Change behavior of render "foo" to render a string instead of a Rails template

[bradgessler]

From Issue #137, this PR makes it such that:

render "foo"

renders a String instead of the rails partial foo. Rendering a partial will be changed to:

render partial: "foo"

This will be a breaking change for Phlex 1.x and ultimately require a path to Phlex 2.x. As such, I'm going to have an Overrides::Version1 and Overrides::Verrsion2 module that get included depending on the setting specified. This will let people upgraded in 1.x to 2.x. When 2.x ships I'll remove 1.x and any updates will break.

[bradgessler]

Turns out render "foo" and render partial: "foo" behavior differently in Rails 🤦.

Given this Erb template:

Hi

<%= render "foo" do %>
The middle
<% end %>

<%= render partial: "foo" do %>
The middle
<% end %>

The first one will work, but the latter will throw this error:

'nil' is not an ActiveModel-compatible object. It must implement :to_partial_path.

[bradgessler]

New approach, which I think could eliminate a lot of the overrides.

render(rails("fizz") { "buzz" })

The rails method returns this object:

# frozen_string_literal: true

# @api private
module Phlex::Rails
  class Rendition
    attr_reader :args, :kwargs, :block

    def initialize(*args, **kwargs, &block)
      @args = args
      @kwargs = kwargs
      @block = block
    end
  end
end

Which is intercepted by the render override:

when Phlex::Rails::Rendition
  captured_block = -> { capture(&renderable.block) } if renderable.block
  @_context.target << @_view_context.render(*renderable.args, **renderable.kwargs, &captured_block)

I would propose moving this into an adapter, which would be included via the kitchen sink, but at least for all-Phlex apps the Rails renderer would never be invoked.

[bradgessler]

I opened a bug at rails/rails#51015. If Rails decides this is a bug and fixes it, we should be set. If they don't then we'll have to figure out how to render from Phlex in a manner that doesn't obstruct the "fidelity" of rendering in Rails.

[joeldrapper]

Had a conversation about the Rails issue here. https://discord.com/channels/849034466856665118/974005005768069211/1218911072522600481

[joeldrapper]

@bradgessler any thoughts on what we can do about this for 2.0? 🤔

[bradgessler]

Yeah we could either take the approach above where we wrap Rails rendering calls in a class via render(rails("fizz") { "buzz" }). This would provide the most "faithful" rendition of Rails content since we'd let people continue calling into this undefined/broken behavior.

Another option is Phlex assumes render "string" does NOT call out to Rails and we instead require a keyword like partial: to call out into Rails renditions.

Another option is we could have a rails.render or rails_render method that makes it clear the call is into Rails and not into Phlex.

I think we should rule out patching Rails since (1) they don't seem interested and (2) getting patches through for stuff like this has historically been a 🥙 for me.

[joeldrapper]

Agh, this is such a tricky issue. None of those options feel quite right, but we have to come up with something. 🤔