Can I do an update that puts the value in another key in a simple way?

@caleb.macdonaldblack mean nested?

update-in

like if I had a key called :ten and I wanted to add 5 and call it :fifteen

change the key name? well, you have to dissoc and then assoc

Nah I still want :ten

want to add a key key, assoc

new key

(assoc m :fifteen (+ (get m :ten) 5))

how are people using CIDER with deps.edn aliases?

Hey, Clojure! You have officially reached maturity. From an HN “who’s hiring?” listing: “some of our legacy code is in ClojureScript”. 🙂

I’ve created a protocol that strips out branches of data structures that end in something which registers as true to empty?. It blows up when I feed it UUIDs. Why is this? I thought java.lang.Object was a kind of catch-all. https://gist.github.com/eneroth/81fe5acf0aab82c355889f28887e08ca

It dies here (filter seq)

Did you mean seq? maybe?

@henrik I think what you want could be done more concisely as something like (clojure.walk/postwalk (fn [x] (cond-> x (map? x) (medley.core/filter-keys #(not (nil? (get x %))) input)

er, cond->>

does anyone not AOT compile their clojure apps? why or why not?

@lilactown I know @seancorfield doesn't 🙂 I don't really see the point in doing so anymore.

I’m not sure I completely understand the pros and cons. I know it improves startup time - is that it?

not just that

it allows you to have a concrete classfile which can be useful when interopping from java in certain cases

ah - like Java calling Clojure?

I used to AOT everything, but that causes problems. Especially if you are releasing libraries. AOT as little as possible, usually just the main class if you need one. I just stopped AOTing my main class for my big project, and now instead boot it with java -cp uberjar.jar clojure.main -e '(do (require 'main.namespace) (main.namespace/main))'

Problems I've seen: functions whose spelling differs only by case collide when compiled on a case-insensitive filesystem; dependency cycles in namespaces cause interfaces or record types to be reloaded multiply, so that two pieces of code reference different versions (sometimes the cycles are caused by REPL tooling which wants to reload things), macros which are expanded based on system or features present create compiled code that doesn't work on other machines.

yeah, I kind of just realized that since I’m using Docker anyways - not even much point to packaging it in an uberjar 😛

we don't AOT. We deploy to docker containers in ECS. We have about 15 seconds of startup time, followed by 25 seconds of downloading data slugs until the healthcheck succeeds.

uberjars are nice for getting all your dependencies in one file and having a SHA256 hash of it for tracking what's running.

I’m interested in understanding what benefits AOT can provide though. like I get the startup time (which is non-negligible IMO) but wondering if there’s any other performance benefits

ah that’s true. dependencies

If you AOT && do direct linking, which isn't enabled by default, you'll get peak performance improvements.

https://github.com/clojure/clojure/blob/master/changes.md#11-direct-linking

otherwise AOT mainly provides better startup time

yeah, I guess one downside to deploying just a docker image is that your container needs to download all dependencies before startup

I wonder if there’s a way to cache those as part of the build

(other than just COPY ~/.m2 😉 )

but then why not just uberjar it I guess

we deploy uberjars and :aot :all, I havent noticed any problems with aot

@lilactown you could make a Docker image with all of your dependencies baked into it that your builds make their images from

that way you just need to rebuild the base image whenever your dependencies change instead of every time someone commits something 😛

that would probably speed CI up at work quite a bit. :thinking_face:

Ah in that case that piece of valuable advice will be $800 😉

do you accept IOUs? 😬

well i do but that also means you owe me a free t-shirt.

boy, I did not understand the terms going into this conversation! :rolling_on_the_floor_laughing:

haha

seriously though, thank you and everyone else for answering my questions 🙂

@lilactown https://twitter.com/smashthepast/status/1010967492393623553

FROM openjdk:10 as jdk
RUN jlink \
   --strip-debug \
   --compress=1 \
   --vm=server \
   --output /tmp/linkedjdk \
   --add-modules 

# multistage docker build
FROM debian:stretch-slim
COPY --from=jdk /tmp/linkedjdk /jdk/
COPY app.jar /app/
ENTRYPOINT ["/jdk/bin/java"]

The whole docker image should be between 70-150MB depending on deps

nice! I’ve been using the clojure:alpine images and have gotten about the same size

those images pin you to java 8 though...

true. I haven’t had any need to update yet; I’ll might have to do that soon though, so then I’ll probably copy what you’re doing 😄

graalvm native-image + scratch = 10mb images 😮

Graal native-image is cool, but it imposes severe limitations that the community rarely mentions

it completely trades away dynamism

many widely used jdk modules are not supported by native-image

yeah it's not for everything, but right now i'm messing with serverless stuff and it's great for that.

it is hard to get a non-trivial app to compile under native-image.

yeah I went down the GraalVM rabbit hole a month or two ago. it’s fun but unwieldly

makes total sense for serverless and some CLI apps, but didn’t fit my usecase

jaotc is another option with fewer caveats than native-image

with clojure.test fixtures is it possible to pass an arg to the function to run?

(use-fixture
  :once
  (fn [f]
    (do-setup)
    (f some-args)
    (do-teardown)))

no, but you can use a dynamic var

This might not be the channel for this sorta thing, but want to say that I’ve been playing around with clojure for a prototype and it has been wonderful 😛