Macros with gensym make not reusing cached results after JVM restart

[pm4r]

I noticed that for some function calls results are not taken from cache after JVM restart. It is especially frustrating when it take a couple of minutes to execute something. Turned out that the reason were macros with auto-gensym (like abc#) because on each JVM start that would evaluate to different symbol, thus analyzer would find such form to be different than what it has already cached
.
The workaround was to change gensym with fixed symbols like ~'abc. I'm not sure if there is better solution for that. It would be good to clarify that in docs.

[mk]

Hi @pm4r, thanks for raising this. I'll need to dig a bit deeper into this. I'd like to get Clerk's caching work for these macros as well.

[mk]

I started working on this with da94bb0. @pm4r can you try if that fixes your problem? If you don't have time, I should get around to it in the coming days.

[pm4r]

Sorry for late reply, I will check this out by the end of this week.

[pm4r]

I tested it and it works with one weird behavior. In each case, after notebook is shown it's get evaluated once and refreshing page uses that cached value - so exactly as it should. when I restart jvm and open notebook it still doesn't evaluate cached macros, however different value is shown in case of macro defined in the same namespace as notebook. After page is refreshed, previous value appears and logs don't indicate enything was re-evaluated - looks like something is cached on browser side only?
For example I have a notebook containing below code:

(defmacro attempt1
  [& body]
  `(try
     (do ~@body)
     (catch Exception e# e#)))

(def a1 (do
          (println "a1")
          (attempt1 (rand-int 9999))))
(def a2 (do
          (println "a2")
          (other-ns/attempt1 (rand-int 9999))))

In previous clerk versions I always see "a1" printed when refreshing page, and "a1" "a2" after jvm restart. In current version nothing is printed after page reload or jvm restart - which means it uses cached values. However cell shows different value for a1 after page is shown first time after jvm restart, and refreshing it (even without any changes in notebook) brings original value back. I tried it several times and always see the same two numbers for a1 (like 1580 for the first time, and 1392 after reloading - which is proper cached value). This happens only for macros defined in notebook.

[mk]

Hi @pm4r, thanks for trying it out and glad to hear it mostly works. The issue you're seeing above is a current limitation of Clerk's analysis and I've created #746 as a follow-up issue to fix or document that.