now / lookout Goto Github PK

Output is slow, as we have to flush it. It would be better to not report progress unless things are going slowly (if then, even). I think that reporting the progress as a percentage makes more sense as well, as those dots really don’t say a whole hell of a lot. This also makes it easier to see what happened, as the interesting output (failures and errors) would be the only output, if any.

Lookout::Expectations currently stores expectations as they are defined to later run them when requested to do so. This is completely unnecessary and I can think of no UI that would desire such behavior. Instead, load the interface as soon as this object is created, then, whenever #expect is called, run the expectation and report and store the result. (The storing of results is also questionable, as this is probably better served by a UI that decides to use Lookout::Results to summarize the results, if necessary. For example, Lookout::UI::Console currently only cares about failures and errors, so no need to store result objects for successes.)

This rewrite also saves us from using at_exit, which is something that I consider to be a worthy cause.

Anyway, for the normalization tests of U, the adjustment below saves us about a gigabyte worth of memory.

class Lookout::Expectations
  undef expect
  def expect(expected, &block)
    @results ||= Lookout::Results.new
    unless defined? @ui
      @ui = Lookout::UI::Console.new
      @ui.start
    end
    file, line = /\A(.*):(\d+)(?::in .*)?\z/.match(caller.first)[1..2]
    @results << Lookout::Expectation.on(expected, file, line, &block).evaluate.tap{ |result| @ui.report result }
    expected
  end

  undef evaluate
  def evaluate(ui = lookout::ui::console.new, results = lookout::results.new)
    @ui.summarize @results, 0
    @results
  end

  undef each
  def each(line = nil)
    self
  end
end

Should keep track of how long expectations run for so that we can output statistics about slow tests.

Delegation support seems like it won’t be used very much, so remove it, unless we find a good use for it.

The whole runner architecture should be redone. The rake loader should simply create an object that manages the state, loads all the files, then asks the runner to finalize the test run (doing what is currently done in the at_exit hook).

Mocha is a rather large library and contains a lot of things that we don’t need. Writing our own mocking framework would also be a good lesson.

It should be possible to expect an exact object:

expect exactly(/abc/) do
  /abc/
end

should pass while

expect exactly(/abc/) do
  'abc'
end

should not.

Having diff as a utility method is stupid. It would be better to add a literal for Strings that could have the diff method instead. Literals of StandardErrors could apply the diff to the messages.

One way to display diffs on random objects is to diff the instance variables like they were a hash (which they more or less are). The question is if we should be diffing the instance variables’ values as well.