I was in the same situation, that I wanted to include logging in the tests. I ended up writing a little macro that gets some meta-data from __CALLER__ and generates log-tuples

assert [{:logger, :error, {4711, %{file: _, line: _}}}] = error(4711)
assert [{:logger, :warning, {'test', %{file: _, line: _}}}] = warning('test')
assert [{:logger, :debug, {[a: 1, b: 2], %{file: _, line: _}}}] = debug([a: 1, b: 2])

The advantage over capture_log is, that you can use structured logging while capture_log can only use the console backend. Also it’s absolutely pure (if you are strict and see logging as impure) so there are no problems with async. But you have to handle the log-data the core produces somehow in the shell (and do the real logging).

There is no such feature at the moment. It would need to be added to Elixir. But keep in mind that, given your tests run concurrently, this could cause Heisenfailures as one test could incorrectly capture the logs of another test and not fail when it should fail (which is ok, it would not be ok to fail when it should not have).

Here’s are examples of captured/uncaptured entries:

test "this is captured" do
  assert capture_log(fn ->
    Logger.warn("example")
  end) =~ "example"
end

@tag capture_log
test "this is captured too" do
  Logger.warn("example")
end

test "this is not captured and should fail" do
  Logger.warn("example")
end

Yeah, we could support a @tag :assert_no_log.

This is where we check for capture log tag: elixir/lib/ex_unit/lib/ex_unit/runner.ex at main · elixir-lang/elixir · GitHub

We probably want to make it a tri-state: :capture, :none, :assert. And then fail the test if it passes but has a log in asset mode.