RogerBW's Blog

I have ended up maintaining a plugin for Discourse. As I got it, it had no tests. Fixing that was fun.

Discourse is largely written in Ruby on Rails, and its plugins need to be in Ruby. Fair enough; it's a procedural language, if more object-flavoured than many, and getting up a bit of fluency isn't hard.

This particular thing is a dice roller for forum-based boardgames and RPGs. Discourse didn't have one to start with, and though it's now grown a built-in one it's pretty basic. So originally I installed one that seems to be generally well-regarded; but that version has some quite major bugs in it (for example, it will reject a roll if the number of dice has a zero in it), and it repeats itself too much for my liking (the regexp definition of "what a request for a roll looks like" is in three different places). So I forked it, amended it a bit and put it on my own github space.

But testing it on Discourse is tricky. To get it into the live server I end up doing a full rebuild, which takes a while. I keep a toybox Discourse instance in a virtual machine at home specifically for this kind of thing; there I can push the code into the right place, then stop and restart to get it reloaded, but this still takes noticeable numbers of seconds (20+). What I want is a fast test wrapper that will exercise the various functions and make sure they're doing sensible things as I'm working on the core code, before I throw it at Discourse for final testing. Save code, run tests, see what's wrong, fix code, repeat.

Ruby has a test system; but the plugin is designed to run within Ruby on Rails. Its general form looks like

after_initialize do

  def # various functions I want to test
  end
  
  on(:post_created) do |post, params|
    # more code I want to test
  end
  
end

OK. So the def parts are easy enough; I define my own function in the test wrapper

def after_initialize
  yield
end

which means "when you meet the after_initialize keyword, run the block of code it's introducing". At that point, the functions get defined, and I can poke at them individually.

But what about on? That clearly introduces a block of code, but I don't want to run that block at this moment; I want to capture it and run it later at my whim. This ended up being:

$onblock=Proc.new { }

def on(post,&block)
  $onblock=block
end

defining a global $onblock which holds the code. I also define a Post class with the few methods of a real post that this plugin cares about. Then I can test it with, for example:

post=Post.new('[roll 3d6]')
srand(1602262750)
$onblock.call(post)
assert_match(/USERNAME asked for a die roll:.*`3d6: 4 \+ 2 \+ 3 = 9`/m,post.raw)

(Note the random seeding for a fixed output; this probably makes the code more fragile, because if the random algorithm changes all my tests will suddenly fail, but it will at least let me know if any part of it isn't doing the right thing.)

So now when I'm extending the plugin I can write the tests and make sure it's doing what it should almost instantly, at the command line, before I start poking it into an actual Discourse installation and doing the second stage testing of it there.

(And I've used it already, because I've just added a new subsystem for the special dice needed by the Genesys RPG – exercising it with tests before I tried it on real Discourse.)