Clojurians Log v2

As in, using TCO? Or using no calls at all?

If you mean using no method/function calls at all, the only thing I can think of is the one I mentioned earlier -- implement your own stack and loop while pushing/popping elements of that stack at appropriate times.

I mean without using recursion

And also without using a stack, since the example doesn't need one

I guess it would just be a for loop where you mutate the collection being looped over as you loop over it

That example definitely needs a stack that has a depth proportional to the depth of the directory structure of the file system, unless I completely misunderstand the code example.

Or, I guess in the proposed Clojure implementation, the use of an explicit stack was avoided by just keeping one big collection of all files and directories not "visited" yet.

The total memory storage required is the same as if it had been done with a stack of parameter values.

In my experience, most places where mutual recursion shows up naturally in code, without forcing it, the maximum call depth is "small", as in the case where the Java code above has a maximum call depth limited by the maximum depth of the file system directory structure.

So TCO is not really needed.

https://twitter.com/cgrand/status/1350063059864346624 👀 thoughts?

Is there a way to pretty print transients in the repl, so I see the actual content and not just the type name?

@rob703 I'm realizing that the reason I don't have an answer to this is that I've always taken transients as a hack (14% or so speed increase), so as soon as I need to debug something I stop using them

someone else might have a better answer

that's exactly my problem now. I have an infinite loop i need to debug

might be easiest to call persistent! while debugging - that makes the next transient op blow up, but it prints clearly

(of course you can create a new transient to use in the place of the old one too)

That's what I did for a while but I find it highly annoying when everything blows up while debugging I'd be ok with "copy-transient-and-print-it"

(unless you are bashing in place, which means you are using transients wrong and that's the likely cause of your bug)

(defn debug-transient [t message] (let [p (persistent! t)] {prn message p) (transient p)) - something like that should be safe, it gives you a new transient back to use with the same contents

the only reason you couldn't use that is if you are using the transient but not using the return value of conj! / assoc! etc., which, in this case, is the probably source of your bug

Hm fair enough, if the sort order remains the same that'll be good enough for now. I made sure to use the return value, so that's fine. The real problem is that I needed to translate a really hardcore imperative algorithm into clojure/fp and it's gotten real ugly (who would have thought)

but more generally, I'd be suspicious of using transients for anything complex enough to need debugging - use normal collections then rewrite with transients when the tests pass

Yes that's what I'll probably do now

if you think transients will help you do an imperative algorithm, you're already on the wrong track

I've chosen them only for speed reasons, instead of an atom... I should have chosen the atom though it seems

the idea that a transient could replace an atom is already an error

they have the same api as normal collections, just different function names

they accidentally let you update in place, but that's not a defined behavior, it's just incidental - they are allowed to do so, and never promise to do so

Yes I found out afterwards and rewritten my code accordingly. But that's the thing. When I want to speed up code that is using an atom right now, I have to rewrite it anyway in order to use transients properly. So I figured I might as well go all in and write my algorithm only with transients. So yah, now I'm here, trying to debug things 🙂

but thank you, I'll try your debug snippet

@rob703 the thing that's meant to be an atom but faster is volatile! - but that only works in single threaded code

but, there's a pattern here - as with transients you get something more brittle and harder to debug, that has the same usage pattern

which means the effective way to use both is to use normal collections and container types first, get your code working, and if it's still too slow swap in a volatile! or transient if it's apropriate

I see, thank you. I'll keep that in mind as a hard earned lesson. I think my main problem was half-assing the translation from the oop/imperative to the clojure version. So I gotta clean that up now

Edit: moved to #ring

@kwladyka I'm not sure if this question belongs in #clojure - also please don't flood the channel with JSON

I mean, you can make this into a gist perhaps.

ok I will ask on #ring then

Hi! I want to build specs from a sequence of maps. Of course, spec/def being a macro, I cannot just map over the sequence. I could achieve it by using a macro myself, but I was wondering if it can be done with functions only. There is spec/def-impl, which is a function, but its docstring discourages its use… any ideas? (I'll totally take no for an answer.)

Try malli instead

Thanks, taking a look at it right now…

@services yeah, there is no way doing that without a macro, or using the discouraged impl

technically it is possible, the most straightforward way being using eval

yeah, macro or eval, kind of the same thing

in theory spec2 is supposed to make that sort of thing easier

but what I would do (and do all the time), is just write some code to generate the s/def forms, run that code in the repl, and then copy and paste those forms back in to your code

(assuming the maps are fairly static)

It's a data-oriented Spec replacement, basically, made by metosin who also made spec-tools: https://github.com/metosin/malli

Got a weird question: I am trying to format a request with Cheshire to have a key "\$oid" , but it only seems possible to get \\$oid. Is there a way to escape $ solely?

you are likely being confused by printing at the repl

When I actually send the request it still is \\$oid

how do you know?

there is no clojure syntax to escape $ in a string, because it's not special

\ prints as \\ unless you are printing for display

user=> "\\$oid"
"\\$oid"
user=> (println "\\$oid")
\$oid

user=> (seq "\\$oid")
(\\ \$ \o \i \d)
user=>

When I actually made the request, the key for the JSON object was still \\$oid

do you mean the it was serialized in the json that way?

because the same escaping happens in json

the string that prints normally (with println) as "\\$oid" would print for reading (with prn) as "\\\\$oid"

s/register can be used for this in spec 2

I guess I am not trying to escape the dollar as I am trying to escape the slash. I need it to be received \$ because in another env I need to escape the dollar.

@grant.isom no clojure string will ever prn as "\" undoubled, and the repl uses prn

@hiredman, @borkdude Thanks to both of you! 1. I'm on ClojureScript, so no eval for me 😔 2. The Sequence is not static, since it is supposed to be edited by my customer who wants to compile the app on his own and generate different versions (we'll see how that works out). Altogether this is not super important, since I already have implmented a working macro-solution… just thought I might simplify it and gain a better understanding of spec @simongray Malli sure does look like an interesting choice fo my use case. One of my goals is to match an existing Swagger2 Schema… which apparently Malli schemas can be transformed into…

@services swagger-like JSON API things is really where Malli probably shines

as malli also supports coercions out of the box, and spec doesn't

and for other reasons maybe too

Okay found out cheshire was escaping my escapes haha

Thanks @noisesmith and @hiredman!

@simongray, @borkdude Looks like the Malli Readme will be my bed-time reading for tonight…thanks! For the record: what I'm trying to do is to declare the constraints on my data in one place and then impose them on my API-request as well as on the input fields in my form (e.g. max-length attributes and field validation)

I haven't used Malli myself, only spec, but it seems like it would fit your use case.

@services See also #malli

Hi everyone! Is there a guide or some sort best practises when it comes to deploying clojure apps to the cloud ? I'm looking at AWS and was wondering if there is some sort of step by step guide to the setup including jar/war creation, deployment and monitoring it... maybe even a CI/CD setup. I've a local docker based setup for experimentation, but I've not tried a production build/setup as yet. Any pointers in this direction would be appreciated. Thanks in advance 🙂

my best results always come from following instructions for java, but using lein or clj to build the fat jar instead of maven

@dionysius.almeida This article appeared a while ago: https://circleci.com/blog/deploy-a-clojure-web-application-to-aws-using-terraform/ (since you asked about AWS)

this generalizes to google cloud, or desktop apps, etc.

how do you deploy ? Do you use containers or something else ?

this is not for work...just experimenting as I'm doing a cloud certification course and wanted to do a clojure deployment as part of the learning process

that varies based on how devops should work

the simplest thing IMHO is to wrap the java service in a daemon

at a certain point of complexity, using docker or whatever plus a container mainagement infrastructure becomes simpler

I've never seen that happen for a personal project

that's true. I'm loathe to setting up a EC2 instance and then install jdk etc and setting it up as that's a lot maintenance further down

you never need a jdk

i meant jre to run the app

right - but a good clojure app never needs more than the jar itself and a jre

anything else means I'm probably doing it wrong

so what's the lightest setup you would recommend ?

(eg. depending on local resources in a fs when I should be using a service for storage)

the lightest setup is a fat jar, a jre, and a systemd script that starts java

if you want "easy" you can maybe try Heroku. If you want complex (load balancers, Terraform, etc) look at the article I posted in the main channel.

yeah - I really don't like the heroku approach, they use git and lein on the server

@borkdude thank you, will check it out

I'm trying to suggest simple rather than easy or complex

i am looking at simple setup which can be scaled in and out based on workload

I've been deploying personal projects as java -jar uberjar.jar and init scripts (now systemd, I guess)

and have databases etc as managed services separately

yeah, sounds like you want some real infrastructure then - I haven't used terraform in anger but it's definitely meant to do what you want

it's a complex pile of brittle crap also (or else someone set it up badly...)

you could also try datomic cloud. your app will live inside the db (as a managed service) ;)

ah i see .. that's interesting. But for experimentation, i would rather stay out of datomic cloud, as I've heard it's quite expensive and the setup is difficult as well

but thanks for all the tips guys 👍:skin-tone-3:

I haven't seen any mass scale app orchestration that isn't hard to set up and expensive - I think it kind of comes with the domain

@borkdude yeah that's what I want to do as well .. If you could share notes of your setup, that would be nice

sorry, I removed the msg. I said: we are migrating from bare metal Docker swarm to AWS docker swarm

yeah i read that before you deleted it 😉 . thanks for the link you posted. Will go through it

I haven't personally done this migration, so I don't have any notes right now. The reason we are moving away from bare metal has to do with someone leaving our team, so we have no choice

but the bills will probably be a lot steeper

there are options with terraform to have config driven cluster spinup

i guess there is always a price for convenience 🙂

@noisesmith thanks .. will research terraform and docker options for deployment

another concern is that docker was not designed for security, newer kubernetes addresses this by using an alternative container format with a compatible config

@noisesmith i see. thanks for the tip

also, even with a container, it's still better to run your build tool and package a jar into the container, rather than making a container that runs your build tool on startup

i agree. that makes more sense

(defmacro functionize [macro]
  `(fn [& args#] (eval (cons '~macro args#))))

(let [a [true]] (apply (functionize and) a))
=> true