Comparing changes

@mockdeep

* Check for puma and capybara when defining system tests Rather than skipping the definition altogether define a fake module that warns about the dependencies. * If we can't load system_test_case for some reason we cant run system tests As per @mockdeep's conversation in #1889 other dependencies might not allow us to even load system_test_case so we fallback to our original strategy for now.

* Follow updated ruby versions in the matrix * Keep 2.2.n matrix as before

Delay loading system tests as late as possible in order to allow people who arent using them to not have to turn off puma or other such things. It also enables other webservers to be configured other than puma which is needed in Rails as of rails/rails@50f6976

Relax the regexp for matching the expected "uninitialized constant" error to fix this spec failure when running on Bundler 1.16.0 and Ruby 2.4.2: expected "NameError: uninitialized constant RSpec::Support" to match /uninitialized constant RSpec::Support \(NameError\)/

* Improve doc about Job specs & rails versions matrix * Add System specs information in doc * doc: add links from directory structure feature to other features * doc: simpler links to other features

* Fix RuboCop violation * Update test harness for Bundler 1.16 compatibility This commit fixes Travis failures due to the recent introduction of Bundler 1.16.0. Both are related to the new style of binstubs that are generated by 1.16.0. First, add `bundler` to the list of binstubs that we explicitly generate when creating the test Rails app. The `bin/bundle` binstub that ships as part of `rails new` is in the pre-1.16 format, and conflicts with the 1.16 format binstubs that are generated for rake and rspec. Fix this by replacing the default `bin/bundle` with a fresh one generated by `bundle binstubs bundler --force`. Second, the `Bundle.with_clean_env` wrapper seems to remove some important information from the environment that causes binstubs to fail with errors like `uninitialized constant Gem (NameError)`. Fix this by using `Bundle.with_original_env` instead (which is what the Bundler documentation recommends). Because Travis also sets its own `BUNDLER_GEMFILE` variable, I had to also reset that one manually. * Choose cuke tags based on correct RAILS_VERSION * Travis needs doc gems in order to run bin/yard * Add --no-rc flag for more reliable local testing The rspec-rails acceptance tests use `rails new` to generate a sample app for running tests. If a developer running the rspec-rails test suite has a `.railsrc` file, this will affect the app that is generated, causing specs to fail. This commit fixes this by passing the `--no-rc` flag so that `rails new` ignores the developer's local `.railsrc`.

If RSpec's `:aggregate_failures` is enabled, a system spec with a failed expectation would not automatically trigger a screenshot as expected. This is because aggregated failures are not exposed via the typical `RSpec.current_example.exception`. To correctly detect failures when aggregation is enabled, we have to do more work. This commit works around this problem by using some behind-the-scenes knowledge of how rspec-expectations does the aggregation. Now we reach into the thread-local storage (`RSpec::Support.failure_notifier`) and discover whether there are in fact failures "queued up". If so, we will consider the example to have failed and take a screenshot. This is the same workaround as used by capybara-screenshot: mattheworiordan/capybara-screenshot#213

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Comparing changes

Open a pull request

Commits on Nov 20, 2017

This comparison is taking too long to generate.