do you understand the error message ?

the java.security.SecureRandom kryo serialization issue is probably separate from clj-http/request var serialization

maybe I do things wrong, but do I need to upload my #‘var everytime it changes ?

It slow down my feedback loop, it’s okay but I am just asking ^^

mount! has :live true

but haven’t gotten to putting that into mount-ring!

thinking about lifting common pieces from mount! for use in mount-ring

ultimately these things are related, mount! is just a special case of proxy

does it upload everything ? * :live true

yes, but only on change of the mounted var, not on change in dependencies

ok, nice

one can “batch” changes that way 🙂

have you looked at my keep stuff, totally wrong ?

haven’t yet have time

ok ok

I have few mini-websites that are going to massively use portkey

at least I am going to try hard

this tool can be life changing for prototyping

🙂

yes

it’s even better than lambda with regular java/node

repl makes a difference

indeed

we will have to get rid of carbonite at some point

why that ? it’s just a kryo wrapper for clojure datastructures right ?

because for some clojure stuff we need our own custom serializer and in such cases we clash with carb

ok ok

@baptiste-from-paris I’m rather bad at faces, have we ever met at the paris clojure group?

so the issue with clj-http.core/request is a kryo one, and trying to fix it (by working around carb) I trigger another one...

nope, I discovered clojure 1 year ago

but i am part of the clojure paris group

we just had a brainstorming to 3 weeks ago

trying to make Clojure just a little more popular in france !

by the way if one of you want to come to paris to talk about portkey it would be awesome

I am sure we can make it all inclusive with the Clojure user group

back in july (iirc) I chatted with @hmadelaine about “monter à la capitale” (going to the big town) to talk about either portkey or unrepl

when ever you want, we plan to make 6 events this year

monthly was too much of a strain?

which is quite hard, there are not a lot of people willing to talk ^^

yes, not enough speakers

and lot’s of people who are really new to the language

I bet there are less than 5 companies using clojure in France

I really want to increase this number

we need to put parens in tap water like they do in Finland 😉

parens means ^^ ?

parenthesis

oh

lol

the uptake of Clojure there is phenomenal

sometimes it’s only a matter of few people/companies that make other people choose a tech

it really amaze me how tech companies choose their stack

I worked for Meetic (Match group); we always did what competition were doing

and vice/verca

Imitation and belief in magic bullet/pixie dust.

However we (collectively as devs) don’t always make rational choices

Offcourse! We are humains ^^

But I think it can be great to have more clojure enthousiast in France

At least just to inject more money in the clojure ecosystem

So that a 28/30 young dev can work on a project like portkey because he already "master" clojure

do you often come to paris @cgrand?

no — for 1.5 year I contracted for SACEM and was in Paris 2 days a week, this ended 2 years ago and since then I didn’t have the opportunity to go to Paris. It has been too long of a time, I should plan a visit to the user group.

nice 🙂

but currently I’m looking into going further in the country side, maybe I should open a (clojure) hackers retreat 🙂

lol

saint etienne right ?

the uptake on Clojure in Finland is probably most visible in ClojuTre, this year and last year attendees in the ~300 range

And how many from Finland itself ?

350 this year, 2/3 Finns I guess

to put it in perspective: same size as EuroClojure

it’s a free conference

this year there was a voluntary 50e patron fee

but there are quite many sponsors

which tells the story of Clojure being commercially in use

and we have sauna 🙂

we have great wine 😉 🍷

UCInterpreter transform clj to .class right ?

I don’t understand also why use vars & classes in here =>

(defn freeze [x]
  (let [classes (atom #{})
        vars (atom #{})
        kryo (mk-kryo)
        baos (java.io.ByteArrayOutputStream.)]
    (with-open [out (com.esotericsoftware.kryo.io.Output. baos)]
      (.writeClassAndObject kryo out x))
    (.toByteArray baos)))

and from what I understand, it might cause problem according to kryo doc

[com.esotericsoftware/kryo "4.0.0"]
                 [com.twitter/carbonite "1.4.0"
                 :exclusions [com.esotericsoftware.kryo/kryo]]

https://github.com/EsotericSoftware/kryo#versioning-semantics-upgrading

thx

Deps tracing/tree-shaking

One starts from a value (a closure), we serialize it with kryo.

To serialize, Kryo performs a traversal of the object graph rooted at our initial value.

As kryo walks the graph we log all classes and vars found. (that’s what you see in freeze)

so now with have bytes (the frozen value) and some deps (classes and vars)

log-dep ?

we add the classes to the package (and we add also deps of deps, that’s UCInterpreter/analysis job, vars and resources may be found too)

then we add resources

then we deref all vars encountered, and we serialize their values

(with kryo)

serialization produces more classes and vars

etc. etc. until nothing new is found

FIN

how did you come with this strategy ^^ ?

and in which phase asm generates .class ?

never

when you eval code at the repl clojure does generate classes (in memory)

we use ouroboros to retrieve them (or any other) from the JVM

let me summarize the problem(s) portkey is trying to solve 1) bring clojure repl to lambda 2) minify/deploy clojure code -> bytecode -> package -> aws lambda

that’s correct but 2) is somehow necessary to make 1) useful

and I guess the hard part is serialization and handling clojure’s data structure ?

not that much, the hard part is class inspection

ok