Honestly this looks fine to me—it’s nice and explicit. It’s not just the when clause that’s different but payload: unquote(pattern) v payload: unquote(payload). What you have is nice and explicit and I think DRYing it up would probably be more confusing (and hence a great case of “when DRY is too DRY”).

Although I see Chris K. is replying and he a notorious committer of macro crimes, so I’m interested to see what he’ll suggest

He did not disappoint, but also made me more confident in my resolve TIL about :elixir_utils.extract_guards which is very cool. But for what it’s worth, in a tense debugging session (or even just coming across it randomly in while trying to understand a system) I would much rather see your code.

Consider the (only semi-public but very stable) :elixir_utils.extract_guards/1 function from Elixir’s compiler. It takes in Macro AST of an expression, and returns a two-tuple of that expression with top-level guards removed, and a list of the guards extracted. Its implementation is similar to yours but handles multiple guards correctly (a rarely used feature in Elixir code, but syntactically valid).

Ex:

payload = quote(do: foo)
{pattern, guards} = :elixir_utils.extract_guards(payload)
#=> {{:foo, [], Elixir}, []}
payload = quote(do: foo when fizz)
{pattern, guards} = :elixir_utils.extract_guards(payload)
#=> {{:foo, [], Elixir}, [{:fizz, [], Elixir}]}
payload = quote(do: foo when fizz when buzz)
{pattern, guards} = :elixir_utils.extract_guards(payload)
#=> {{:foo, [], Elixir}, [{:fizz, [], Elixir}, {:buzz, [], Elixir}]}

You can then rebuild the guard AST around a new expression for insertion where desired without thinking about if there are zero or more by reducing over them:

new_guarded_expression = quote(do: bar)
Enum.reduce(guards, new_guarded_expression, fn guard, expression ->
  {:when, [], [expression, guard]}
end)
|> Macro.to_string
#=> "(bar when fizz) when buzz"

Yep, I’m glossing over a lot here, especially since OP’s code is clearly a test assertion. There is absolutely a huge difference in what’s acceptable in library v. application code and I was assuming this was for the latter (even if it is library-esque code within application code).