The sequence's head might be garbage collected if there are no references to it.

uberjar is just an archive with all the source and necessary class files. when you add uberjar as a dependency there is a very high chance it will not work because those files might collide if root project trying to add different version of the same dep

this ☝️ also if you need both A and B to coexist within the same jvm and each use their particular version of C, https://github.com/benedekfazekas/mrandersonhelps with that

Thanks a lot, how do that relate to shaded/unshaded. Like I have some bit of knowledge that java artifacts can be in both forms (shaded/unshaded).

I've done it in the past, I applied https://github.com/ptaoussanis/tufte over each public var of a given ns sadly it's closed-source but it shouldn't be hard to implement

Yeah I figured I could cobble together something using ns-publics; that requires me to have everything I want to measure bound to a var, but that's probably good practice for what I'm doing anyway for a fair number of reasons!

Only problem with that approach as far as I can see is that it may require two passes over a namespace that might have expensive computation in it (my use case is data science code): one to produce all the public vars, and a second to profile them.

> that requires me to have everything I want to measure bound to a var not really, tufte can also work at the form level. You could read forms from a .clj file, walk over them, wrap them in tufte, and then eval each top-level form now the question is: do you really want that? you'd get an exhaustingly detailed result

so it's better to start at the var level, and only focus on the slow vars, then split them, recur

it's not entirely about just identifying bottlenecks / hot spots. ML performance metrics + result coefficients might change on each eval too, so for those erring on the side of "exhaustingly detailed" is maybe OK

> Only problem with that approach as far as I can see is that it may require two passes over a namespace that might have expensive computation in it (my use case is data science code): one to produce all the public vars, and a second to profile them. Producing a public var is generally cheap - that's clojure compiler's job. Not data scientists' :) The profiling itself is very cheap - it's part of tufte's pitch So your code is compiled once, and ran once - just as usual

you don't need too, it will be a background process, not a jvm thread. You may want to spawn a thread that wait for completion though

Found it! https://github.com/borkdude/speculative

note that using it in all its fullness may have a significant performance impact

That's fine – I'm actually just looking at it for examples on how to write certain kinds of fdefs

And the ones I write will of course only be instrumented during development

then dive in!

zprint has a lot of options and sane default. For code generation, just building the data structure and writing with pr should be enough. I don't know any code generation tool in clojure (except rewrite-clj but it could be too complex) if it's for runtime only, you can have a look at how hugsql does it with def-db-fns

thanks @U02F0C62TC1. I wonder how I can support comments. ;;

but it's not critical

https://github.com/Cirru/sepal.clj/ looks cool

yeah, it does, thanks for the link

(.addShutdownHook (Runtime/getRuntime) (Thread. #(println "shutdown")))

that's the good one

I found this one recently:

clojure.repl/set-break-handler!
([] [f])
  Register INT signal handler.  After calling this, Ctrl-C will cause
  the given function f to be called with a single argument, the signal.
  Uses thread-stopper if no function given.

not sure shutdownhook will work when killing container though

I don’t know, but clojure.repl doesn’t sound like something to use in production application.

read the source. it’s very simple.

> not sure shutdownhook will work when killing container though yeah and I am curious how much time the app have

10s

(defn set-break-handler!
  "Register INT signal handler.  After calling this, Ctrl-C will cause
  the given function f to be called with a single argument, the signal.
  Uses thread-stopper if no function given."
  ([] (set-break-handler! (thread-stopper)))
  ([f]
   (sun.misc.Signal/handle
     (sun.misc.Signal. "INT")
     (proxy [sun.misc.SignalHandler] []
       (handle [signal]
         (f (str "-- caught signal " signal)))))))

> 10s Is it serious answer?

i think so, but it's configurable

looks like it works. thank you

pretty sure almost no has ever used that :) also, per https://ask.clojure.org/index.php/11576/clojure-repl-set-break-handler-bitrotted I think it does not actually work

❯ clj
Clojure 1.11.1
user=> (clojure.repl/set-break-handler! (fn [sig] (println "received: " sig)))
#object[sun.misc.Signal$SunMiscHandler 0x773dab28 "java.lang.Terminator$1@af78c87"]
user=> received:  -- caught signal SIGINT
received:  -- caught signal SIGINT
received:  -- caught signal SIGINT
received:  -- caught signal SIGINT

user=> ;; i hit Ctrl-c for the above
user=> (System/getProperty "java.version")
"17.0.1"
user=>

i was sending some coworkers my favorites from there and was surprised by it in (dir clojure.repl)

I have been using it to clean up my servers: https://github.com/bob-cd/bob/blob/main/apiserver/src/apiserver/main.clj#L23 but now I'm doubting it. Works well for my needs.

Should I not be using it?

ha. that’s like the one organic usage that http://grep.app found

(answered too quickly with group-by sorry)

group-by shouldn't really be a problem if you use the normal clojure idioms, e.g., (if (seq (get results true)) :something ...)

you could also make a little wrapper around group-by that wraps the pred in a (comp boolean f) so you always get back true and false. Can then return whatever structure you want: a vector of true coll and false coll, keep the map but ensure it has both true and false keys, etc

And or something like this if you absolutely must have all the possible keys:

(merge-with into 
  {true [] false []} 
  (group-by odd? [1 3 5]))

I'd just merge not merge with. Seems extra work