I'm trying to use environ, but using it at compile time(our build machine have all env vars), if I wrote (def ^:const x (env :x)) it works, x will be compile to a constant. but (def ^:const x (Integer/parseInt (env :x)))will not work.

@doglooksgood are you doing AOT compiling? Or is this ClojureScript?

@bfabry : why should delete be more expensive than assoc? can't we just 'mark it with a tombstone' ?

@danielcompton I'm doing AOT compile in Clojure.

@qqq A look at the source for assoc on vectors should convince you that delete would be more expensive -- assoc doesn't change any of the indices for the vector so only the segment containing the "changed" item needs to be updated. delete would cascade a change of indices across the segments of the vector. As I understand how vectors are implemented in Clojure -- happy to be corrected.

@doglooksgood Aside from all the myriad problems associated with AOT and why you should avoid it... ahem ...could you be a bit more specific about "will not work"? What exactly is the error/behavior you get?

@seancorfield : I failed to consider indexing. Does vector maintain the invariant "the only non-full 32-item block is the last block" ?

if I use (def ^:const x (env :x)) and :x in env is "hello", it will be compile to (def ^:const x "hello") i think, because our build machine have environment variables, and production machine dont, this is the way we want. if I use (def ^:const x (Integer/parseInt (env :x))), x will not be compiled to a literal constant. so at runtime, this (Integer/parseInt (env :x)) will run again.

no

^:const has a very small very specific use, that is not it, it only happens to "work" for (env :x) but that could change at anytime

what's the explain for this, what's the difference between these two?

(I'm a bit surprised (def ^:const x (env :x)) works, to be honest)

^:const causes the value of def to be inlined at the invocation site, it doesn't care about how the value is produced

(env :x) is fine if that returns an value that can be used as a constant

const doesn't mean that the literal expression passed to def will be inlined, (def ^:const x (do (println "foo") 2)) the println will only ever be runned once

and x will be 2?

yes

and x will be 2?

I know how to write macros. Is it easy to write reader macros?

I'm trying to implement a special macro (:: a t-sig) where it always evals to a, regarxless of where the :: is located at

so I can write something like

(defn [ (::a int) (:: b int)] .... )

and it should become

(defn [a b] ...)

@qqq Not sure I'm understanding you but Clojure does not support "reader macros"...

@seancorfield : I also thoguht it might be impossible, but then I read: https://stackoverflow.com/questions/20677055/define-my-own-reader-macro-in-clojure

Tagged literals begin with # -- like #inst and #uuid -- but that's not "reader macros".

Tagged literals have #, a namespace-qualified symbol, and a regular Clojure expression. The regular Clojure expression is read, then passed to the function associated with that symbol.

@qqq We use tagged literals in our configuration library at work so we can define values in "special" ways.

@seancorfield : (I know nothing about tagged literals / reader macros) -- so what you're saying is that (1) what tagged literals get is after macro expansion and (2) it's basically a function call with a single argument ?

How come tools.deps.alpha is using a map (which is merged with the default deps)? Doesn't that mean I can't specify the order of the deps? Which, IMO is crucial on the JVM.

Just tried out Compojure API (2.0). https://github.com/metosin/compojure-api

I had no idea it was this easy to set up an API. Is this normal?

Paranormal

@rauh. Do you mean the order of maven coordinates in project.clj? I wasn't aware that ordered mattered there

it matters anywhere (boot, lein etc) I think, it's a maven property

@pesterhazy No I mean the new tools.deps.alpha project that just got released. In maven all of the order matters, which makes projects predictable. Though there is a bug report for leiningen too: https://github.com/technomancy/leiningen/issues/2283

but yes, I think @rauh is right about that

It's going to seeminly work for projects that have <=8 dependencies and then all in a sudden the depedencies will be semi random and even change if you just change the version number.

So IMO, deps shouldn't be specified by a map.

it's prolly worth raising the issue with @alexmiller & co

Yeah I'll wait until they wake up in the states.

@rauh, I was aware of tools.deps but didn't know order influenced the maven algorithm

Maybe just order apathetically?

Whoops alphabetically

A wonderful typo though

sorting wouldn't help

> Dependency mediation - this determines what version of a dependency will be used when multiple versions of an artifact are encountered. Currently, Maven 2.0 only supports using the "nearest definition" which means that it will use the version of the closest dependency to your project in the tree of dependencies. You can always guarantee a version by declaring it explicitly in your project's POM. Note that if two dependency versions are at the same depth in the dependency tree, until Maven 2.0.8 it was not defined which one would win, but since Maven 2.0.9 it's the order in the declaration that counts: the first declaration wins. https://maven.apache.org/guides/introduction/introduction-to-dependency-mechanism.html

At least it would make it predictable

but it prevents you to have control over it

But yeah deviating from maven here doesn't seem like a good idea

I imagine deps could support a vector of "tuples" and map transparently, use the former if you care, the other if not (but since maps make ordering unpredictable I am not sure that even make sense to support them at all)

Yeah I wish the new t.d.alpha was leaning heavily on maven and would just provide a hiccup wrapper around their XML config. That'd be neat.

to be fair I've never been in a situation where I had to re-order maven coordinates in clojure to fix dependencies. Have you?

Can't you accomplish the same thing through exclusions?

imho it doesn't really matter, if the aim is to have a simple tool that kind of maps 1-1 with maven without too much indirection, that ordering issue could be considered a flaw

but no, never had to mess with ordering myself I think

It's seldom and only matters if you have the same files in two jars. Which you shouldn't have. But then, it's not clear how often this problems has been the issue with user reported problems who never figured it out what the issue was.

@pesterhazy we have had to do this many times when using xml

deps conflicts is quite common, ex between guava versions in transitive deps it happens to me all the time

IMO predictability is crucial, so deps should neither be put into a set (as leiningen does currently) nor into a map.

hi everyone,…i want ask, sqlkorma how set date and datetime mysql format timezone GMT ?

@rauh, wait why is it a set? do you mean internally?

@rauh my (limited) understanding is that you're meant to resolve all dependency conflicts completely, so it would actually be ok to store them as a map

@danielcompton how would that work given that there are transitive dependencies, java dependnecies, etc.?

do you mean a complete spec in the style of yarn.lock?

I thought the idea was that you completely resolved all of the transitive dependencies explicitly

more like, being explicit about exclusions to solve these kind of issues, but still

(which is the best way to fix these in general anyway)

@danielcompton ahh? didn't understand it that way.

yeah I think you're closer @mpenet

but I thought idea is that there is no conflicts

sort of like lein pedantic

I see, that would make sense, and actually an improvement over raw maven definitions (arguably, since way more verbose)

@danielcompton My issue isn't only about transitive deps, but about the non-predictable ORDER of the jar's on the classpath. (See my leiningen ticket)

hmm, yeah

Does the order matter for you because you have several JARs which provide the same namespaces/classes?

alex mentioned the classpath order in the talk here in berlin

iirc, he said that the order is useful for some things (like overriding packages a la carte) but is not a great fit for other things, like making sure the right dependencies are picked up

so it sounded to me like tools.deps is intended to take classpath order out of the equation somehow

hey, anyone here who can help out with a boot-clj related question?

well... probably someone, just ask, but maybe in the #boot channel...

what could be an easy way to some a random choice given some probabilities I just need a function that given something like this: {:a 1/2 :b 1/4 :c 1/4}

returns me a random element from [:a ':b :c] respecting the probabilities for each of them

put the cnhoices in arbitrary order

sample randomly from [0, 1]

index into the cumulative probability

yeah I thought about that, so I guess there is no easier way?

wouldnt repeating each choice depending on the probability and then using rand-nth do the trick?

that would work in a simple case like this one

but in the general case I would need to find the LCD of all the fractions

and generate a massive vector with all the repeated stuff

right, i see

suppose an item has probability .3957 , do you repeat it 3957 times? 🙂

fair point

Something like this gives me the ranges I can then check

user> (def probs {0 1/2 1 1/4 2 1/4})
#'user/probs
user> (into {} (map-indexed
                (fn [idx [el prob]]
                  [el
                   [(apply + (take idx (vals probs)))
                   (apply + (take (inc idx) (vals probs)))]])
                probs))
{0 [0 1/2], 1 [1/2 3/4], 2 [3/4 1N]}

quite sure it can be improved a lot though

https://stackoverflow.com/questions/14464011/idiomatic-clojure-for-picking-between-random-weighted-choices

nice thanks

I am pondering about how I would implement a simple compiler and/or interpreter in Clojure. In particular, I am wondering what would be the best way to model the abstract syntax tree. The first (and most obvious) option would need to represent AST nodes as maps, tagged with their type/label (e.g. :type key). I would then define the various functions that operate on the AST as multimethods using the node’s types as dispatch keys. Another option would be to represent each type of node as a record, and to then have them all implement a protocol (e.g. IExpr) which would contain have implementations for all the functions operation on AST nodes. This comes wth the added benefits that AST nodes could implement existing protocols too. Which of these options would be most idiomatic in Clojure? Or do you have another one to suggest?

Take a look at Instaparse

clj-antlr isn't bad either

Which parts in particular? I am not looking to write a parser, just to understand what the best way to model an AST in clojure would be

multimethods are nice, I think

I made a notice about instaparse because if you use it — you don't have to think about AST representation yourself

@hmaurer take a look at tools.analyzer

https://github.com/clojure/tools.analyzer https://github.com/clojure/tools.analyzer.jvm

Another, completely unrelated question… Does anyone have advice/examples on how to structure application business logic? This is a fairly general question (not specific to Clojure), but there might be Clojure-specific idioms coming into play. Roughly speaking I am wondering if I should take a CQRS-y road and try to have a clear distinction between functions that modify the app’s state and functions that read that state. The application I am currently working on is a fairly simple, CRUD-y Clojure+Datomic web app, but I would like to use it as a learning ground for good practices

A link to an open-source application that you consider well-structured would be helpful, or advice / link to blog posts / resources

@hmaurer have you seen the docs for re-frame? He talks extensively about CQRS-y architecture (even though it’s cljs not clj) https://github.com/Day8/re-frame

There’s a pretty detailed readme doc right up front, but don’t be fooled, the real meat is in the docs/ folder

@manutter51 ah, let me check it now, thanks! By the way, something I should have specified: I am looking for a way to structure my code while keeping it fairly straightforward. E.g. no crazy async at the moment; ideally I would like to keep everything synchronous

When I do some calculation with pmap and want to use another pmap on the result. How I can I wait until the first pmap is done?

Does anyone know of a clean way to assoc in a field only if the field isn't null; otherwise return the map unmodified?

@ghopper merge

👍

Thanks, @dpsutton

merge will overwrite existing keys in the original map... this may (or may not) be the expected behaviour.

change the order

I'm aware. That is what I want. It's just an optional field that should be tacked on only if it's not nil. (An authorization http header)

(merge {:field "value if not present"} my-map)

yeah exactly

right 🙂

I don't suppose there's a merge that would do (merge {:a 1} {:b nil}) and return {:a 1}?

that sounds orthogonal to merge. you want to prune your map after the merge?

if it's just top level keys you could (into {} (map (fn [[k v]] (when v [k v])) my-map) would probably work

there's probably an even better sort with filter

(into {} (filter (fn [[k v]] v) my-map)

and if you have possibility of false change the body of the fn to (some v)

All great ideas. I've gone with (merge xs (when a {:a a})).

that will clobber your original value i thought

you need your original map second

@ghopper In that case you could do: (cond-> xs a (assoc :a a))

Sorry, I'm mixing names. xs doesn't have a :a yet. There's no possibility of clobbering.

👍

@rauh I quite like that one. :thumbsup:

Is there a better way to do (symbol *ns* 'symbol)?

Huh, nevermind. Apparently I can do syntax quoting for the dispatch value of a multimethod. Not sure why I thought I couldn't earlier.

@ghopper Not sure if it would help, but https://github.com/nathanmarz/specter has some really 'navigators' to do all kinds of transformations on data structures. (Maybe not worth the effort for easy navigation, although I think it may be if you have at least 2 steps to navigate, including predicates 🙂 )

@kurt-o-sys Huh, that's cool. Definitely overkill for this though. 🙂

@ghopper for associng a value if not null, with specter it's (setval [:a some?] my-value amap)

@kurt-o-sys are those basically lenses?

for associng only if field exists, it would be (setval (must :a) my-val amap)

will be way more performant than merge as well

I didn't use specter much yet - only found about it about a month ago - but it's definitely cool. Also for small cases.

Style question: say I have three items (def a ["a"]) (def b [["b" "c"] ["d" "e"]]) (def c "f"), and I want to put them into a list like so:

`[~a ~@b ~c]

(i.e. the desired result is [["a"] ["b" "c"] ["d" "e"] "f"])

so you just want to append them?

the middle one is unpacked

one option is (concat [a] b [c]) - I admit that looks weird

@noisesmith actually, that's not nearly as bad as what I was doing

Does suck a bit that you have to pack the elements you don't want unpacked just to...well...yeah

@jballanc depending on your actual problem, you could use something like (def prepacked [x] (if (and (coll? x) (vector? (first x)))) x [x]) (mapcat prepacked [a b c])

hah...yeah, that probably won't work because some of these are multiple levels nested, and only some of the levels need to be unpacked

my personal preference is to "say what you mean" with syntax-quote, but understandably the team is worried about maintainability...

Hey there. How to update a key in a map to hold a empty vector?

=> (def Person {:person-id "person-1" :category "customer" :purchase ["p1"] :dates ["d1"]})
#'user/Person
=> (update :purchase Person [])
IllegalArgumentException Key must be integer  clojure.lang.APersistentVector.invoke (APersistentVector.java:292)

look at the docs for update

(you seem to have picked random arguments in a random order)

(and you want assoc not update, but go ahead and read the docs for both assoc and update)

by docs I meant the docstring you access using doc in the repl, but you may also want to start here https://clojure.org/reference/data_structures#Maps

Hi! I am getting a cryptic error which I do not understand. Here’s the piece of code: https://gist.github.com/hmaurer/8e786bfd507798393c8be45ffb3a1b46. The error is in the comments. Could someone take a look please?

The error is Can't let qualified name, but I don’t get why the gensym’ed variable gets qualified by the syntax quote

I would double check to make sure the code that is being run actually matches the code you are reading (restart your jvms) and then I would suspect this is actually coming from some other macro

I restarted lein’s repl. Is that enough?

yes

I would put a println at the start of the macro to print out ctx-binding because I suspect it is already being passed in a fully qualified symbol

your docstring is also in the wrong place

ah, thanks for pointing that out

it is a super common thing to do

ctx-binding seems fine. The println prints ctx

which matches how I am using the macro, e.g.

(deftest test-authentication
  (with-scratch-ctx ctx
      ...

Still throwing :cause "Can't let qualified name: wef-backend.test-util/conn# " though

is d/db a macro?

no

but I get the same error even if I deleted the whole ~ctx-binding line

I never got that error with uri# or datomic# by the way, it only occured when I added conn#

It’s probably something silly

do you have a var named conn# in that namespace?

@hmaurer What does macroexpand give you?

@hiredman no, I checked. Let me check again just to make sure..

java.lang.RuntimeException: No such var: wef-backend.test-util/conn#

you could fix it immediately by replacing most of the macro with a function that returns the context, and have the macro expand in to invoking that and then deleting it

but that wouldn't tell you what is wrong

@rauh

(let*
 [uri__16604__auto__
  (clojure.core/str "datomic:mem://hello-test-" (java.util.UUID/randomUUID))
  datomic__16605__auto__
  (integrant.core/init-key :wef-backend/datomic {:uri uri__16604__auto__})
  wef-backend.test-util/conn#
  (:conn datomic__16605__auto__)
  ctx
  {:auth nil :conn conn__16606__auto__ :db (datomic.api/db conn__16606__auto__)}]
 (try (+ 1 2) (finally (datomic.api/delete-database uri__16604__auto__))))

when running

(macroexpand '(wef-backend.test-util/with-scratch-ctx ctx (+ 1 2)))

What happens when you rename it to some-other-conn#?

tried that, didn’t work. Trying again now just to make sure

is that a real pound sign, or some unicode nonsense?

Well the second usage of conn# expands properly. (in the map)

yeah, which makes me suspect the first has some weird shenanigans

Wait, I think I found the issue

some-other-conn# (:conn datomic#)]

the “space” between # and ( has charcode 160

instead of 32

oh lol. 🙂

hah!

unicode malarky

did you copy past some of this from a github page?

nope, I wrote it myself

no idea what happened there

well, I copied an earlier version of that code, which worked

then added that line myself

from where?

it might have picked up a character from somewhere

github

oh

weird

I just checked and none of the characters from the snippet I copy/pasted had character 160

oh well

sorry for the trouble

there is actually an open jira issue that has had some recent activity about changing how clojure handles unicode whitespaces

char code 160 would be  

the fun thing about this is, thats 160 is not even considered whitespace by java

oh, I guess the issue it got declined

Probably what happened is that OS X and/or Atom (the editor I am using) has some shortcut or some way to enter this type of whitespace, and it fat-fingered on the shortcut

employee-resizer.core> (char 160)
\ 
employee-resizer.core> \ 
\space

https://dev.clojure.org/jira/browse/CLJ-2207

it had been a while since I debugged a unicode issue; almost forgot how fun it is

apparently atom will enter a non-breaking space with alt-space

probably pretty easy to inadvertently hit

yep, I most definitely inadvertently hit that

If Clojure’s reader threw an error saying “unknown character at position X” it would have been easy/easier to debug, but it considered it as part of the symbol

that actually may be just be a bug in the reader

Now I know how to confuse the hell out of people though

wef-backend.core=> (def     42)
#'wef-backend.core/
wef-backend.core=>
42
wef-backend.core=>

this works

this too:

wef-backend.core=> (def hello world this is a long variable 99)
#'wef-backend.core/hello world this is a long variable
wef-backend.core=>

eh

What is a fast way to do this without the repeated nested pointer?

(-> state
    (assoc-in [:a :b :b/field0] "")
    (assoc-in [:a :b :b/field1] "")
    (assoc-in [:a :b :b/field2] ""))

;; maybe?
(assoc-in [:a :b] {:b/field1 "" :b/field1 "" :b/field2 ""})

fast as in performance?

Fast as in, using standard library. I don't want to add any external libs or anything.

Performance isn't a concern.

Sorry, I shouldn't be using that work. 😛

(merge state {:b/field1 "" :b/field1 "" :b/field2 ""})

err

sorry, (update-in state [:a b] merge {:b/field1 "" :b/field1 "" :b/field2 ""})

forgot a few words

(let [nav [:a :b]]
  (-> {:a {:b {}}}
      (assoc-in (conj nav :b/field0) "")
      (assoc-in (conj nav :b/field1) "")))
{:a {:b {:b/field0 "", :b/field1 ""}}}

the navigator is just a vector, so create it with the shared navigation parts

if i'm understanding correctly

@dpsutton That's what I thought I was going to need to do. @bfabry I'm not sure why I didn't think of update-in... That should work nicely.

Not the question is update-in with assoc or merge. Is there any performance difference?

I suspect they'd be identical

I figured they would be. Anyways, thanks all!

merge is implemented in terms of conj so yes I'd say identical https://github.com/clojure/clojure/blob/clojure-1.9.0-alpha14/src/clj/clojure/core.clj#L3022

When I use lein, where will the clojure stdlib be located on my file system?

.clj files that is

(let [nav [:a :b]]
  (reduce (fn [m [k v]] (assoc-in m (conj nav k) v))
          {:a {:b {}}}
          [[:b/field0 ""]
           [:b/field1 ""]
           [:b/field2 ""]]))
{:a {:b {:b/field0 "", :b/field1 "", :b/field2 ""}}}

@hmaurer You mean the jar files?

ah right, so the std lib will be within jar files. Nevermind then

jar files are just zip files and you can unzip them and the .clj files will be inside, if that matters to you

@hmaurer lein runs 2 jvms, lein's jvm and your project's jvm, the clojure runtime that lein uses is part of the lein jar and lives in ~/.lein, the clojure runtime for your project is Just Another Jar(tm)

lots of interesting jar files in ~/.m2/repository too

what’s a good way to xor two sets?

(set/union (set/difference s1 s2) (set/difference s2 s1))

a silly way to do it

=> (into #{} (comp cat (take 2)) (clojure.data/diff #{1 2 3} #{2 3 4}))
#{1 4}

is it possible to use clojure.java.shell to invoke a shell command like tr < infile.txt -d '\000' > outfile.txt ?

https://github.com/clojure/clojure/blob/master/src/clj/clojure/java/shell.clj

https://twitter.com/clojure_conj/status/889899202767147008

^ How cool is that?

yeah got the email earlier. super super cool

@aaelony in order to use things like < you need to invoke /bin/sh

clojure.java.shell/sh, despite the name, is a raw system command that doesn’t use sh

of course that’s not portable, and then you’ve created code that only works on a *nix system or reasonable facsimile

yeah, @noisesmith there is a thorny translation that tr handles well at the command line that I'd like to shell out within a clojure program ( that does a whole laundry list of things). Haven't yet found the right syntax for calling tr via clojure.java.shell/sh though

@aaelony you send a normal shell command to sh (clojure.java.shell/sh "/bin/sh" "-c" "tr < infile.txt -d '\000' > outfile.txt")

wow, thank-you. That works. I tried a lot of other combinations but without the "/bin/sh" "-c"

right, sh -c says “find and run the sh executable, tell it to run this”

that class of example should probably be added to the other examples (https://github.com/clojure/clojure/blob/master/src/clj/clojure/java/shell.clj#L130-L142)

thanks again

np, glad I could help, I wonder if it would be worth submitting a patch to JIRA for adding another println to that comment block

agreed - I think that would help others. Not sure how easy that is to do...

@aaelony there is an example of using sh -c as an arg to sh on the clojuredocs page https://clojuredocs.org/clojure.java.shell/sh

true enough, perhaps a few words there describing why it is useful/needed would work too

@noisesmith thanks for your help yesterday, you were correct !!!!