If all the paths specified in `FactoryBot.definition_file_paths` do not
exist, an empty array is passed to the reloader.
If an empty array is specified, reloader passes it to `Listen.to` as it
is. This causes the application root to be watched and `node_modules` to
become a target of listening.
The behavior when an empty array is passed to reloader depends on the
Rails side. Fixes to not register a reloader when file paths do not exist
consistent behavior regardless of Rails side.
Fixes#328.
Currently, `factories`, `test/factories` and `spec/factories` are
specified by default as reload watching paths.
In many applications, there are directories which do not exist (perhaps
do not have` spec` if using minitest and do not have` spec` if using
minitest).
In Rails 5 series, if specify a path that does not exist in
`EventedFileUpdateChecker`, its parent is added to the watching path.
As a result, `node_modules` is also included in the watching path, and
unexpectedly many directories and files are included in listen's watching
targets. Also, if symlink is included in `node_modules`, it also causes
warning of listen.
This issue is solved in Rails 6. However, since many applications use
this gem in Rails 5 and below, it is good not to add directories that
do not exist on the gem side.
Ref: https://github.com/rails/rails/issues/32700
Closes#236
This commit uses ActiveSupport's FileUpdateChecker to allow reloading
FactoryBot of definitions whenever a file in
`FactoryBot.definition_file_paths` gets updated.
This is similar to reloading for
[I18n](ced104d579/activesupport/lib/active_support/i18n_railtie.rb (L60-L70))
and [react rails](83b6175460/lib/react/rails/railtie.rb (L35-L41))
This allows us to get rid of any Spring-specific logic in the railtie,
since [Spring hooks into the application
reloader](0c711ff10b/lib/spring/application.rb (L161)).
This partly solves #211, since we no longer call `FactoryBot.reload` at
all in `after_initialize`. Instead, we will only call it when one of the
files in `definition_file_paths` gets updated. I say it partly
solves #211 because once a definition file gets updated, reloading
would still give warnings about redefining any constants in the
definition files. I wonder how common it is to define constants in the
same file as factory definitions. It's probably better to keep constants
in the autoload path, allowing Rails to handle unloading and reloading
them in development. I would want to see some specific examples before
worrying too much about it.
I would also like to offer a better way to configure the definition file
paths (see #165 and #166) and possibly an option to opt out of loading
definitions at all (could help with issues like #192).
A couple of quirks here:
* the to_prepare block could potentially get
[called multiple times](https://github.com/rails/rails/issues/28108).
It shouldn't matter, since `execute_if_updated` is a no-op if no
updates have been made.
* I noticed that the first time I call `reload!` in the console the
factory definitions get reloaded even when I made no updates to the
definition files. I think it is related to
[this](f7c5a8ce26/activesupport/lib/active_support/evented_file_update_checker.rb (L18)). After the first call
`reload!` works as expected, only reloading the factory definitions
if the definition files were updated.
* Rails uses execute rather than execute_if_updated for the
[route
reloader](https://github.com/rails/rails/blob/master/railties/lib/rails/application/finisher.rb#L133)
and for the [watchable file
reloader](https://github.com/rails/rails/blob/master/railties/lib/rails/application/finisher.rb#L173).
This means that changes to factory definitions will cause reloading of
the whole application. I think this is fine.
