chrisbetz
commented
9 years ago Hi Dima,
quoting from my mail for others who follow to read:
One thing coming to my mind (aka stuff I fell over on my way to Spark) could be compilation issue: You need to make sure to have the right classes at hand, because (e.g. at the repl) clojure does recompile stuff, resulting in new classes (with the same name, but a different class). So, it might happen that you registered Class clojure.core$comparator$fn4661.class in version "A", wheras actual data contains class clojure.core$comparator$fn4661.class in version "B". I think something similar happens with (comparator … ), but with two different classes (something like clojure.core$comparator$fn4661 and clojure.core$comparator$fn2737.
Thus, two calls to (comparator …) do return two classes. Make sure to use only one comparator (aka def a var with comparator instance and use that all over the place.
It's really hard to debug that kind of stuff. You could log (.identityHashCode (class your-comparator)) while registering it and also the instance you use in your data structure to catch both types of errors - if the two logged hashCodes differ, you know you didn't register the right class.
Hope this helps
Cheers,
Chris
P.S.: Would you mind opening up pull requests against the develop branch? This would simplify things for me. Thanks for that!
Hi Chris,
I added take-ordered to sparkling.api. I also fiddled with additional test which accepts comparator but couldn't make it work due to serialization issues. You can see what I tried here. Hope that helps.