calls that map existing file names to -their conventional constant name. - -There are a number of reasons that prevent Rails from using that implementation. - -For example, `Module#autoload` is only capable of loading files using `require`, -so reloading would not be possible. Not only that, it uses an internal `require` -which is not `Kernel#require`. - -Then, it provides no way to remove declarations in case a file is deleted. If a -constant gets removed with `Module#remove_const` its `autoload` is not triggered -again. Also, it doesn't support qualified names, so files with namespaces should -be interpreted during the walk tree to install their own `autoload` calls, but -those files could have constant references not yet configured. - -An implementation based on `Module#autoload` would be awesome but, as you see, -at least as of today it is not possible. Constant autoloading in Rails is -implemented with `Module#const_missing`, and that's why it has its own contract, -documented in this guide. - - -Common Gotchas --------------- - -### Nesting and Qualified Constants - -Let's consider - -```ruby -module Admin - class UsersController < ApplicationController - def index - @users = User.all + # Constantizes all types present in the database. There might be more on + # disk, but that does not matter in practice as far as the STI API is + # concerned. + # + # Assumes store_full_sti_class is true, the default. + def preload_sti + types_in_db = \ + base_class. + select(inheritance_column). + distinct. + pluck(inheritance_column). + compact. + each(&:constantize) + + types_in_db.each do |type| + logger.debug("Preloading STI type #{type}") + type.constantize + end + + self.preloaded = true + end end end end ``` -and +and then include it in the STI root classes of your project: ```ruby -class Admin::UsersController < ApplicationController - def index - @users = User.all - end -end -``` - -To resolve `User` Ruby checks `Admin` in the former case, but it does not in -the latter because it does not belong to the nesting (see [Nesting](#nesting) -and [Resolution Algorithms](#resolution-algorithms)). - -Unfortunately Rails autoloading does not know the nesting in the spot where the -constant was missing and so it is not able to act as Ruby would. In particular, -`Admin::User` will get autoloaded in either case. - -Albeit qualified constants with `class` and `module` keywords may technically -work with autoloading in some cases, it is preferable to use relative constants -instead: +# app/models/shape.rb +require "sti_preload" -```ruby -module Admin - class UsersController < ApplicationController - def index - @users = User.all - end - end +class Shape < ApplicationRecord + include StiPreload # Only in the root class. end -``` -### Autoloading and STI - -Single Table Inheritance (STI) is a feature of Active Record that enables -storing a hierarchy of models in one single table. The API of such models is -aware of the hierarchy and encapsulates some common needs. For example, given -these classes: - -```ruby # app/models/polygon.rb -class Polygon < ApplicationRecord +class Polygon < Shape end # app/models/triangle.rb class Triangle < Polygon end - -# app/models/rectangle.rb -class Rectangle < Polygon -end -``` - -`Triangle.create` creates a row that represents a triangle, and -`Rectangle.create` creates a row that represents a rectangle. If `id` is the -ID of an existing record, `Polygon.find(id)` returns an object of the correct -type. - -Methods that operate on collections are also aware of the hierarchy. For -example, `Polygon.all` returns all the records of the table, because all -rectangles and triangles are polygons. Active Record takes care of returning -instances of their corresponding class in the result set. - -Types are autoloaded as needed. For example, if `Polygon.first` is a rectangle -and `Rectangle` has not yet been loaded, Active Record autoloads it and the -record is correctly instantiated. - -All good, but if instead of performing queries based on the root class we need -to work on some subclass, things get interesting. - -While working with `Polygon` you do not need to be aware of all its descendants, -because anything in the table is by definition a polygon, but when working with -subclasses Active Record needs to be able to enumerate the types it is looking -for. Let's see an example. - -`Rectangle.all` only loads rectangles by adding a type constraint to the query: - -```sql -SELECT "polygons".* FROM "polygons" -WHERE "polygons"."type" IN ("Rectangle") -``` - -Let's introduce now a subclass of `Rectangle`: - -```ruby -# app/models/square.rb -class Square < Rectangle -end -``` - -`Rectangle.all` should now return rectangles **and** squares: - -```sql -SELECT "polygons".* FROM "polygons" -WHERE "polygons"."type" IN ("Rectangle", "Square") -``` - -But there's a caveat here: How does Active Record know that the class `Square` -exists at all? - -Even if the file `app/models/square.rb` exists and defines the `Square` class, -if no code yet used that class, `Rectangle.all` issues the query - -```sql -SELECT "polygons".* FROM "polygons" -WHERE "polygons"."type" IN ("Rectangle") -``` - -That is not a bug, the query includes all *known* descendants of `Rectangle`. - -A way to ensure this works correctly regardless of the order of execution is to -manually load the direct subclasses at the bottom of the file that defines each -intermediate class: - -```ruby -# app/models/rectangle.rb -class Rectangle < Polygon -end -require_dependency 'square' -``` - -This needs to happen for every intermediate (non-root and non-leaf) class. The -root class does not scope the query by type, and therefore does not necessarily -have to know all its descendants. - -### Autoloading and `require` - -Files defining constants to be autoloaded should never be `require`d: - -```ruby -require 'user' # DO NOT DO THIS - -class UsersController < ApplicationController - ... -end -``` - -There are two possible gotchas here in development mode: - -1. If `User` is autoloaded before reaching the `require`, `app/models/user.rb` -runs again because `load` does not update `$LOADED_FEATURES`. - -2. If the `require` runs first Rails does not mark `User` as an autoloaded -constant and changes to `app/models/user.rb` aren't reloaded. - -Just follow the flow and use constant autoloading always, never mix -autoloading and `require`. As a last resort, if some file absolutely needs to -load a certain file use `require_dependency` to play nice with constant -autoloading. This option is rarely needed in practice, though. - -Of course, using `require` in autoloaded files to load ordinary 3rd party -libraries is fine, and Rails is able to distinguish their constants, they are -not marked as autoloaded. - -### Autoloading and Initializers - -Consider this assignment in `config/initializers/set_auth_service.rb`: - -```ruby -AUTH_SERVICE = if Rails.env.production? - RealAuthService -else - MockedAuthService -end -``` - -The purpose of this setup would be that the application uses the class that -corresponds to the environment via `AUTH_SERVICE`. In development mode -`MockedAuthService` gets autoloaded when the initializer runs. Let's suppose -we do some requests, change its implementation, and hit the application again. -To our surprise the changes are not reflected. Why? - -As [we saw earlier](#constant-reloading), Rails removes autoloaded constants, -but `AUTH_SERVICE` stores the original class object. Stale, non-reachable -using the original constant, but perfectly functional. - -The following code summarizes the situation: - -```ruby -class C - def quack - 'quack!' - end -end - -X = C -Object.instance_eval { remove_const(:C) } -X.new.quack # => quack! -X.name # => C -C # => uninitialized constant C (NameError) -``` - -Because of that, it is not a good idea to autoload constants on application -initialization. - -In the case above we could implement a dynamic access point: - -```ruby -# app/models/auth_service.rb -class AuthService - if Rails.env.production? - def self.instance - RealAuthService - end - else - def self.instance - MockedAuthService - end - end -end -``` - -and have the application use `AuthService.instance` instead. `AuthService` -would be loaded on demand and be autoload-friendly. - -### `require_dependency` and Initializers - -As we saw before, `require_dependency` loads files in an autoloading-friendly -way. Normally, though, such a call does not make sense in an initializer. - -One could think about doing some [`require_dependency`](#require-dependency) -calls in an initializer to make sure certain constants are loaded upfront, for -example as an attempt to address the [gotcha with STIs](#autoloading-and-sti). - -Problem is, in development mode [autoloaded constants are wiped](#constant-reloading) -if there is any relevant change in the file system. If that happens then -we are in the very same situation the initializer wanted to avoid! - -Calls to `require_dependency` have to be strategically written in autoloaded -spots. - -### When Constants aren't Missed - -#### Relative References - -Let's consider a flight simulator. The application has a default flight model - -```ruby -# app/models/flight_model.rb -class FlightModel -end -``` - -that can be overridden by each airplane, for instance - -```ruby -# app/models/bell_x1/flight_model.rb -module BellX1 - class FlightModel < FlightModel - end -end - -# app/models/bell_x1/aircraft.rb -module BellX1 - class Aircraft - def initialize - @flight_model = FlightModel.new - end - end -end -``` - -The initializer wants to create a `BellX1::FlightModel` and nesting has -`BellX1`, that looks good. But if the default flight model is loaded and the -one for the Bell-X1 is not, the interpreter is able to resolve the top-level -`FlightModel` and autoloading is thus not triggered for `BellX1::FlightModel`. - -That code depends on the execution path. - -These kind of ambiguities can often be resolved using qualified constants: - -```ruby -module BellX1 - class Plane - def flight_model - @flight_model ||= BellX1::FlightModel.new - end - end -end -``` - -Also, `require_dependency` is a solution: - -```ruby -require_dependency 'bell_x1/flight_model' - -module BellX1 - class Plane - def flight_model - @flight_model ||= FlightModel.new - end - end -end -``` - -#### Qualified References - -WARNING. This gotcha is only possible in Ruby < 2.5. - -Given - -```ruby -# app/models/hotel.rb -class Hotel -end - -# app/models/image.rb -class Image -end - -# app/models/hotel/image.rb -class Hotel - class Image < Image - end -end -``` - -the expression `Hotel::Image` is ambiguous because it depends on the execution -path. - -As [we saw before](#resolution-algorithm-for-qualified-constants), Ruby looks -up the constant in `Hotel` and its ancestors. If `app/models/image.rb` has -been loaded but `app/models/hotel/image.rb` hasn't, Ruby does not find `Image` -in `Hotel`, but it does in `Object`: - -``` -$ rails r 'Image; p Hotel::Image' 2>/dev/null -Image # NOT Hotel::Image! -``` - -The code evaluating `Hotel::Image` needs to make sure -`app/models/hotel/image.rb` has been loaded, possibly with -`require_dependency`. - -In these cases the interpreter issues a warning though: - -``` -warning: toplevel constant Image referenced by Hotel::Image -``` - -This surprising constant resolution can be observed with any qualifying class: - -``` -2.1.5 :001 > String::Array -(irb):1: warning: toplevel constant Array referenced by String::Array - => Array -``` - -WARNING. To find this gotcha the qualifying namespace has to be a class, -`Object` is not an ancestor of modules. - -### Autoloading within Singleton Classes - -Let's suppose we have these class definitions: - -```ruby -# app/models/hotel/services.rb -module Hotel - class Services - end -end - -# app/models/hotel/geo_location.rb -module Hotel - class GeoLocation - class << self - Services - end - end -end ``` -If `Hotel::Services` is known by the time `app/models/hotel/geo_location.rb` -is being loaded, `Services` is resolved by Ruby because `Hotel` belongs to the -nesting when the singleton class of `Hotel::GeoLocation` is opened. +Rails.autoloaders +----------------- -But if `Hotel::Services` is not known, Rails is not able to autoload it, the -application raises `NameError`. - -The reason is that autoloading is triggered for the singleton class, which is -anonymous, and as [we saw before](#generic-procedure), Rails only checks the -top-level namespace in that edge case. - -An easy solution to this caveat is to qualify the constant: +The Zeitwerk instances managing your application are availabe at ```ruby -module Hotel - class GeoLocation - class << self - Hotel::Services - end - end -end +Rails.autoloaders.main +Rails.autoloaders.once ``` -### Autoloading in `BasicObject` - -Direct descendants of `BasicObject` do not have `Object` among their ancestors -and cannot resolve top-level constants: - -```ruby -class C < BasicObject - String # NameError: uninitialized constant C::String -end -``` +The former is the main one. The latter is there mostly for backwards compatibily reasons, in case the application has something in `config.autoload_once_paths` (this is discouraged nowadays). -When autoloading is involved that plot has a twist. Let's consider: +You can check if `zeitwerk` mode is enabled with ```ruby -class C < BasicObject - def user - User # WRONG - end -end +Rails.autoloaders.zeitwerk_enabled? ``` -Since Rails checks the top-level namespace `User` gets autoloaded just fine the -first time the `user` method is invoked. You only get the exception if the -`User` constant is known at that point, in particular in a *second* call to -`user`: - -```ruby -c = C.new -c.user # surprisingly fine, User -c.user # NameError: uninitialized constant C::User -``` - -because it detects that a parent namespace already has the constant (see [Qualified -References](#autoloading-algorithms-qualified-references)). - -As with pure Ruby, within the body of a direct descendant of `BasicObject` use -always absolute constant paths: - -```ruby -class C < BasicObject - ::String # RIGHT - - def user - ::User # RIGHT - end -end -``` - -### Autoloading in the Test Environment - -When configuring the `test` environment for autoloading you might consider multiple factors. - -For example it might be worth running your tests with an identical setup to production (`config.eager_load = true`, `config.cache_classes = true`) in order to catch any problems before they hit production (this is compensation for the lack of dev-prod parity). However this will slow down the boot time for individual tests on a dev machine (and is not immediately compatible with spring see below). So one possibility is to do this on a -[CI](https://en.wikipedia.org/wiki/Continuous_integration) machine only (which should run without spring). - -On a development machine you can then have your tests running with whatever is fastest (ideally `config.eager_load = false`). - -With the [Spring](https://github.com/rails/spring) pre-loader (included with new Rails apps), you ideally keep `config.eager_load = false` as per development. Sometimes you may end up with a hybrid configuration (`config.eager_load = true`, `config.cache_classes = true` AND `config.enable_dependency_loading = true`), see [spring issue](https://github.com/rails/spring/issues/519#issuecomment-348324369). However it might be simpler to keep the same configuration as development, and work out whatever it is that is causing autoloading to fail (perhaps by the results of your CI test results). - -Occasionally you may need to explicitly eager_load by using `Rails -.application.eager_load!` in the setup of your tests -- this might occur if your [tests involve multithreading](https://stackoverflow.com/questions/25796409/in-rails-how-can-i-eager-load-all-code-before-a-specific-rspec-test). - -## Troubleshooting - -### Tracing Autoloads - -Active Support is able to report constants as they are autoloaded. To enable these traces in a Rails application, put the following two lines in some initializer: - -```ruby -ActiveSupport::Dependencies.logger = Rails.logger -ActiveSupport::Dependencies.verbose = true -``` - -### Where is a Given Autoload Triggered? +Opting Out +---------- -If constant `Foo` is being autoloaded, and you'd like to know where is that autoload coming from, just throw +Applications can load Rails 6 defaults and still use the classic autoloader this way: ```ruby -puts caller +# config/application.rb +config.load_defaults "6.x" +config.autoloader = :classic ``` -at the top of `foo.rb` and inspect the printed stack trace. - -### Which Constants Have Been Autoloaded? - -At any given time, - -```ruby -ActiveSupport::Dependencies.autoloaded_constants -``` +That may be handy if upgrading to Rails 6 in different phases, but classic mode is discouraged for new applications. -has the collection of constants that have been autoloaded so far. +`zeitwerk` mode is not available in versions of Rails previous to 6.0. diff --git a/guides/source/autoloading_and_reloading_constants_classic_mode.md b/guides/source/autoloading_and_reloading_constants_classic_mode.md new file mode 100644 index 0000000000..d0d9e076d6 --- /dev/null +++ b/guides/source/autoloading_and_reloading_constants_classic_mode.md @@ -0,0 +1,1351 @@ +**DO NOT READ THIS FILE ON GITHUB, GUIDES ARE PUBLISHED ON https://guides.rubyonrails.org.** + +Autoloading and Reloading Constants (Classic Mode) +================================================== + +This guide documents how constant autoloading and reloading works in `classic` mode. + +After reading this guide, you will know: + +* Key aspects of Ruby constants +* What are the `autoload_paths` and how does eager loading work in production? +* How constant autoloading works +* What is `require_dependency` +* How constant reloading works +* Solutions to common autoloading gotchas + +-------------------------------------------------------------------------------- + + +Introduction +------------ + +INFO. This guide documents autoloading in `classic` mode, which is the traditional one. If you'd like to read about `zeiwerk` mode instead, the new one in Rails 6, please check [Autoloading and Reloading Constants (Zeitwerk Mode)](autoloading_and_reloading_constants.html). + +Ruby on Rails allows applications to be written as if their code was preloaded. + +In a normal Ruby program classes need to load their dependencies: + +```ruby +require 'application_controller' +require 'post' + +class PostsController < ApplicationController + def index + @posts = Post.all + end +end +``` + +Our Rubyist instinct quickly sees some redundancy in there: If classes were +defined in files matching their name, couldn't their loading be automated +somehow? We could save scanning the file for dependencies, which is brittle. + +Moreover, `Kernel#require` loads files once, but development is much more smooth +if code gets refreshed when it changes without restarting the server. It would +be nice to be able to use `Kernel#load` in development, and `Kernel#require` in +production. + +Indeed, those features are provided by Ruby on Rails, where we just write + +```ruby +class PostsController < ApplicationController + def index + @posts = Post.all + end +end +``` + +This guide documents how that works. + + +Constants Refresher +------------------- + +While constants are trivial in most programming languages, they are a rich +topic in Ruby. + +It is beyond the scope of this guide to document Ruby constants, but we are +nevertheless going to highlight a few key topics. Truly grasping the following +sections is instrumental to understanding constant autoloading and reloading. + +### Nesting + +Class and module definitions can be nested to create namespaces: + +```ruby +module XML + class SAXParser + # (1) + end +end +``` + +The *nesting* at any given place is the collection of enclosing nested class and +module objects outwards. The nesting at any given place can be inspected with +`Module.nesting`. For example, in the previous example, the nesting at +(1) is + +```ruby +[XML::SAXParser, XML] +``` + +It is important to understand that the nesting is composed of class and module +*objects*, it has nothing to do with the constants used to access them, and is +also unrelated to their names. + +For instance, while this definition is similar to the previous one: + +```ruby +class XML::SAXParser + # (2) +end +``` + +the nesting in (2) is different: + +```ruby +[XML::SAXParser] +``` + +`XML` does not belong to it. + +We can see in this example that the name of a class or module that belongs to a +certain nesting does not necessarily correlate with the namespaces at the spot. + +Even more, they are totally independent, take for instance + +```ruby +module X + module Y + end +end + +module A + module B + end +end + +module X::Y + module A::B + # (3) + end +end +``` + +The nesting in (3) consists of two module objects: + +```ruby +[A::B, X::Y] +``` + +So, it not only doesn't end in `A`, which does not even belong to the nesting, +but it also contains `X::Y`, which is independent from `A::B`. + +The nesting is an internal stack maintained by the interpreter, and it gets +modified according to these rules: + +* The class object following a `class` keyword gets pushed when its body is +executed, and popped after it. + +* The module object following a `module` keyword gets pushed when its body is +executed, and popped after it. + +* A singleton class opened with `class << object` gets pushed, and popped later. + +* When `instance_eval` is called using a string argument, +the singleton class of the receiver is pushed to the nesting of the eval'ed +code. When `class_eval` or `module_eval` is called using a string argument, +the receiver is pushed to the nesting of the eval'ed code. + +* The nesting at the top-level of code interpreted by `Kernel#load` is empty +unless the `load` call receives a true value as second argument, in which case +a newly created anonymous module is pushed by Ruby. + +It is interesting to observe that blocks do not modify the stack. In particular +the blocks that may be passed to `Class.new` and `Module.new` do not get the +class or module being defined pushed to their nesting. That's one of the +differences between defining classes and modules in one way or another. + +### Class and Module Definitions are Constant Assignments + +Let's suppose the following snippet creates a class (rather than reopening it): + +```ruby +class C +end +``` + +Ruby creates a constant `C` in `Object` and stores in that constant a class +object. The name of the class instance is "C", a string, named after the +constant. + +That is, + +```ruby +class Project < ApplicationRecord +end +``` + +performs a constant assignment equivalent to + +```ruby +Project = Class.new(ApplicationRecord) +``` + +including setting the name of the class as a side-effect: + +```ruby +Project.name # => "Project" +``` + +Constant assignment has a special rule to make that happen: if the object +being assigned is an anonymous class or module, Ruby sets the object's name to +the name of the constant. + +INFO. From then on, what happens to the constant and the instance does not +matter. For example, the constant could be deleted, the class object could be +assigned to a different constant, be stored in no constant anymore, etc. Once +the name is set, it doesn't change. + +Similarly, module creation using the `module` keyword as in + +```ruby +module Admin +end +``` + +performs a constant assignment equivalent to + +```ruby +Admin = Module.new +``` + +including setting the name as a side-effect: + +```ruby +Admin.name # => "Admin" +``` + +WARNING. The execution context of a block passed to `Class.new` or `Module.new` +is not entirely equivalent to the one of the body of the definitions using the +`class` and `module` keywords. But both idioms result in the same constant +assignment. + +Thus, an informal expression like "the `String` class" technically means the +class object stored in the constant called "String". That constant, in turn, +belongs to the class object stored in the constant called "Object". + +`String` is an ordinary constant, and everything related to them such as +resolution algorithms applies to it. + +Likewise, in the controller + +```ruby +class PostsController < ApplicationController + def index + @posts = Post.all + end +end +``` + +`Post` is not syntax for a class. Rather, `Post` is a regular Ruby constant. If +all is good, the constant is evaluated to an object that responds to `all`. + +That is why we talk about *constant* autoloading, Rails has the ability to +load constants on the fly. + +### Constants are Stored in Modules + +Constants belong to modules in a very literal sense. Classes and modules have +a constant table; think of it as a hash table. + +Let's analyze an example to really understand what that means. While common +abuses of language like "the `String` class" are convenient, the exposition is +going to be precise here for didactic purposes. + +Let's consider the following module definition: + +```ruby +module Colors + RED = '0xff0000' +end +``` + +First, when the `module` keyword is processed, the interpreter creates a new +entry in the constant table of the class object stored in the `Object` constant. +Said entry associates the name "Colors" to a newly created module object. +Furthermore, the interpreter sets the name of the new module object to be the +string "Colors". + +Later, when the body of the module definition is interpreted, a new entry is +created in the constant table of the module object stored in the `Colors` +constant. That entry maps the name "RED" to the string "0xff0000". + +In particular, `Colors::RED` is totally unrelated to any other `RED` constant +that may live in any other class or module object. If there were any, they +would have separate entries in their respective constant tables. + +Pay special attention in the previous paragraphs to the distinction between +class and module objects, constant names, and value objects associated to them +in constant tables. + +### Resolution Algorithms + +#### Resolution Algorithm for Relative Constants + +At any given place in the code, let's define *cref* to be the first element of +the nesting if it is not empty, or `Object` otherwise. + +Without getting too much into the details, the resolution algorithm for relative +constant references goes like this: + +1. If the nesting is not empty the constant is looked up in its elements and in +order. The ancestors of those elements are ignored. + +2. If not found, then the algorithm walks up the ancestor chain of the cref. + +3. If not found and the cref is a module, the constant is looked up in `Object`. + +4. If not found, `const_missing` is invoked on the cref. The default +implementation of `const_missing` raises `NameError`, but it can be overridden. + +Rails autoloading **does not emulate this algorithm**, but its starting point is +the name of the constant to be autoloaded, and the cref. See more in [Relative +References](#autoloading-algorithms-relative-references). + +#### Resolution Algorithm for Qualified Constants + +Qualified constants look like this: + +```ruby +Billing::Invoice +``` + +`Billing::Invoice` is composed of two constants: `Billing` is relative and is +resolved using the algorithm of the previous section. + +INFO. Leading colons would make the first segment absolute rather than +relative: `::Billing::Invoice`. That would force `Billing` to be looked up +only as a top-level constant. + +`Invoice` on the other hand is qualified by `Billing` and we are going to see +its resolution next. Let's define *parent* to be that qualifying class or module +object, that is, `Billing` in the example above. The algorithm for qualified +constants goes like this: + +1. The constant is looked up in the parent and its ancestors. In Ruby >= 2.5, +`Object` is skipped if present among the ancestors. `Kernel` and `BasicObject` +are still checked though. + +2. If the lookup fails, `const_missing` is invoked in the parent. The default +implementation of `const_missing` raises `NameError`, but it can be overridden. + +INFO. In Ruby < 2.5 `String::Hash` evaluates to `Hash` and the interpreter +issues a warning: "toplevel constant Hash referenced by String::Hash". Starting +with 2.5, `String::Hash` raises `NameError` because `Object` is skipped. + +As you see, this algorithm is simpler than the one for relative constants. In +particular, the nesting plays no role here, and modules are not special-cased, +if neither they nor their ancestors have the constants, `Object` is **not** +checked. + +Rails autoloading **does not emulate this algorithm**, but its starting point is +the name of the constant to be autoloaded, and the parent. See more in +[Qualified References](#autoloading-algorithms-qualified-references). + + +Vocabulary +---------- + +### Parent Namespaces + +Given a string with a constant path we define its *parent namespace* to be the +string that results from removing its rightmost segment. + +For example, the parent namespace of the string "A::B::C" is the string "A::B", +the parent namespace of "A::B" is "A", and the parent namespace of "A" is "". + +The interpretation of a parent namespace when thinking about classes and modules +is tricky though. Let's consider a module M named "A::B": + +* The parent namespace, "A", may not reflect nesting at a given spot. + +* The constant `A` may no longer exist, some code could have removed it from +`Object`. + +* If `A` exists, the class or module that was originally in `A` may not be there +anymore. For example, if after a constant removal there was another constant +assignment there would generally be a different object in there. + +* In such case, it could even happen that the reassigned `A` held a new class or +module called also "A"! + +* In the previous scenarios M would no longer be reachable through `A::B` but +the module object itself could still be alive somewhere and its name would +still be "A::B". + +The idea of a parent namespace is at the core of the autoloading algorithms +and helps explain and understand their motivation intuitively, but as you see +that metaphor leaks easily. Given an edge case to reason about, take always into +account that by "parent namespace" the guide means exactly that specific string +derivation. + +### Loading Mechanism + +Rails autoloads files with `Kernel#load` when `config.cache_classes` is false, +the default in development mode, and with `Kernel#require` otherwise, the +default in production mode. + +`Kernel#load` allows Rails to execute files more than once if [constant +reloading](#constant-reloading) is enabled. + +This guide uses the word "load" freely to mean a given file is interpreted, but +the actual mechanism can be `Kernel#load` or `Kernel#require` depending on that +flag. + + +Autoloading Availability +------------------------ + +Rails is always able to autoload provided its environment is in place. For +example the `runner` command autoloads: + +``` +$ rails runner 'p User.column_names' +["id", "email", "created_at", "updated_at"] +``` + +The console autoloads, the test suite autoloads, and of course the application +autoloads. + +By default, Rails eager loads the application files when it boots in production +mode, so most of the autoloading going on in development does not happen. But +autoloading may still be triggered during eager loading. + +For example, given + +```ruby +class BeachHouse < House +end +``` + +if `House` is still unknown when `app/models/beach_house.rb` is being eager +loaded, Rails autoloads it. + + +autoload_paths and eager_load_paths +----------------------------------- + +As you probably know, when `require` gets a relative file name: + +```ruby +require 'erb' +``` + +Ruby looks for the file in the directories listed in `$LOAD_PATH`. That is, Ruby +iterates over all its directories and for each one of them checks whether they +have a file called "erb.rb", or "erb.so", or "erb.o", or "erb.dll". If it finds +any of them, the interpreter loads it and ends the search. Otherwise, it tries +again in the next directory of the list. If the list gets exhausted, `LoadError` +is raised. + +We are going to cover how constant autoloading works in more detail later, but +the idea is that when a constant like `Post` is hit and missing, if there's a +`post.rb` file for example in `app/models` Rails is going to find it, evaluate +it, and have `Post` defined as a side-effect. + +All right, Rails has a collection of directories similar to `$LOAD_PATH` in which +to look up `post.rb`. That collection is called `autoload_paths` and by +default it contains: + +* All subdirectories of `app` in the application and engines present at boot + time. For example, `app/controllers`. They do not need to be the default + ones, any custom directories like `app/workers` belong automatically to + `autoload_paths`. + +* Any existing second level directories called `app/*/concerns` in the + application and engines. + +* The directory `test/mailers/previews`. + +`eager_load_paths` is initially the `app` paths above + +How files are autoloaded depends on `eager_load` and `cache_classes` config settings which typically vary in development, production, and test modes: + + * In **development**, you want quicker startup with incremental loading of application code. So `eager_load` should be set to `false`, and Rails will autoload files as needed (see [Autoloading Algorithms](#autoloading-algorithms) below) -- and then reload them when they change (see [Constant Reloading](#constant-reloading) below). + * In **production**, however, you want consistency and thread-safety and can live with a longer boot time. So `eager_load` is set to `true`, and then during boot (before the app is ready to receive requests) Rails loads all files in the `eager_load_paths` and then turns off auto loading (NB: autoloading may be needed during eager loading). Not autoloading after boot is a `good thing`, as autoloading can cause the app to be have thread-safety problems. + * In **test**, for speed of execution (of individual tests) `eager_load` is `false`, so Rails follows development behaviour. + +What is described above are the defaults with a newly generated Rails app. There are multiple ways this can be configured differently (see [Configuring Rails Applications](configuring.html#rails-general-configuration). +). But using `autoload_paths` on its own in the past (before Rails 5) developers might configure `autoload_paths` to add in extra locations (e.g. `lib` which used to be an autoload path list years ago, but no longer is). However this is now discouraged for most purposes, as it is likely to lead to production-only errors. It is possible to add new locations to both `config.eager_load_paths` and `config.autoload_paths` but use at your own risk. + +See also [Autoloading in the Test Environment](#autoloading-in-the-test-environment). + +`config.autoload_paths` is not changeable from environment-specific configuration files. + +The value of `autoload_paths` can be inspected. In a just-generated application +it is (edited): + +``` +$ rails r 'puts ActiveSupport::Dependencies.autoload_paths' +.../app/assets +.../app/channels +.../app/controllers +.../app/controllers/concerns +.../app/helpers +.../app/jobs +.../app/mailers +.../app/models +.../app/models/concerns +.../activestorage/app/assets +.../activestorage/app/controllers +.../activestorage/app/javascript +.../activestorage/app/jobs +.../activestorage/app/models +.../actioncable/app/assets +.../actionview/app/assets +.../test/mailers/previews +``` + +INFO. `autoload_paths` is computed and cached during the initialization process. +The application needs to be restarted to reflect any changes in the directory +structure. + + +Autoloading Algorithms +---------------------- + +### Relative References + +A relative constant reference may appear in several places, for example, in + +```ruby +class PostsController < ApplicationController + def index + @posts = Post.all + end +end +``` + +all three constant references are relative. + +#### Constants after the `class` and `module` Keywords + +Ruby performs a lookup for the constant that follows a `class` or `module` +keyword because it needs to know if the class or module is going to be created +or reopened. + +If the constant is not defined at that point it is not considered to be a +missing constant, autoloading is **not** triggered. + +So, in the previous example, if `PostsController` is not defined when the file +is interpreted Rails autoloading is not going to be triggered, Ruby will just +define the controller. + +#### Top-Level Constants + +On the contrary, if `ApplicationController` is unknown, the constant is +considered missing and an autoload is going to be attempted by Rails. + +In order to load `ApplicationController`, Rails iterates over `autoload_paths`. +First it checks if `app/assets/application_controller.rb` exists. If it does not, +which is normally the case, it continues and finds +`app/controllers/application_controller.rb`. + +If the file defines the constant `ApplicationController` all is fine, otherwise +`LoadError` is raised: + +``` +unable to autoload constant ApplicationController, expected +<full path to application_controller.rb> to define it (LoadError) +``` + +INFO. Rails does not require the value of autoloaded constants to be a class or +module object. For example, if the file `app/models/max_clients.rb` defines +`MAX_CLIENTS = 100` autoloading `MAX_CLIENTS` works just fine. + +#### Namespaces + +Autoloading `ApplicationController` looks directly under the directories of +`autoload_paths` because the nesting in that spot is empty. The situation of +`Post` is different, the nesting in that line is `[PostsController]` and support +for namespaces comes into play. + +The basic idea is that given + +```ruby +module Admin + class BaseController < ApplicationController + @@all_roles = Role.all + end +end +``` + +to autoload `Role` we are going to check if it is defined in the current or +parent namespaces, one at a time. So, conceptually we want to try to autoload +any of + +``` +Admin::BaseController::Role +Admin::Role +Role +``` + +in that order. That's the idea. To do so, Rails looks in `autoload_paths` +respectively for file names like these: + +``` +admin/base_controller/role.rb +admin/role.rb +role.rb +``` + +modulus some additional directory lookups we are going to cover soon. + +INFO. `'Constant::Name'.underscore` gives the relative path without extension of +the file name where `Constant::Name` is expected to be defined. + +Let's see how Rails autoloads the `Post` constant in the `PostsController` +above assuming the application has a `Post` model defined in +`app/models/post.rb`. + +First it checks for `posts_controller/post.rb` in `autoload_paths`: + +``` +app/assets/posts_controller/post.rb +app/controllers/posts_controller/post.rb +app/helpers/posts_controller/post.rb +... +test/mailers/previews/posts_controller/post.rb +``` + +Since the lookup is exhausted without success, a similar search for a directory +is performed, we are going to see why in the [next section](#automatic-modules): + +``` +app/assets/posts_controller/post +app/controllers/posts_controller/post +app/helpers/posts_controller/post +... +test/mailers/previews/posts_controller/post +``` + +If all those attempts fail, then Rails starts the lookup again in the parent +namespace. In this case only the top-level remains: + +``` +app/assets/post.rb +app/controllers/post.rb +app/helpers/post.rb +app/mailers/post.rb +app/models/post.rb +``` + +A matching file is found in `app/models/post.rb`. The lookup stops there and the +file is loaded. If the file actually defines `Post` all is fine, otherwise +`LoadError` is raised. + +### Qualified References + +When a qualified constant is missing Rails does not look for it in the parent +namespaces. But there is a caveat: when a constant is missing, Rails is +unable to tell if the trigger was a relative reference or a qualified one. + +For example, consider + +```ruby +module Admin + User +end +``` + +and + +```ruby +Admin::User +``` + +If `User` is missing, in either case all Rails knows is that a constant called +"User" was missing in a module called "Admin". + +If there is a top-level `User` Ruby would resolve it in the former example, but +wouldn't in the latter. In general, Rails does not emulate the Ruby constant +resolution algorithms, but in this case it tries using the following heuristic: + +> If none of the parent namespaces of the class or module has the missing +> constant then Rails assumes the reference is relative. Otherwise qualified. + +For example, if this code triggers autoloading + +```ruby +Admin::User +``` + +and the `User` constant is already present in `Object`, it is not possible that +the situation is + +```ruby +module Admin + User +end +``` + +because otherwise Ruby would have resolved `User` and no autoloading would have +been triggered in the first place. Thus, Rails assumes a qualified reference and +considers the file `admin/user.rb` and directory `admin/user` to be the only +valid options. + +In practice, this works quite well as long as the nesting matches all parent +namespaces respectively and the constants that make the rule apply are known at +that time. + +However, autoloading happens on demand. If by chance the top-level `User` was +not yet loaded, then Rails assumes a relative reference by contract. + +Naming conflicts of this kind are rare in practice, but if one occurs, +`require_dependency` provides a solution by ensuring that the constant needed +to trigger the heuristic is defined in the conflicting place. + +### Automatic Modules + +When a module acts as a namespace, Rails does not require the application to +define a file for it, a directory matching the namespace is enough. + +Suppose an application has a back office whose controllers are stored in +`app/controllers/admin`. If the `Admin` module is not yet loaded when +`Admin::UsersController` is hit, Rails needs first to autoload the constant +`Admin`. + +If `autoload_paths` has a file called `admin.rb` Rails is going to load that +one, but if there's no such file and a directory called `admin` is found, Rails +creates an empty module and assigns it to the `Admin` constant on the fly. + +### Generic Procedure + +Relative references are reported to be missing in the cref where they were hit, +and qualified references are reported to be missing in their parent (see +[Resolution Algorithm for Relative +Constants](#resolution-algorithm-for-relative-constants) at the beginning of +this guide for the definition of *cref*, and [Resolution Algorithm for Qualified +Constants](#resolution-algorithm-for-qualified-constants) for the definition of +*parent*). + +The procedure to autoload constant `C` in an arbitrary situation is as follows: + +``` +if the class or module in which C is missing is Object + let ns = '' +else + let M = the class or module in which C is missing + + if M is anonymous + let ns = '' + else + let ns = M.name + end +end + +loop do + # Look for a regular file. + for dir in autoload_paths + if the file "#{dir}/#{ns.underscore}/c.rb" exists + load/require "#{dir}/#{ns.underscore}/c.rb" + + if C is now defined + return + else + raise LoadError + end + end + end + + # Look for an automatic module. + for dir in autoload_paths + if the directory "#{dir}/#{ns.underscore}/c" exists + if ns is an empty string + let C = Module.new in Object and return + else + let C = Module.new in ns.constantize and return + end + end + end + + if ns is empty + # We reached the top-level without finding the constant. + raise NameError + else + if C exists in any of the parent namespaces + # Qualified constants heuristic. + raise NameError + else + # Try again in the parent namespace. + let ns = the parent namespace of ns and retry + end + end +end +``` + + +require_dependency +------------------ + +Constant autoloading is triggered on demand and therefore code that uses a +certain constant may have it already defined or may trigger an autoload. That +depends on the execution path and it may vary between runs. + +There are times, however, in which you want to make sure a certain constant is +known when the execution reaches some code. `require_dependency` provides a way +to load a file using the current [loading mechanism](#loading-mechanism), and +keeping track of constants defined in that file as if they were autoloaded to +have them reloaded as needed. + +`require_dependency` is rarely needed, but see a couple of use cases in +[Autoloading and STI](#autoloading-and-sti) and [When Constants aren't +Triggered](#when-constants-aren-t-missed). + +WARNING. Unlike autoloading, `require_dependency` does not expect the file to +define any particular constant. Exploiting this behavior would be a bad practice +though, file and constant paths should match. + + +Constant Reloading +------------------ + +When `config.cache_classes` is false Rails is able to reload autoloaded +constants. + +For example, if you're in a console session and edit some file behind the +scenes, the code can be reloaded with the `reload!` command: + +``` +> reload! +``` + +When the application runs, code is reloaded when something relevant to this +logic changes. In order to do that, Rails monitors a number of things: + +* `config/routes.rb`. + +* Locales. + +* Ruby files under `autoload_paths`. + +* `db/schema.rb` and `db/structure.sql`. + +If anything in there changes, there is a middleware that detects it and reloads +the code. + +Autoloading keeps track of autoloaded constants. Reloading is implemented by +removing them all from their respective classes and modules using +`Module#remove_const`. That way, when the code goes on, those constants are +going to be unknown again, and files reloaded on demand. + +INFO. This is an all-or-nothing operation, Rails does not attempt to reload only +what changed since dependencies between classes makes that really tricky. +Instead, everything is wiped. + +Common Gotchas +-------------- + +### Nesting and Qualified Constants + +Let's consider + +```ruby +module Admin + class UsersController < ApplicationController + def index + @users = User.all + end + end +end +``` + +and + +```ruby +class Admin::UsersController < ApplicationController + def index + @users = User.all + end +end +``` + +To resolve `User` Ruby checks `Admin` in the former case, but it does not in +the latter because it does not belong to the nesting (see [Nesting](#nesting) +and [Resolution Algorithms](#resolution-algorithms)). + +Unfortunately Rails autoloading does not know the nesting in the spot where the +constant was missing and so it is not able to act as Ruby would. In particular, +`Admin::User` will get autoloaded in either case. + +Albeit qualified constants with `class` and `module` keywords may technically +work with autoloading in some cases, it is preferable to use relative constants +instead: + +```ruby +module Admin + class UsersController < ApplicationController + def index + @users = User.all + end + end +end +``` + +### Autoloading and STI + +Single Table Inheritance (STI) is a feature of Active Record that enables +storing a hierarchy of models in one single table. The API of such models is +aware of the hierarchy and encapsulates some common needs. For example, given +these classes: + +```ruby +# app/models/polygon.rb +class Polygon < ApplicationRecord +end + +# app/models/triangle.rb +class Triangle < Polygon +end + +# app/models/rectangle.rb +class Rectangle < Polygon +end +``` + +`Triangle.create` creates a row that represents a triangle, and +`Rectangle.create` creates a row that represents a rectangle. If `id` is the +ID of an existing record, `Polygon.find(id)` returns an object of the correct +type. + +Methods that operate on collections are also aware of the hierarchy. For +example, `Polygon.all` returns all the records of the table, because all +rectangles and triangles are polygons. Active Record takes care of returning +instances of their corresponding class in the result set. + +Types are autoloaded as needed. For example, if `Polygon.first` is a rectangle +and `Rectangle` has not yet been loaded, Active Record autoloads it and the +record is correctly instantiated. + +All good, but if instead of performing queries based on the root class we need +to work on some subclass, things get interesting. + +While working with `Polygon` you do not need to be aware of all its descendants, +because anything in the table is by definition a polygon, but when working with +subclasses Active Record needs to be able to enumerate the types it is looking +for. Let's see an example. + +`Rectangle.all` only loads rectangles by adding a type constraint to the query: + +```sql +SELECT "polygons".* FROM "polygons" +WHERE "polygons"."type" IN ("Rectangle") +``` + +Let's introduce now a subclass of `Rectangle`: + +```ruby +# app/models/square.rb +class Square < Rectangle +end +``` + +`Rectangle.all` should now return rectangles **and** squares: + +```sql +SELECT "polygons".* FROM "polygons" +WHERE "polygons"."type" IN ("Rectangle", "Square") +``` + +But there's a caveat here: How does Active Record know that the class `Square` +exists at all? + +Even if the file `app/models/square.rb` exists and defines the `Square` class, +if no code yet used that class, `Rectangle.all` issues the query + +```sql +SELECT "polygons".* FROM "polygons" +WHERE "polygons"."type" IN ("Rectangle") +``` + +That is not a bug, the query includes all *known* descendants of `Rectangle`. + +A way to ensure this works correctly regardless of the order of execution is to +manually load the direct subclasses at the bottom of the file that defines each +intermediate class: + +```ruby +# app/models/rectangle.rb +class Rectangle < Polygon +end +require_dependency 'square' +``` + +This needs to happen for every intermediate (non-root and non-leaf) class. The +root class does not scope the query by type, and therefore does not necessarily +have to know all its descendants. + +### Autoloading and `require` + +Files defining constants to be autoloaded should never be `require`d: + +```ruby +require 'user' # DO NOT DO THIS + +class UsersController < ApplicationController + ... +end +``` + +There are two possible gotchas here in development mode: + +1. If `User` is autoloaded before reaching the `require`, `app/models/user.rb` +runs again because `load` does not update `$LOADED_FEATURES`. + +2. If the `require` runs first Rails does not mark `User` as an autoloaded +constant and changes to `app/models/user.rb` aren't reloaded. + +Just follow the flow and use constant autoloading always, never mix +autoloading and `require`. As a last resort, if some file absolutely needs to +load a certain file use `require_dependency` to play nice with constant +autoloading. This option is rarely needed in practice, though. + +Of course, using `require` in autoloaded files to load ordinary 3rd party +libraries is fine, and Rails is able to distinguish their constants, they are +not marked as autoloaded. + +### Autoloading and Initializers + +Consider this assignment in `config/initializers/set_auth_service.rb`: + +```ruby +AUTH_SERVICE = if Rails.env.production? + RealAuthService +else + MockedAuthService +end +``` + +The purpose of this setup would be that the application uses the class that +corresponds to the environment via `AUTH_SERVICE`. In development mode +`MockedAuthService` gets autoloaded when the initializer runs. Let's suppose +we do some requests, change its implementation, and hit the application again. +To our surprise the changes are not reflected. Why? + +As [we saw earlier](#constant-reloading), Rails removes autoloaded constants, +but `AUTH_SERVICE` stores the original class object. Stale, non-reachable +using the original constant, but perfectly functional. + +The following code summarizes the situation: + +```ruby +class C + def quack + 'quack!' + end +end + +X = C +Object.instance_eval { remove_const(:C) } +X.new.quack # => quack! +X.name # => C +C # => uninitialized constant C (NameError) +``` + +Because of that, it is not a good idea to autoload constants on application +initialization. + +In the case above we could implement a dynamic access point: + +```ruby +# app/models/auth_service.rb +class AuthService + if Rails.env.production? + def self.instance + RealAuthService + end + else + def self.instance + MockedAuthService + end + end +end +``` + +and have the application use `AuthService.instance` instead. `AuthService` +would be loaded on demand and be autoload-friendly. + +### `require_dependency` and Initializers + +As we saw before, `require_dependency` loads files in an autoloading-friendly +way. Normally, though, such a call does not make sense in an initializer. + +One could think about doing some [`require_dependency`](#require-dependency) +calls in an initializer to make sure certain constants are loaded upfront, for +example as an attempt to address the [gotcha with STIs](#autoloading-and-sti). + +Problem is, in development mode [autoloaded constants are wiped](#constant-reloading) +if there is any relevant change in the file system. If that happens then +we are in the very same situation the initializer wanted to avoid! + +Calls to `require_dependency` have to be strategically written in autoloaded +spots. + +### When Constants aren't Missed + +#### Relative References + +Let's consider a flight simulator. The application has a default flight model + +```ruby +# app/models/flight_model.rb +class FlightModel +end +``` + +that can be overridden by each airplane, for instance + +```ruby +# app/models/bell_x1/flight_model.rb +module BellX1 + class FlightModel < FlightModel + end +end + +# app/models/bell_x1/aircraft.rb +module BellX1 + class Aircraft + def initialize + @flight_model = FlightModel.new + end + end +end +``` + +The initializer wants to create a `BellX1::FlightModel` and nesting has +`BellX1`, that looks good. But if the default flight model is loaded and the +one for the Bell-X1 is not, the interpreter is able to resolve the top-level +`FlightModel` and autoloading is thus not triggered for `BellX1::FlightModel`. + +That code depends on the execution path. + +These kind of ambiguities can often be resolved using qualified constants: + +```ruby +module BellX1 + class Plane + def flight_model + @flight_model ||= BellX1::FlightModel.new + end + end +end +``` + +Also, `require_dependency` is a solution: + +```ruby +require_dependency 'bell_x1/flight_model' + +module BellX1 + class Plane + def flight_model + @flight_model ||= FlightModel.new + end + end +end +``` + +#### Qualified References + +WARNING. This gotcha is only possible in Ruby < 2.5. + +Given + +```ruby +# app/models/hotel.rb +class Hotel +end + +# app/models/image.rb +class Image +end + +# app/models/hotel/image.rb +class Hotel + class Image < Image + end +end +``` + +the expression `Hotel::Image` is ambiguous because it depends on the execution +path. + +As [we saw before](#resolution-algorithm-for-qualified-constants), Ruby looks +up the constant in `Hotel` and its ancestors. If `app/models/image.rb` has +been loaded but `app/models/hotel/image.rb` hasn't, Ruby does not find `Image` +in `Hotel`, but it does in `Object`: + +``` +$ rails r 'Image; p Hotel::Image' 2>/dev/null +Image # NOT Hotel::Image! +``` + +The code evaluating `Hotel::Image` needs to make sure +`app/models/hotel/image.rb` has been loaded, possibly with +`require_dependency`. + +In these cases the interpreter issues a warning though: + +``` +warning: toplevel constant Image referenced by Hotel::Image +``` + +This surprising constant resolution can be observed with any qualifying class: + +``` +2.1.5 :001 > String::Array +(irb):1: warning: toplevel constant Array referenced by String::Array + => Array +``` + +WARNING. To find this gotcha the qualifying namespace has to be a class, +`Object` is not an ancestor of modules. + +### Autoloading within Singleton Classes + +Let's suppose we have these class definitions: + +```ruby +# app/models/hotel/services.rb +module Hotel + class Services + end +end + +# app/models/hotel/geo_location.rb +module Hotel + class GeoLocation + class << self + Services + end + end +end +``` + +If `Hotel::Services` is known by the time `app/models/hotel/geo_location.rb` +is being loaded, `Services` is resolved by Ruby because `Hotel` belongs to the +nesting when the singleton class of `Hotel::GeoLocation` is opened. + +But if `Hotel::Services` is not known, Rails is not able to autoload it, the +application raises `NameError`. + +The reason is that autoloading is triggered for the singleton class, which is +anonymous, and as [we saw before](#generic-procedure), Rails only checks the +top-level namespace in that edge case. + +An easy solution to this caveat is to qualify the constant: + +```ruby +module Hotel + class GeoLocation + class << self + Hotel::Services + end + end +end +``` + +### Autoloading in `BasicObject` + +Direct descendants of `BasicObject` do not have `Object` among their ancestors +and cannot resolve top-level constants: + +```ruby +class C < BasicObject + String # NameError: uninitialized constant C::String +end +``` + +When autoloading is involved that plot has a twist. Let's consider: + +```ruby +class C < BasicObject + def user + User # WRONG + end +end +``` + +Since Rails checks the top-level namespace `User` gets autoloaded just fine the +first time the `user` method is invoked. You only get the exception if the +`User` constant is known at that point, in particular in a *second* call to +`user`: + +```ruby +c = C.new +c.user # surprisingly fine, User +c.user # NameError: uninitialized constant C::User +``` + +because it detects that a parent namespace already has the constant (see [Qualified +References](#autoloading-algorithms-qualified-references)). + +As with pure Ruby, within the body of a direct descendant of `BasicObject` use +always absolute constant paths: + +```ruby +class C < BasicObject + ::String # RIGHT + + def user + ::User # RIGHT + end +end +``` + +### Autoloading in the Test Environment + +When configuring the `test` environment for autoloading you might consider multiple factors. + +For example it might be worth running your tests with an identical setup to production (`config.eager_load = true`, `config.cache_classes = true`) in order to catch any problems before they hit production (this is compensation for the lack of dev-prod parity). However this will slow down the boot time for individual tests on a dev machine (and is not immediately compatible with spring see below). So one possibility is to do this on a +[CI](https://en.wikipedia.org/wiki/Continuous_integration) machine only (which should run without spring). + +On a development machine you can then have your tests running with whatever is fastest (ideally `config.eager_load = false`). + +With the [Spring](https://github.com/rails/spring) pre-loader (included with new Rails apps), you ideally keep `config.eager_load = false` as per development. Sometimes you may end up with a hybrid configuration (`config.eager_load = true`, `config.cache_classes = true` AND `config.enable_dependency_loading = true`), see [spring issue](https://github.com/rails/spring/issues/519#issuecomment-348324369). However it might be simpler to keep the same configuration as development, and work out whatever it is that is causing autoloading to fail (perhaps by the results of your CI test results). + +Occasionally you may need to explicitly eager_load by using `Rails +.application.eager_load!` in the setup of your tests -- this might occur if your [tests involve multithreading](https://stackoverflow.com/questions/25796409/in-rails-how-can-i-eager-load-all-code-before-a-specific-rspec-test). + +## Troubleshooting + +### Tracing Autoloads + +Active Support is able to report constants as they are autoloaded. To enable these traces in a Rails application, put the following two lines in some initializer: + +```ruby +ActiveSupport::Dependencies.logger = Rails.logger +ActiveSupport::Dependencies.verbose = true +``` + +### Where is a Given Autoload Triggered? + +If constant `Foo` is being autoloaded, and you'd like to know where is that autoload coming from, just throw + +```ruby +puts caller +``` + +at the top of `foo.rb` and inspect the printed stack trace. + +### Which Constants Have Been Autoloaded? + +At any given time, + +```ruby +ActiveSupport::Dependencies.autoloaded_constants +``` + +has the collection of constants that have been autoloaded so far. diff --git a/guides/source/command_line.md b/guides/source/command_line.md index 4681574edd..60d0de17bc 100644 --- a/guides/source/command_line.md +++ b/guides/source/command_line.md @@ -601,6 +601,8 @@ The `tmp:` namespaced commands will help you clear and create the `Rails.root/tm ### Miscellaneous +* `rails initializers` prints out all defined initializers in the order they are invoked by Rails. +* `rails middleware` lists Rack middleware stack enabled for your app. * `rails stats` is great for looking at statistics on your code, displaying things like KLOCs (thousands of lines of code) and your code to test ratio. * `rails secret` will give you a pseudo-random key to use for your session secret. * `rails time:zones:all` lists all the timezones Rails knows about. diff --git a/guides/source/configuring.md b/guides/source/configuring.md index e53e8b4e92..c5d3d09bd0 100644 --- a/guides/source/configuring.md +++ b/guides/source/configuring.md @@ -353,7 +353,7 @@ All these configuration options are delegated to the `I18n` library. * `config.active_record.lock_optimistically` controls whether Active Record will use optimistic locking and is `true` by default. -* `config.active_record.cache_timestamp_format` controls the format of the timestamp value in the cache key. Default is `:nsec`. +* `config.active_record.cache_timestamp_format` controls the format of the timestamp value in the cache key. Default is `:usec`. * `config.active_record.record_timestamps` is a boolean value which controls whether or not timestamping of `create` and `update` operations on a model occur. The default value is `true`. @@ -844,6 +844,8 @@ You can find more detailed configuration options in the * `config.active_storage.content_types_to_serve_as_binary` accepts an array of strings indicating the content types that Active Storage will always serve as an attachment, rather than inline. The default is `%w(text/html text/javascript image/svg+xml application/postscript application/x-shockwave-flash text/xml application/xml application/xhtml+xml application/mathml+xml text/cache-manifest)`. +* `config.active_storage.content_types_allowed_inline` accepts an array of strings indicating the content types that Active Storage allows to serve as inline. The default is `%w(image/png image/gif image/jpg image/jpeg image/vnd.adobe.photoshop image/vnd.microsoft.icon application/pdf)`. + * `config.active_storage.queues.analysis` accepts a symbol indicating the Active Job queue to use for analysis jobs. When this option is `nil`, analysis jobs are sent to the default Active Job queue (see `config.active_job.default_queue_name`). ```ruby @@ -881,7 +883,11 @@ text/javascript image/svg+xml application/postscript application/x-shockwave-fla config.active_storage.routes_prefix = '/files' ``` - The default is `/rails/active_storage` + The default is `/rails/active_storage`. + +* `config.active_storage.replace_on_assign_to_many` determines whether assigning to a collection of attachments declared with `has_many_attached` replaces any existing attachments or appends to them. The default is `true`. + +* `config.active_storage.draw_routes` can be used to toggle Active Storage route generation. The default is `true`. ### Results of `load_defaults` @@ -917,6 +923,7 @@ text/javascript image/svg+xml application/postscript application/x-shockwave-fla - `config.active_job.return_false_on_aborted_enqueue`: `true` - `config.active_storage.queues.analysis`: `:active_storage_analysis` - `config.active_storage.queues.purge`: `:active_storage_purge` +- `config.active_storage.replace_on_assign_to_many`: `true` - `config.active_record.collection_cache_versioning`: `true` ### Configuring a Database diff --git a/guides/source/contributing_to_ruby_on_rails.md b/guides/source/contributing_to_ruby_on_rails.md index d3706a4dbf..9c28ff6a9c 100644 --- a/guides/source/contributing_to_ruby_on_rails.md +++ b/guides/source/contributing_to_ruby_on_rails.md @@ -139,7 +139,7 @@ changes to the master branch. When working with documentation, please take into account the [API Documentation Guidelines](api_documentation_guidelines.html) and the [Ruby on Rails Guides Guidelines](ruby_on_rails_guides_guidelines.html). -NOTE: To help our CI servers you should add [ci skip] to your documentation commit message to skip build on that commit. Please remember to use it for commits containing only documentation changes. +NOTE: For documentation changes, your commit message should include [ci skip]. This will skip running the test suite, helping us to cut down on our server costs. Keep in mind that you should only skip CI when your change touches documentation exclusively. Translating Rails Guides ------------------------ diff --git a/guides/source/documents.yaml b/guides/source/documents.yaml index 8df1c601a7..da11236064 100644 --- a/guides/source/documents.yaml +++ b/guides/source/documents.yaml @@ -139,9 +139,13 @@ url: initialization.html description: This guide explains the internals of the Rails initialization process. - - name: Autoloading and Reloading Constants + name: Autoloading and Reloading Constants (Zeitwerk Mode) url: autoloading_and_reloading_constants.html - description: This guide documents how autoloading and reloading constants work. + description: This guide documents how autoloading and reloading constants work (Zeitwerk mode). + - + name: Autoloading and Reloading Constants (Classic Mode) + url: autoloading_and_reloading_constants_classic_mode.html + description: This guide documents how autoloading and reloading constants work (Classic mode). - name: "Caching with Rails: An Overview" url: caching_with_rails.html diff --git a/guides/source/engines.md b/guides/source/engines.md index 90b08b00f0..8961a079b5 100644 --- a/guides/source/engines.md +++ b/guides/source/engines.md @@ -220,7 +220,7 @@ Inside the `app` directory are the standard `assets`, `controllers`, `helpers`, `jobs`, `mailers`, `models`, and `views` directories that you should be familiar with from an application. We'll look more into models in a future section, when we're writing the engine. -Within the `app/assets` directory, there are the `images`, `javascripts` and +Within the `app/assets` directory, there are the `images` and `stylesheets` directories which, again, you should be familiar with due to their similarity to an application. One difference here, however, is that each directory contains a sub-directory with the engine name. Because this engine is diff --git a/guides/source/layouts_and_rendering.md b/guides/source/layouts_and_rendering.md index 39935cd2ef..ce90a60e36 100644 --- a/guides/source/layouts_and_rendering.md +++ b/guides/source/layouts_and_rendering.md @@ -149,25 +149,6 @@ Rails knows that this view belongs to a different controller because of the embe render template: "products/show" ``` -#### Rendering an Arbitrary File - -The `render` method can also use a view that's entirely outside of your application: - -```ruby -render file: "/u/apps/warehouse_app/current/app/views/products/show" -``` - -The `:file` option takes an absolute file-system path. Of course, you need to have rights -to the view that you're using to render the content. - -NOTE: Using the `:file` option in combination with users input can lead to security problems -since an attacker could use this action to access security sensitive files in your file system. - -NOTE: By default, the file is rendered using the current layout. - -TIP: If you're running Rails on Microsoft Windows, you should use the `:file` option to -render a file, because Windows filenames do not have the same format as Unix filenames. - #### Wrapping it up The above three ways of rendering (rendering another template within the controller, rendering a template within another controller, and rendering an arbitrary file on the file system) are actually variants of the same action. @@ -178,17 +159,9 @@ In fact, in the BooksController class, inside of the update action where we want render :edit render action: :edit render "edit" -render "edit.html.erb" render action: "edit" -render action: "edit.html.erb" render "books/edit" -render "books/edit.html.erb" render template: "books/edit" -render template: "books/edit.html.erb" -render "/path/to/rails/app/views/books/edit" -render "/path/to/rails/app/views/books/edit.html.erb" -render file: "/path/to/rails/app/views/books/edit" -render file: "/path/to/rails/app/views/books/edit.html.erb" ``` Which one you use is really a matter of style and convention, but the rule of thumb is to use the simplest one that makes sense for the code you are writing. @@ -287,6 +260,23 @@ time. NOTE: Unless overridden, your response returned from this render option will be `text/plain`, as that is the default content type of Action Dispatch response. +#### Rendering raw file + +Rails can render a raw file from an absolute path. This is useful for +conditionally rendering static files like error pages. + +```ruby +render file: "#{Rails.root}/public/404.html", layout: false +``` + +This renders the raw file (it doesn't support ERB or other handlers). By +default it is rendered within the current layout. + +WARNING: Using the `:file` option in combination with users input can lead to security problems +since an attacker could use this action to access security sensitive files in your file system. + +TIP: `send_file` is often a faster and better option if a layout isn't required. + #### Options for `render` Calls to the `render` method generally accept five options: @@ -303,7 +293,7 @@ Calls to the `render` method generally accept five options: By default, Rails will serve the results of a rendering operation with the MIME content-type of `text/html` (or `application/json` if you use the `:json` option, or `application/xml` for the `:xml` option.). There are times when you might like to change this, and you can do so by setting the `:content_type` option: ```ruby -render file: filename, content_type: "application/rss" +render template: "feed", content_type: "application/rss" ``` ##### The `:layout` Option diff --git a/guides/source/rails_application_templates.md b/guides/source/rails_application_templates.md index 982df26987..e0e79fc41b 100644 --- a/guides/source/rails_application_templates.md +++ b/guides/source/rails_application_templates.md @@ -82,10 +82,10 @@ end Adds the given source to the generated application's `Gemfile`. -For example, if you need to source a gem from `"http://code.whytheluckystiff.net"`: +For example, if you need to source a gem from `"http://gems.github.com"`: ```ruby -add_source "http://code.whytheluckystiff.net" +add_source "http://gems.github.com" ``` If block is given, gem entries in block are wrapped into the source group. diff --git a/guides/source/upgrading_ruby_on_rails.md b/guides/source/upgrading_ruby_on_rails.md index 54f014293d..05980d1614 100644 --- a/guides/source/upgrading_ruby_on_rails.md +++ b/guides/source/upgrading_ruby_on_rails.md @@ -184,41 +184,21 @@ That may be handy if you need to preload STIs or configure a custom inflector, f If the application being upgraded autoloads correctly, the project structure should be already mostly compatible. -However, `classic` mode infers file names from missing constant names (`underscore`), whereas `zeitwerk` mode infers constant names from file names (`camelize`). These helpers are not always inverse of each other, in particular if acronyms are involved. For instance, `"FOO".underscore` is `"foo"`, but `"foo".camelize` is `"Foo"`, not `"FOO"`. Compatibility can be checked by setting `classic` mode first temporarily: +However, `classic` mode infers file names from missing constant names (`underscore`), whereas `zeitwerk` mode infers constant names from file names (`camelize`). These helpers are not always inverse of each other, in particular if acronyms are involved. For instance, `"FOO".underscore` is `"foo"`, but `"foo".camelize` is `"Foo"`, not `"FOO"`. -```ruby -# config/application.rb +Compatibility can be checked with the `zeitwerk:check` task: -config.load_defaults "6.0" -config.autoloader = :classic ``` - -and then running - +$ bin/rails zeitwerk:check +Hold on, I am eager loading the application. +All is good! ``` -bin/rails zeitwerk:check -``` - -When all is good, you can delete `config.autoloader = :classic`. #### require_dependency All known use cases of `require_dependency` have been eliminated, you should grep the project and delete them. -In the case of STIs with a hierarchy of more than two levels, you can preload the leaves of the hierarchy in an initializer: - -```ruby -# config/initializers/preload_stis.rb - -# By preloading leaves, the hierarchy is loaded upwards following -# the references to superclasses in the class definitions. -sti_leaves = %w( - app/models/leaf1.rb - app/models/leaf2.rb - app/models/leaf3.rb -) -Rails.autoloaders.main.preload(sti_leaves) -``` +If your application has STIs, please check their section in the guide [Autoloading and Reloading Constants (Zeitwerk Mode)](autoloading_and_reloading_constants.html#single-table-inheritance). #### Qualified names in class and module definitions @@ -270,6 +250,16 @@ In that case, `app/models/concerns` is assumed to be a root directory (because i The `Concerns::` namespace worked with the classic autoloader as a side-effect of the implementation, but it was not really an intended behavior. An application using `Concerns::` needs to rename those classes and modules to be able to run in `zeitwerk` mode. +#### Having `app` in the autoload paths + +Some projects want something like `app/api/base.rb` to define `API::Base`, and add `app` to the autoload paths to accomplish that in `classic` mode. Since Rails adds all subdirectories of `app` to the autoload paths automatically, we have another situation in which there are nested root directories, so that setup no longer works. Similar principle we explained above with `concerns`. + +If you want to keep that structure, you'll need to delete the subdirectory from the autoload paths in an initializer: + +```ruby +ActiveSupport::Dependencies.autoload_paths.delete("#{Rails.root}/app/api") +``` + #### Autoloaded Constants and Explicit Namespaces If a namespace is defined in a file, as `Hotel` is here: @@ -391,6 +381,12 @@ To fix this, just remove the wildcards: config.autoload_paths << "#{config.root}/lib" ``` +#### Eager loading and autoloading are consistent + +In `classic` mode, if `app/models/foo.rb` defines `Bar`, you won't be able to autoload that file, but eager loading will work because it loads files recursively blindly. This can be a source of errors if you test things first eager loading, execution may fail later autoloading. + +In `zeitwerk` mode both loading modes are consistent, they fail and err in the same files. + #### How to Use the Classic Autoloader in Rails 6 Applications can load Rails 6 defaults and still use the classic autoloader by setting `config.autoloader` this way: @@ -402,6 +398,54 @@ config.load_defaults "6.0" config.autoloader = :classic ``` +### Active Storage assignment behavior change + +In Rails 5.2, assigning to a collection of attachments declared with `has_many_attached` appended new files: + +```ruby +class User < ApplicationRecord + has_many_attached :highlights +end + +user.highlights.attach(filename: "funky.jpg", ...) +user.higlights.count # => 1 + +blob = ActiveStorage::Blob.create_after_upload!(filename: "town.jpg", ...) +user.update!(highlights: [ blob ]) + +user.highlights.count # => 2 +user.highlights.first.filename # => "funky.jpg" +user.highlights.second.filename # => "town.jpg" +``` + +With the default configuration for Rails 6.0, assigning to a collection of attachments replaces existing files +instead of appending to them. This matches Active Record behavior when assigning to a collection association: + +```ruby +user.highlights.attach(filename: "funky.jpg", ...) +user.highlights.count # => 1 + +blob = ActiveStorage::Blob.create_after_upload!(filename: "town.jpg", ...) +user.update!(highlights: [ blob ]) + +user.highlights.count # => 1 +user.highlights.first.filename # => "town.jpg" +``` + +`#attach` can be used to add new attachments without removing the existing ones: + +```ruby +blob = ActiveStorage::Blob.create_after_upload!(filename: "town.jpg", ...) +user.highlights.attach(blob) + +user.highlights.count # => 2 +user.highlights.first.filename # => "funky.jpg" +user.highlights.second.filename # => "town.jpg" +``` + +Opt in to the new default behavior by setting `config.active_storage.replace_on_assign_to_many` to `true`. +The old behavior will be deprecated in Rails 6.1 and removed in a subsequent release. + Upgrading from Rails 5.1 to Rails 5.2 ------------------------------------- |