Does this definition make sense? Trying to make a wrapper similar to priority-map-keyfn without falling into pitfalls with custom comparators

(defn sorted-set-keyfn
  [keyfn & keys]
  (apply sorted-set-by
    (fn [a b]
      (let [c (compare (keyfn a) (keyfn b))]
        (if (zero? c)
          ;; break the tie consistently
          (compare (hash a) (hash b))
          c)))
    keys))

Regarding this: https://clojure.org/guides/dev_startup_time I recently tried using clj-kondo metadata to find the subtree’s of ns dependencies that are referenced by my project, but I get weird errors when running with the classes directory on my cp. Stuff like ClassNotFoundException’s. One thing to note: the project uses ig/load-namespaces

I don't know what that is

if you are partially compiling things and missing things you depend on, it's possible to get into bad scenarios. what happens if you do what's in the guide?

I can’t literally do what is in the guide; because the app uses ig/load-namespaces; essentially the require tree is built dynamically by wiring in integrant config… esssentially stuart sierra’s dependency.

ig is for integrant

can integrant be a red herring?

I’m not suggesting it is a herring; just that it and also perhaps the presence of tools.namespaces.repl/clear etc might be 🙂

load-namespaces just ends up calling require — same as ns macro ends up calling require

@alexmiller Yeah ok it makes total sense that compile assumes all transitive requires are also compiled. It’s most likely in my manual pass over the first results that I messed something up regarding transitive namespaces… I had to for example remove some shadow-cljs npm requires; and also some cljs namespaces etc by hand; as it’s a mixed clj/cljs project.

It remembers me of an old https://groups.google.com/d/msg/clojure/NtQ9BJXbziA/AlONiMe9I0AJ discussion related to calling lein uberjar (which adds compiled classes to the classpath) and then starting a REPL. Same class not found error and same relationship with having a component system in place. Never spent time to find the root cause, but now I always lein clean after uberjarring.

There's a lot of fear around aot. Tools namespace explicitly discourages it.

Random ClassNotFoundException was the exact reason I stopped using AOT for third-party libraries.

there is a lot of irrational fear around aot

Yeah I’ve always AOT’d apps, never really had any issues with it; that weren’t issues worth fixing anyway… e.g. side effects happening at compile/load time.

Integrant with dynamic namespaces/components makes it somewhat more difficult though — I’ve been wanting to resolve that for a while.

One thing the startup time link above doesn’t discuss is the pitfalls of doing it… e.g. presumably an app that downgraded a dependency to a prior version might have some problems

is there a way to use namespace aliases to creating a record that is defined in another ns?

(ns app.worker)

(defrecord WorkerA [x] ...)

(pr-str (map->WorkerA {:x 1})) ;; => #app.worker.WorkerA{:x 1}

(ns app.other
  (:require [app.worker :as w]))

#app.worker.WorkerA{:x 1} ;; works

#::w.WorkerA{:x 1} ;; doesn't work

no

you should really prefer using the constructor functions, which are normal functions and can be accessed via aliasing in the same way as any other functions

so here (w/map->WorkerA {:x 1}) would work fine

right. Just checking for completeness. Thanks, Alex

the #app.worker.WorkerA is a syntax for a Java class name constructor and Java class names have nothing to do with namespace aliasing

Definitely I've experienced AOT issues. AOT on your app is fine, uberjar style. But for libraries it's caused us issues in the past for sure

Mostly because compile is transitive

So what happens is you end up if dependency A uses B v1, and you depend directly on A and B v2. Then the AOT jar of A will actually contain the compiled classes of B v1

that does not seem right. by what process would that happen?

that’s an AOT uberjar, no?

No it's not. That happens just if you compile A and jar the classes-dir folder

I’m making a form to upload a file. Simple stuff, only a field for a name, and a field for uploading. I already have some code in place to do this:

[:input {:type "file" :name "uploaded-file" :required ""
                    :accept ".xls, .xlsx, .ods, .csv, .tsv, .pages,
                             application/vnd.openxmlformats-officedocument.spreadsheetml.sheet,
                             application/vnd.ms-excel, text/csv,
                             application/vnd.oasis.opendocument.spreadsheet"}]

In the case where B is not vending itself as AOT

So because B is not compiled

The compile of A will transitively include all of B

And all those classes will go on the classes-dir

And then get Jared as a lib (not uberjar)

And then when you depend on A, you get the classes of B as well. And if A depends on a different B, and because B is a source lib, and Clojure load favours classes over sources you will get the version of B defined by A

yeah, that seems like a problematic way to do aot, but I see how you get there

aot'ing from an app namespace makes you dependent on the classpath, which has already been resolved with version selection (only B v2 is available)

seems like if you are running all your compile's in that classpath context, same would be true right?

Yes, but A specifies a dependency on B v1

what tooling were you using for this or was it something custom?

So the classpath resolves to B v1. And then A is AOT compiled

And then my App depends on A and B v2

but the classpath will not resolve to B v1 here

It won't, but the Jar of A will have classes of B v1 in it

And Clojure Load will see that each namespace on B v2 (whose Jar only has sources) also have a .class on the classpath coming from A's Jar

maybe I was misunderstanding what you were talking about then. are you talking about publishing aot'ed library artifacts? or a process for aot'ing an app?

I'm talking about publishing AOT libs

ah, don't do that

and if you do, they should only include that lib's classes

Yeah this is with custom tooling. But I remember looking what lein does and it seemed it would suffer the same faith

yeah, it is very easy to do it wrong (and it takes away choices from a downstream app builder)

compiled classes are inherently dependent on some version of the clojure compiler

if the lib compiles, the app doesn't have the opportunity to pick that

Ya, that's what I meant by libraries shouldn't be AOTed.

Application can and I've never had issues there

ok, then we are in violent agreement :)

sorry I misunderstood :)

No worries, honestly it hurts my brain everytime I try and rethink about this issue. Dependencies are so hard

For us, we had a case of a dependency doing AOT because it also gen-classes itself so it can be consumed by Java and Clojure

So that was even more fun to deal with! 😅

yeah, that's hard

a couple of the contrib libs actually publish both source and aot versions (the latter under aot classifier)

well, actually clojure itself too has a "slim" version that is source-only

I did face this issue with Clojure actually. When going from 1.9 to 1.10 I think. We had a library that was doing an AOT for a gen-class. Even though it was only compiling the gen-class namespace and that had no requires. Transitively, it would compile the only sources not AOT compiled in Clojure 1.9, the core specs.

And somehow on 1.10 some core macro changed in a way that the 1.9 spec would fail to validate calls to it

So we were getting the 1.9 specs of Clojure indirectly through this.

Was very confusing haha

Anyway, our solution now is our build have you list out the namespace you want to gen-class explicitly. And it will do an AOT of them, and then select only the classes file of that namespace and not of any transitive dependencies. And only include that in the Jar

that's how basically all aot tooling ends up working

that's what we do with the maven clojure plugin for the contrib libs

Ya, I feel I had checked what lein did and it didn't appear to do that. Like it does only AOT the gen-class namespaces. But then it packages everything resulting from that into the Jar

yeah, I think lein is really targeted more towards the app casee

but it's been a while since I looked at it

I should go see what the maven contrib Clojure plugin does. Might be some good learnings in there

https://github.com/talios/clojure-maven-plugin#namespace-configuration

the config is pretty fiddly to make it do exactly what you want

Thanks

https://github.com/clojure/tools.reader/blob/master/pom.xml is the tools.reader config which is fairly complicated as it's both building a source jar, then aot'ing, and building an additional aot'ed jar with the assembly plugin

I do think I'm seeing some of the code that cleans up the classes that weren't for the declared namespaces, seems this does it here: https://github.com/talios/clojure-maven-plugin/blob/develop/src/main/java/com/theoryinpractise/clojure/ClojureCompilerMojo.java#L64

Hello, everyone. If anyone here uses vim, can you tell if it's normal for clojure to have different rules about how words are defined? Currently, for me, w pretty much behaves the way I'd expect W to work, and * and # match along similar definitions? I've had some plugins installed from when I tried clojure years ago and I don't know if I've missed one as I've progessively uninstalled everything Clojure or Lisp related

Did clojure.data.xml not get the 1.0.0 treatment?

Hello, Does anyone know how to add and configure gatling-maven-plugin in project.clj https://mvnrepository.com/artifact/io.gatling/gatling-maven-plugin/3.0.5

Hi all,curl -X GET ' '"https://reqres.in/api/users?page=2' -H 'API-Key:MySecret'

that was not supposed to happen dmanit

damn it±

oooooh that string doesn't look like it should be on a public slack channel 😛

yes - it fortunately only a test one

damn

hopefuly it is gone now?

what a mistake!!!! GRRRR

when I use this on a cli curl -X GET '"https://reqres.in/api/users?page=2' -H 'API-Key:MySecret' I get the correct results in clojure snippet ((client/get "https://reqres.in/api/users?page=2" {:body-encoding "UTF-8" :accept :json :jwt ["API-Key" "MySecret"]})) it fails now it might be the auth-type is a jwt but I cannot tell - I have tried all variations of :auth-type "oauth1" :auth and similar and cannot get it right what am I doing wrong? I have been all over the clj-http github docs but ....

why are you putting what should be a header into a :jwt key?

:headers {"API-KEY" "SECRET"}

:headers {"API-KEY" "SECRET"} results in failuer Key must be integer

hold typo? on my side?

"key must be integer" means that some clojure function is doing a map lookup but you gave it a vector

hello everyone. can someone direct me to whom is in charge of keeping the ubuntu leiningen package up to date?

https://packages.ubuntu.com/source/bionic/misc/leiningen-clojure

is showing 2.8.1

but i would like 2.9.1

will bionic forever be on 2.8.1?

lein is very easy to manage outside the ubuntu package management system: you put a shell script in ~/bin/lein and then use lein upgrade X to make it run a specific version

ok i will try that then!

Yes it works thanks - the docs stated [`"API-KEY" "SECRET"] not {"API-KEY" "SECRET"} many thanks`

but it results in a 401 Unathorised

in your original you had API-Key - is it picky about case?

that is my bad typing

So no it is not the case

I have it consistent in the actual test item

I have tried the :digest-auth [api secret]/ the aws version/ etc.... so it is very frustrating

if the curl you showed works none of that is even relevant

that is what I find odd

I have also used httpie and got a result

Other docs on a sister site stated they use jwt but as you state - the keys are in the header

if you really want quality help, and to value people's (free) time, - Post what the intent was - Post the code you ran - Post the error you received (in full, don't leave out things you don't think consider helpful)

like be really specific, rather than saying stuff like "it failed" or "it worked"

otherwise it's the Miss Cleo school of debugging

thanks, at 9:30 I hope i did state the problem - it is one line of clojure where I am attempting to covert a curl call to clojure and I do not understand how I convert it - It seems the api keys are referred in the header and I am using the clj-http library for the one line of code and attempting to get the syntax right . There is not much else except I receive 401 responses.

As a noobie in clojure I wish I knew how to look under the covers so to speak

If you're really just getting started with Clojure, you might have better luck asking in #beginners -- folks there will show you more patience and will be more forgiving...

ok thanks

This channel tends to assume a certain level of scientific method in debugging etc and some solid ability to read source code in libraries...