Skip to content

Latest commit

 

History

History
286 lines (195 loc) · 8.5 KB

upgrade-from-opal-0.11-rails-5.md

File metadata and controls

286 lines (195 loc) · 8.5 KB
Raw

Progress:

  • hyper-spec: passing
  • hyperstack-config: passing
  • hyper-state: passing
  • hyper-store: passing
  • hyper-router: passing
  • hyper-component: passing
  • hyper-operation: passing
  • hyper-model: first couple of specs pass, suspect more issues because splat operator is now working properly.

Multiple Dependency Environments Supported

set env OPAL_VERSION='~ > 0.11' (for example) and then bundle update
set env RAILS_VERSION='~> 5.0' (for example) and then bundle update

note that when testing hyper-component with OPAL_VERSION 0.11 you must also set the environment variable to the same when running the specs (because we DONT want to fetch opal-browser from github see below)

note that Opal 0.11 will NOT work with rails 6.x, so that leaves three test environments

  1. no specification = latest opal (1.x) and latest rails (6.x)
  2. OPAL_VERSION='~> 0.11' = opal 0.11 and forces rails 5.x
  3. RAILS_VERSION='~> 5.0' = rails 5.x and defaults to latest opal (1.0)

Opal require implementation changed

replace

Opal.load('xxxx')

with

Opal.loaded(OpalLoaded || []);
Opal.require("components");

require 'opal' must be executed before any ruby code (including other requires.) This was not true in Opal 0.11

hyperstack-config semantically figures out the right method, so if you use hyperstack-config consistently to load the opal code, all will be well!

Typically:

  • app/assets/javascript/application.js
    should have the line:
    //= require hyperstack-loader
  • add remove other dependencies using Hyperstack.import

@@vars properly implemented

@@vars: Are now implemented correctly (I guess) so that they are lexically scoped, thus this does not work:

Foo.class_eval do
  def self.get_error
    @@error
  end
end

because its lexically scoped, the compiler looks for an inclosing class definition for @@foo, not what you want.

in this case simply replace Foo.class_eval do with class Foo

opal-browser must be taken from master to work with 1.0

opal-browser not released for opal 1.0 yet, causing deprecation warnings. To work around this we include browser in the hypercomponent Gemfile (not gemspec) unless the OPAL_VERSION environment variable is set to 0.11

TODO: Release opal-browser!

compiler checks for JS undefined, empty objects, and null in more places

This code will no longer work as expected:

x = `{}`
"foo(#{x})"  # <- generates "foo("+x+")" which
# results in the following string "foo([Object Object])"

not sure how this worked before, but it did.

array.each do |e|
  # be careful any values e that are `null` will
  # get rewritten as `nil` fine I guess most of the time
  # but caused some problems in specs where we are
  # explicitly trying to iterate over JS values
  # see /spec/client_features/component_spec.rb:656

back quotes operator now works properly in hyper-spec blocks

This now works properly:

  evaluate_ruby do
    ...
    foo = `JSON.stringify(x)`
    ...
  end

TODO: inconsistent dependency on server_rendering.js

search around for references to config.react.server_renderer_options and/or "server_rendering.js" inconsistent... should now default to hyperstack_prerender_loader.js and more over there is a hyperstack option as well, so the server_renderer file option should match

Opal splat works properly in 1.0!

foo(*[1, 2], [[3, 4]])

def foo(*x)
  # 0.9 x == [1, 2, [3, 4]]
  # 1.0 x == [1, 2, [[3, 4]]]
end

is there a fix that works for both 1.0 and 0.9 ? or do we need a SWITCH???

Found this first in hyper-model, so in hyper-model these are going to be marked with # SPLAT BUG with the old code until things get sorted

Can't inherit directly from ApplicationController::Base in Rails 6.0 ...

TODO pull in next-gen-hyperspec branch

TODO figure out how to make libv8 dependency optional

Patch selenium webdriver

Hyper-spec currently depends on an older version of selenium webdriver, and so requires this patch be applied to get logging to work:

module Selenium
  module WebDriver
    module Chrome
      module Bridge
        COMMANDS = remove_const(:COMMANDS).dup
        COMMANDS[:get_log] = [:post, 'session/:session_id/log']
        COMMANDS.freeze

        def log(type)
          data = execute :get_log, {}, {type: type.to_s}

          Array(data).map do |l|
            begin
              LogEntry.new l.fetch('level', 'UNKNOWN'), l.fetch('timestamp'), l.fetch('message')
            rescue KeyError
              next
            end
          end
        end
      end
    end
  end
end

TODO: figure out if we have to depend on this old version

BasicObject responds_to the class method

This means that a DummyValue returns DummyValue instead of the class its wrapping. Simple fix is to add

def class
  notify
  @object.class
end

which is what the method missing would do...

Opal 1.0 has trouble with super and method_missing

If super is called within a method that is not defined in the super class, the superclasses' method missing is called, but the method name is not provided.

This has been worked around, and a bug has been filed.

Object.present? should not have been defined

The DummyValue system was defining present? in both DummyValue and Object. This should not be necessary as present? is defined by activesupport. In 0.11 it didnt' matter because the load order defined activesupport's version AFTER the bogus one in the dummy value system. In 1.0 the order for some reason is reversed, so the correct version in activesupport was being overwritten.

I don't believe this will have any effect on application code.

HyperSpec is not using ApplicationController

HyperModel had its own version of HyperSpec. In that version the test controller was subclassed from ApplicationController. In the new version its subclassed from ActionController::Base.

In some HyperModel specs it was setting up mocks in ApplicationController, and these were not being used.

so now hyper-spec tries to use ApplicationController and falls back to ::ActionController::Base.

TODO: verify this works in all cases...

HyperSpec ApplicationController fix breaks client_driver.rb

A couple of specs were setting ApplicationController.acting_user = true. But HyperSpec was not using the ApplicationController when mounting (see above problem fix) so acting_user remained nil as far as the client code was concerned.

Once the above fix was made however client_driver was attempting to do a controller.acting_user.id where acting_user was true, causing a method missing. The code in client_driver was checking for acting_user being nil, but is now changed to check for acting_user.respond_to? :id

Trying to set window size outside of a JS spec no longer works

Who knows why, but for now we just rescue any failure, and keep going.

page.evaluate_ruby no longer works but plain evaluate_ruby does.

Only one spec files was doing that, so just upgraded.
You can sort of fix it by doing an include instead of config.include in the hyper_spec.rb file but this causes other problems, including a warning from rspec to not do it.

Probably will require a general cleanup of application code changing page.evaluate_ruby to just evaluate_ruby.

Related to this there was a method called attributes_on_client that was being added to ActiveRecord::Base. But in order for it to work the "page" was being passed so that you could do a page.evaluate_ruby, but the whole method was backwards. It should be a capybara helper method that takes a active record model. This is fixed.

Hyperspec changes behavior of on_client

New hyperspec aliases evaluate_ruby as on_client for readability, the old on_client is now called before_mount.

You can either update calls to on_client to be before_mount, or

You can get legacy behavior by executing this line in the spec_helper:

HyperSpec::ComponentTestHelpers.alias_method :on_client, :before_mount

HyperSpec todos

TODO: Hyperspec is dependent on HyperComponent for mounting components, and initializing the system.

However we want to make sure you can use HyperSpec without HyperComponent

TODO: Investigate why we need before_mount...

TODO: add_locals method only works on simple cases. For now make it experimental (i.e. make the method called experimental_add_locals, which you can alias, or include a module that switches on the behavior.)