can i have a record implement multiple protocols? example pls 🙂

https://clojure.org/reference/protocols <-- see section "extend"

I mean, this looks like a function that takes any arbitrary function, and returns another anonymous function, that has some state in the form of an atom

without trickery, no

although most people would recommend not writing fib as a recursive function in the first place

specifically in line 13, when function 'f' is applied, I wanted to know if the recursive calls are also memoized

no, they will not be

those internal calls will be calling the original function

memoization is tricky to use with recursive functions

but again, most people would suggest trying to avoid recursive functions

you can write fib as a tail-recursive loop/recur instead

then memoizing it is fine

okay that makes a lot of sense @alexmiller! got it, I'm not going to use this in production - just for educational purposes 🙂

if you google for it, there are a large number of fib variants out there in Clojure

okay - another quick question - if I repeatedly call the same function - how is the original function f not invoked again? How does the atom persist across multiple function calls?

for example if we do (defn mfoo (memoize foo))

if you notice there is a let around the creation of the fn

and I invoke (mfoo bar) multiple times, how does the second function call access that cache?

the fn closes over the local binding mem

so the function itself contains a reference to the atom

it’s a stateful function

so for somebody coming from Object oriented languages - atom is like a global variable?

kind of in effect here, but not really

not really? can you please elaborate?

what’s actually happening under the hood is that when you create a fn it creates a Java class

that class has a field that corresponds to mem

the instance of the class that you get back has the mem field populated with an instance of an atom

that field is not publicly accessible - it’s a private field of the function class

to “use” it, you have to invoke the function which controls access to it

okay - makes sense. It's a kind of a global but only visible from within the function, kind of like class static

only accessible by instances of that class

in c# or java..

yeah, but it’s not static, it’s a private field

got it

okay - so I tried running this example https://clojure.org/reference/atoms

It looks like this example is misleading - it claims that without memoization, the recursive function takes more time

and with memoization - the very first call takes an order of magnitude less

what am I missing here?

I agree that that’s a misleading example :)

all it’s doing is memoizing the single outer call, not the inner recursive calls

well, actually that’s kind of tricky as it’s actually rebinding fib to the memoized fib

depending on your compiler mode (direct linking) that won’t actually work

so it will work in the case presented there (but with some caveats)

seems to work for me - I haven't tinkered with the compiler mode (default) - but I see no difference in times

thank you, this makes a lot of sense. This was driving me nuts for the last 2 hours. peace is resetored

How can I fix this example? Is all this documentation hosted on github or some place like that?

so like I corrected above, this is actually doing what it says

it’s just not particularly obvious how

whoa, it's memoizing the recursive calls?? how?

(def fib (memoize fib))

how does using the same name make it any different?

def will find the existing var if there is one and re-bind it’s root value (the function)

so is this actually replacing the var fib with a memoized fib

my goodness - is the internal function also the memoized variant?

and the internal call resolves the recursive call via the var (which has changed)

yes

(assuming you’re not using direct linking, which would not do that)

I want to buy you a beer

no need!

btw, the docs are hosted in github at https://github.com/clojure/clojure-site and issues and prs are welcome there

okay - I will expand this example to include a little bit of this discussion, so it helps noobs like me

if you mean on this atom page, I would keep that brief, but happy to take a look. if you’re sending a PR, you also need to sign the contributor agreement https://clojure.org/community/contributing_site

thanks!

just one final question - so rebinding the memoized function to itself is then a neat trick to memoize recursive functions?

yes

the big thing to always remember with memoization is that you have created a place to put unbounded unmanaged state

aka a memory leak

this was just posted: https://techblog.roomkey.com/posts/memory-leak.html

it’s exactly that

is it not possible to put a limit on the size of the atom (kind of like chan buffer size in core.async)

there are contrib libs (core.memoize which uses core.cache) that give you memoization but also give you control over the cache, so it’s this idea but productionized a bit

no, you can’t limit the size of the atom

or make it safe by checking the size of the atom inside the memoized function

yeah, that’s basically what core.memoize adds, with the option for eviction, expiration, etc

what I want is a (fixed-sized-atom {} ) - I'm thinking I can write a macro to achieve this easily?

I will try that - good exercise for me

thank you - this discussion has helped me a lot

I don’t think you need a macro for that - a function would fit the bill fine

I'm taking the Up And Running With Clojure course on http://Lynda.com. The person who created the course (Arthur Ulfeldt) suggested I join slack, to get some help with my init.el file. The course has a file downloads section, of which contains the code to pull down the files to setup emacs as a Clojure IDE. Well, apparently the settings in the init.el file are bad, and Emacs displays an error message about retrieving a package. I'm very much new to Clojure, Emacs, needless to say lisp & init.el settings

@cappy2112 you can can ask here: #emacs 🙂

clojure.spec question: is it possible to change what is returned from conform'ing an input? I have a string in format 'e1234' (so 'e' prefixed to '1234'). I'd like to have '1234' returned when calling conform on that string, is that possible (or idiomatic at all?)

I thought about making a predicate that returned 1234, but that didn't seem to work 🙂

@kah0ona you can wrap your predicate with s/conformer

tx!

Oh how nice it would be to have this patch merged: http://dev.clojure.org/jira/browse/CLJ-1496

Get rid of all those (throw (ex-info "Boom" {}))

@pesterhazy i do agree, on the other hand is it not that difficult to wrap exinfo yourself?

@cappy2112 I would suggest to not learn clojure and emacs at the same time. Pick an easier to setup IDE like Lighttable or cursive.

@cappy2112 , checkout spacemacs. lots of stuff pre configured

is anyone aware of how I can setup compojure to write the log msgs into a file instead of stdout??

@plins this is usually configured at the level of your logging system, not Compojure

well theres some logging automatically happening on stdout when I start a compojure-app, any ideas where i should look for instructions?

compojure-api app*

plins what does that logging say?

it is probably the underlying java server that is logging stuff

@plins usually on the JVM you need to configure the logging, depending on your library. Mostly its a configuration file in the resources folder like so: https://github.com/sveri/closp/blob/master/resources/leiningen/new/closp/resources/log4j.properties

You could add a file appender there too

$ lein ring server-headless
2017-02-10 11:28:05.317:INFO::main: Logging initialized @1286ms
2017-02-10 11:28:14.157:INFO:oejs.Server:main: jetty-9.2.10.v20150310
2017-02-10 11:28:14.208:INFO:oejs.ServerConnector:main: Started ServerConnector@66b561f1{HTTP/1.1}{0.0.0.0:3000}
2017-02-10 11:28:14.210:INFO:oejs.Server:main: Started @10179ms
Started server on port 3000

@plins here is one with a file appender: http://pastebin.com/hjQY3qCJ

thanks! 🙂

Although I have to admit I never tried it 😄

I need to write a parser for a file. I've heard good things about instaparse, but I've never used it. Would y'all recommend it, or is there a different library/approach that people are moving to instead? Would anybody recommend using clojure spec for this? (Happy for threaded replies to keep chatter down on the main thread.)

@jeff.terrell I’d recommend instaparse. I think it’s maybe the best parser I’ve used.

Does anyone know of a library that contains a function to lowercase the keys in an existing map of values? The problem is that when destructuring a map with key of :thekey the actual key might be case-insensitive (e.g. :theKEY or :Thekey or other permutations)

perhaps Specter...

@aaelony used this, can’t complain: https://github.com/qerub/camel-snake-kebab

@aaelony

boot.user=> (require '[clojure.string :as s])

boot.user=> (defn mlk [m] (into {} (for [[k v] m] [((comp keyword s/lower-case name) k) v])))
#'boot.user/mlk

boot.user=> (mlk {:a 42 :B 34 :some-Key 28})
{:a 42, :b 34, :some-key 28}

@aaelony medley (http://weavejester.github.io/medley/medley.core.html) has a map-keys function which you could use.

(meta (assoc (with-meta {} :a) :foo :bar))

Thank-you @jromine

@aaelony: or reduce if you care about number of iterations over it

boot.user=> (defn mlk [m] (reduce (fn [a [k v]] (assoc a (keyword (s/lower-case (name k))) v)) {} m))
#'boot.user/mlk
boot.user=> (mlk {:a 42 :B 34 :some-Key 28})
{:a 42, :b 34, :some-key 28}

you'll need something a little better I'll bet

{:KEY {:NESTED 1}}

thanks @tolitius, @dpsutton string values is fine for now.

i wasn't talking about string but about nested keys

wasn't sure if you needed to descend into maps and check for case or if just a top level sweep was enough

the values are mostly numeric in my specific case, it's just that the key names can be case-insensitive

correct

my point is if you have a "flat" map or nested maps

if you need to walk your map, descending into nested maps and standardize the keys, you'll need something more along the lines of this:

(defn keywordize-keys
  "Recursively transforms all map keys from strings to keywords."
  {:added "1.1"}
  [m]
  (let [f (fn [[k v]] (if (string? k) [(keyword k) v] [k v]))]
    ;; only apply to maps
    (postwalk (fn [x] (if (map? x) (into {} (map f x)) x)) m)))

awesome. Yeah got it. thanks for swift responses.

all good

👍

Anyone have any links or guides on best practices for using refs? I’m using one as an in-memory DB and am curious if there’s any gotchas I should be aware of

Examples of using refs in the wild would be nice too

@aaelony (if nested needed check out specter):

boot.user=> (require '[clojure.string :as s])
nil
boot.user=> (use 'com.rpl.specter)
nil
boot.user=> (transform (walker keyword?) (comp keyword s/lower-case name) {:a 42 :b {:and-thIs 28 :c {:some-Key 34}}})
{:a 42, :b {:and-this 28, :c {:some-key 34}}}

thank-you

@jsselman riemann uses refs

It's the only open source project I know of that's using them

"serious project"

Looking now, thanks

I’m guessing refs aren’t that popular because most of the time atoms will do, but when they don’t it’s preferred to just use a database?

On closer look, it’s more than I’m using a ref where a DB makes more sense; for example I need to persist the data across deployments, take backups, etc...

@tolitius, in the off-chance of duplicate keys, I want to concatenate.. e.g. {:ABC 100 :aBC 200 :abc 300} => {:abc "100,200,300"} will need to add that to the specter version

if the map is flat you can just use reduce (to get you started):

(defn mlk [m] 
  (reduce (fn [a [k v]] 
            (update a (keyword (s/lower-case (name k))) conj v))
          {} m))

=> (mlk {:ABC 100 :aBC 200 :abc 300})
{:abc (300 200 100)}

A Clojure book I co-wrote is on sale today. Is that appropriate material for #announcements or too commercial/self-promoting?

sure

at least in my opinion (I’m not an admin here)

Can I use you as a reference later 😉

I think it's probably fine, too, from the channel description

if anyone here has submitted a talk to a Cognitect Clojure conf in the past, I have a question on #conf-proposals for you

in clojure, is it possible to xreate an object x such that (identical? x (:self (meta x))) returns true ?

@qqq

(let [x (atom nil)]
  (alter-meta! x #(assoc % :self x))
  (identical? x (:self (meta x))))

@favila: the user of alter-meta! is clever; thanks!

hmmm. leiningen is suddenly complaining that

Release versions may not depend upon snapshots.
Freeze snapshots to dated versions or set the LEIN_SNAPSHOTS_IN_RELEASE environment variable to override.

trying to do a lein uberjar

@aaelony i think that happens when you're referencing something you shouldnt in your deps in your project.clj

i think

hmmmm. ok. wonder what that might be...

export LEIN_SNAPSHOTS_IN_RELEASE=override; lein uberjar should allow you to actually get your build to run if you're in a rush, but you should probably check out the output from lein deps :tree to see what snapshot you have in your project

thank-you @tanzoniteblack, that got my build working. What would I look for in lein deps :tree ?

try searching for -SNAPSHOT

nothing SNAPSHOT there as far as I can tell though

if you don't mind sharing the output of that command, I'm happy to take a quick look at it

ah, I see Possibly confusing dependencies found: ...

that's normal when your dependencies have some of the same shared dependencies, but require different versions of them

not the direct cause of the snapshot issue

ok. will override for now until I get a chance to dig in

many thanks

if you're ok with putting the output of the lein deps :tree command in a gist on github, I'm willing to take a minute to look over it if you'd like

sure

version ranges can also get you into this

how do you disable logging for your test namespaces, with either tools.logging or timbre?

I don't

my preference is to have separate logging configuration in tests that redirects log output to a log file

how do you do that?

it will depend on how you are logging

my current project uses tools.logging and log4j2, so I have a log4j2 configuration file under test/, and because it is under test/ it is used when running tests or a repl or whatever in development from a checkout. I have another log4j2 configuration file that is baked in to the docker container with the uberjar for deployment

is it possible to have one for the repl and one for tests?

it would be tricky, and the exact way to do would depend on what build tool you are using

a related question: right now I’m doing something like (clojure.tools.logging/info msg), but what I want is (clojure.tools.logging/info a-logger msg), so I can inject a null logger in the tests

doesn’t look like either tools.logging or timbre has such an API though

why do you want to do that?

so can control logging through dependency injection, not through config files

shrug

anyway, tools.logging doesn't do that, so you'll have to use something else

heh, to each his own, I want to gauge my eyes out when dealing with all these Java logger config files 😛

that is what everyone says

the thing is, it takes maybe a day tops of messing around with it, then you never touch it

"10. Get into a rut early: Do the same process the same way. Accumulate idioms. Standardize. The only difference(!) between Shakespeare and you was the size of his idiom list - not the size of his vocabulary."

+1 day everytime you want to tweak anything

nope

I have a set of logging dependencies and configuration files that I just copy around everywhere. Some times I do have to update them when, for example libraries started using log4j2 instead of log4j

and it just works, and I shake my head in confusion and wonder what everyone else is doing wrong when they get so annoyed with logginig

Does anyone have experience using clojurefx with an FXML file? I'm having an issue where it loads the FXML, but keeps throwing this exception on load: CompilerException java.lang.NoClassDefFoundError: Could not initialize class javafx.scene.control.ToolBar, compiling:

are there any serious alternatives to tools.logging and timbre in the clojure logging department?

fwiw, I “solved” my problem by creating a test_helpers.clj containing the line (timbre/merge-config! {:appenders nil}) and requiring that in each test ns

which I guess might break horribly in some way I’ll find out later ¯\(ツ)/¯

aaaand that is right now

I was having a look at Transit

but I find this sort of CONTRIBUTING file really strange https://github.com/cognitect/transit-python/blob/master/CONTRIBUTING.md

very odd policy for an open source library and even the test files that are needed to run the tests are not in the repo

I probably won't consider transit for Python if I can't even run tests for it, which is a shame since it looked quite interesting

Lets say I have a vec with x elements and want to get the 4 elements from it at arbitrary positions, how would I do that? Basically I want something similar to select-keys, but vor vecs and positions.

nth?

(map #(nth my-vec %) [2, 5, 6])

ok, thats pretty succinct, thank you 🙂

make sure you're aware of the fact that nth will throw java.lang.IndexOutOfBoundsException if the number is out of bounds

(common source of bugs whenever I use nth)

(map #(nth my-vec % default-value) [2 5 6]) will prevent the exception from being thrown

Yes, I know, thanks again 🙂

I'm not sure that I knew that nth took a default value before just looking that up 😄

Isn’t nth a linear time lookup? With a vector you can use get.

Or just the vector itself as a function.

(map my-vec [2 5 6])

if the coll is indexed, nth will do a lookup by index

so not linear

@weavejester you are absolutely correct that that would work as well. And possibly be more performant. I generally try to make sure that all of my code will work with either a list/seq/vector instead of making the code specific to one

@alexmiller Oh really? I thought it was a seq operation and O(n)

that way I can switch between a vector and a sequence produced by having done map/`remove`/`filter` on it without having to change my code

nope

Has it always been that way?

nth is super weird

yes

Huh. TIL 🙂

it has many special cases

nth is a pre-protocol protocol

well, that’s not the weird part imo

as other coll functions do similar

it’s weird in that most coll functions give you a perf expectation and are not implemented when they can’t

nth does an indexed lookup in a bunch of special cases but falls back to sequential lookup

so it kind of is both an indexed collection function AND a sequence function

so is there a standard way to make a *in* reader with a non-determinant length sequence that I can write to?

https://github.com/clojure/clojure/blob/clojure-1.9.0-alpha14/src/jvm/clojure/lang/APersistentVector.java#L170 apparently there's not really a difference between nth and get on vectors 🙂

I have been killing myself trying todo this.

I wrote a custom inputstream with a blockingqueue

thinking I was clever

that didn't work out very well

@bhauman: PipedInputStream + ByteArrayOutputStream?

I tried pipedinput stream with pipedoutput stream

that worked but it had strange timeout exceptions

it wanted it to be alive and not wait to long

it still worked

@weavejester I'm going to give it a try

You should only need PipedInputStream, not PipedOutputStream… I think!

Bytearrayoutputstream automatically grows!!

geez

yeah 🙂

Can anyone see the problem with this code:

(defprotocol Logger
  (log
    [logger level event]
    [logger level event data])
  (log-ex [logger level exception]))

(extend-protocol Logger
  nil
  (log [_ _ _])
  (log [_ _ _ _])
  (log-ex [_ _ _]))

user=> (log nil :info ::foo)

ArityException Wrong number of args (3) passed to: protocols/eval1345/fn--1346  clojure.lang.AFn.throwArity (AFn.java:429)

Am I going mad, or should it not do that?

It works with four arguments.

ByteArrayOutputStream doesn't do it

😞

@bhauman I think my recollection of pipedinputstream is inaccurate

@weavejester I appreciate the help

Hm, weird, protocols with variable argument lists work on records, but not on nil.

it's kinda interesting how difficult this is to do

@bhauman: So a pipedinputstream+pipedoutputstream didn’t work?

it doesn't like to hang for long periods of time

There’s nothing in the docs to suggest that piped streams have timeouts.

I know 🙂

it may just be internal bookeeping

because the behavior doesn't change

can defmulti have a :post clause?

I found the solution to my protocol problem though. Turns out that the syntax when using extend is slightly different to the syntax when inlining protocol implementations in a vector when it comes to overloaded functions.

Sometime I need to write a better-multimethods library 🙂

I remember there being a good reason why multimethods had pre/post conditions on the defmethods rather than defmulti but can't recall it off the top of my head

It might be another implementation detail; the defmethods just attach functions to the multimethod object.

And pre/post are implemented on the fn

I have a fun puzzle at the moment, I have an atom that has some vectors in it... I want to grab the entirety of one vector that has a specific key... how can I "grab the whole thing when just knowing one attribute" ? let me know if that's not clear enough a query...

@weavejester not sure about that not being a deliberate design choice tho

@sova how are the vectors stored?

note that you can have pre conditions on the multimethods by placing them in the dispatch function

@bronsa Yeah, but again, isn’t that because the dispatch function is a function, and therefore has pre/post?

speaking of better-*, I discovered better-cond the other day and found it pretty nice to use

I'd say that because of the polymorphic nature of multimethods having post condition for each method makes more sense than havingn one to handle all the future defmethods

IE it's the only way to allow the multimethods to be open in its return type

Yeah, but no reason you can’t have both.

you could a single post condition for multimethods with a multi-spec

Also multimethods don’t play nice with namespace reloads. Haven’t figured out a good way of solving that one, though.

remove-ns 😕

@wei doesn’t work, unless you remove the namespace that has defmulti. The defmethods aren’t attached to the namespaces they’re defined in.

@weavejester (def my-multi nil) (defmulti my-multi dispatch-fn) is what people usually do

:c

Right, but again you need to wipe the defmulti

yeah

to get around the defonce-like semantics

My current theory is to have better-multimethods create hidden vars in the namespaces they were defined in.

I think the counter argument is that if you're redefining a defmulti, all the defmethods should be invalidated because you've changed your dispatch funcntion

and the dispatch value of your old defmethods might not make sense anymore

(CLOS has a way of dealing with that IIRC but we're nowhere near that level of sophistication)

i might be completely misunderstanding what you're proposing tho

Like:

(defmacro defmethod [sym dispatch & fn-tail]
  (let [hidden-sym (gensym (str sym))]
    `(do (defn ~hidden-sym ~@fn-tail)
         (.addMethod ~sym ~dispatch ~hidden-sym))))

what would that buy you?

If the namespace is cleared or removed, the multimethods in it are cleared and removed as well.

ah, I see

Well, clearly my implementation isn’t great - but add some find-vars etc. in there

Make the gensym predictable based on the key...

But then it would place nice with tools.namespace.

Alternatively, attach the namespace to the metadata of the function, and let tools.namespace clear away all multimethods with that metadata.

you can kind of do that already if you're feeling hacky

But that would require changes to clojure.core and tools.namespace, so...

user=> (defmulti foo identity)
#'user/foo
user=> (ns foo)
nil
foo=> (defmethod user/foo 1 [_])
#object[clojure.lang.MultiFn 0x15615099 "clojure.lang.MultiFn@15615099"]
foo=> (methods user/foo)
{1 #object[foo$eval20$fn__21 0x41975e01 "foo$eval20$fn__21@41975e01"]}

unmunge the fn class name to get the namespace

and then it's just a matter of remove-method

not pretty but possible

Oh, huh. Yeah, I hadn’t considered unmunging the function name.

really not pretty tho :)

No, but for a dev-time tool that approach definitely has promise.

I just want it for refresh

I might steal that hack, @bronsa 🙂

feel free to do so :)

There is a public demunge method in Compiler btw :)

@alexmiller if there was any chance of having defmethod include the namespace as meta on the method function that hack would be unnecessary.. just planting the idea :P

Well I'd be fine with that :)

But I just work here

now that I think about it for more than 12 seconds, metadata on fn objects has potentially non-insignificant performance overhead

it very much does :c

see the last time we messed with this