We run this as part of our CI:
mix xref graph --format cycles --label compile-connected --min-cycle-size 2
Running with 1.19.0-rc.2 detected a huge cycle (length 171) that was not there before (1.18.4). I notice that only some modules in the cycle have the “compile” annotation, which seems suspicious.
an excerpt of the output
2 cycles found. Showing them in decreasing size:
Cycle of length 171:
lib/fonar/aanvullen.ex
lib/fonar/aanvullen/aanvullen_vanuit_wenslijst_worker.ex
lib/fonar/boeksuggesties.ex
lib/fonar/bronnen.ex
lib/fonar/daisy_cd_leveringen.ex
lib/fonar/daisy_cd_leveringen/daisy_cd_leveren_worker.ex
lib/fonar/email_activities.ex
lib/fonar/instanties.ex
lib/fonar/lezers.ex
lib/fonar/lezers/lezer_notifier.ex
lib/fonar/lezers/lezer_notifier_worker.ex
lib/fonar/notifier.ex (export)
lib/fonar/organisaties.ex
lib/fonar/organisaties/medewerker_notifier.ex
lib/fonar/reservaties.ex (export)
lib/fonar/reservaties/mail_reservatie_beschikbaar_worker.ex
lib/fonar/reservaties/reservatie.ex
lib/fonar/reservaties/verwerk_reservaties_worker.ex
lib/fonar/uitleningen.ex
lib/fonar/uitleningen/start_uitlening.ex
lib/fonar/uitleningen/uitlening_state_change.ex
lib/fonar/verwervingen.ex (export)
lib/fonar/verwervingen/formaat_verwerving.ex (compile)
lib/fonar/verwervingen/formaat_verwerving_states.ex
lib/fonar/verwervingen/productiebestelling.ex
lib/fonar/verwervingen/productiebestelling/item.ex
...
I doubt this is an actual compile-connected cycle. Could this be regression?
. The first time I ran it the real time was slower too (by around 10–20% iirc), but when I ran it while not doing anything else at the same time it evened out. Still seems slightly odd that it’s doing that much more work overall though, even if it’s making better use of all the cores.
