I'm working on a project where there are different test paths under different profiles. Is this a common paradigm? Right now cider can't run these tests in the project when you say run all tests in project, and i'm wondering if this would be a common request for people, or if tests under different profiles isn't that common of a setup

Anyone know the best posgres lib for Clojure?

just look for a decent JDBC/SQL library. What postgres specific stuff would you need in a Clojure lib?

@dpsutton: this is something you can do pretty well, but yes, you'll need to write some logic that handles this

Is there a more Clojure-y version of TimeLimiter - https://google.github.io/guava/releases/snapshot/api/docs/com/google/common/util/concurrent/TimeLimiter.html If you’re unfamiliar with the library, the concept is limiting the amount of time functions can run and giving them, essentially, a timeout value. So I’m interested in something along the lines of:

(limit my-function 5000 “default-value”)

seems like a promise/future with a call to deref with a timeout would do it

Ah, thanks! I wasn’t aware of the deref w/ timeout option, that looks pretty perfect.

just be carefull with future, it uses an cachedthreadpool (basically an unbounded thread pool)

you might want to roll you own executor and go a bit lower level for this stuff

let's say you had a java method that took java lambdas as an argument, how would you pass it clojure functions instead?

I think you can just reify the Function interface

(in java.util.function)

you could write a tiny wrapper to do this

@mpenet: got it, looks like it just needs to implement apply, thanks!

I think I recall somebody wrote a lib to ease java8 interop

@josh_tackett: Please have a look at https://github.com/bsvingen/squirrel

anyone use expectationsfor testing? have this dummy test

(let [x 10]
  (expect 20 x))

(let [x 20]
  (expect 20 x))

it actually passed bc it only ran the latter one

wonder why is that?

expect expands to a defn so having it inside a let may well confuse the symbol generation it uses for the name.

Hey, how can I achieve something like (reduce and [true false true])

Like trying something like a haskell fold

and is a macro, so you may need to do something like (reduce #(and %1 %2) [true false true])

Someone else may have a more clever way though

Simpler would be (every? identity [true false true])

hmmm let me try that

I want something like (not (contains? [false false true false] true))

Basically I want to check whether a true is contained in a coll

If it is return false

not-any?

(some identity '(false false true))

(not (some true? coll))

contains? is only for sets and maps.

(contains? [:a :b :c :d] 2) ;=> true) because vectors are associative collections with integer indices as keys.

Hello; I’m seeing some very strange things with transit-clj decoding msgpacked data

The minimal case I can reduce it to is if there’s a string of a certain length (only happing to me when it’s around 140+ bytes long) transit-clj fails to decode msgpacked data if the string is stored like D9 <len>, but works if it’s encoded as DA 00 <len>,

msgpack can read the data though, so it seems to be something specific to transit parsing

does anyone have an idea where I should look? Or do I just need to report a bug to transit-clj?

Maybe @alexmiller can point in the right direction if he's around

Best to file an issue on transit-clj

Then everyone gets notified

If you have a simple repro that would be ideal

alexmiller: I will do that then, working on making the smallest repro I can now

I'm trying to use a transducer to do some work on some meshes. I have something like (transduce (map first) min (:collision-mesh player))

the issue is that min will be asked for a value to begin with, but has no 0 arity

i could provide an init value, but it would look ugly. I would need to select the highest number possible, to ensure that it is immediately replaced in the reduction

another approach would be to let the init be the first (or any) value of the collision-mesh, which feels a little better but still feels kind of wonky

is there a non-wonky way to do this?

In the initial value case you could use Long/MAX_VALUE or double or whatever is appropriate

There is a actually a monoid function in the reducers namespace that can also be used to patch in the no arg case iirc

i think you might be talking about completing

and it patches the 1 arity case

its a shame that min doesn't have a 0 arity that defaults to the maximum value representable

I'm aware of completing but I thought we were talking about 0 arity

we are

is there a different function in reducers that completes for 0 arity?

https://clojuredocs.org/clojure.core.reducers/monoid

that looks like a handy utilty

It's for this need - adding the 0 arity to a reducing function

i'd still have the ugliness of "X/MAX-VALE

where X isn't known

i had considered taking the first value as the init, and letting the collection that the transducer operates on be the "rest"

that feels a little better, but i was just wondering if you guys have ever used transducers this way

and if there is a better way

Just hardcode initial value as 100 and assume it's good enough 🙂

ecks dee nice maymay

so right now i'm considering (transduce (map first) min (first (:collision-mesh player)) (rest (:collision-mesh player)))

i like being able to just tell what code does without thinking about what i'm looking at, and the transducer here is better than the hard-coded init type, but it isn't quite as good as i'd like

Do you need transduce at all?

You could just min of map first?

i suppose i could do a reduction on a map

but thats the definition of transduce

well, i guess transduction is more general

its a reduction on an arbitrary series of stuff

what did you have in mind? what would i gain from doing a (reduce fn init (map first (:collision-mesh player)))

(apply min (map first (:collision-mesh player))) ?

that is much more in line with what i'd like to do thank you very much alex

(reduce min (map first (:collision-mesh player))) ?

that one’s probably more efficient

than apply

i don't really know if thats the case

I do :)

the reason i figured on using transducers in the first place was trying to come up with something more efficient than reduce

apply is going to build an intermediate sequence before applying

the reduce method will involve two sequences

I don’t think so?

the one you feed it and the one generated from the map

oh, well apply will be one more

transduce would be better, but you just need to cover the initial case

i think i'll just put it in a let and do [some-point (first (:collision-mesh player)) other-points (rest (:collision-mesh player))

then feed those into the transduction

that first/rest can be destructured

(let [[f & r] (:collision-mesh player)] (transduce (map first) min f r))

is that more efficient or just more compact?

more compact, should be same efficiency-wise

i'd like to ask a different question, do you guys think this is an abuse of transduce?

if you saw it in code you were collaborating on, would you call it hacky?

nope, makes sense

no, it’s fine

for the reasons you mentioned above - you want to make a single pass through the input, map out the first of each, and then use min to decide which to keep

so it’s more efficient than either the reduce or apply versions above

so, I’d say: not hacky, but good use

the non-0 arity of min was the only sticking point

its regrettable that i have to work around the non-0 arity of min

but that’s not something that can be done (as min is polymorphic across all number types)

If you wanted to you could wrap min' in a function that answers for the 0-arity case, but that's the hacky version compared to what's been suggested

so it isn't possible to establish a notion of "replace me with anything-ness"

the only thing about the transducer version i don't like

is the fact that we arbitrarily chose the first of the sequence

we could have chosen any point whatsoever

imagine we chose the last as the init

then the first seems equally good :)

in fact, based on the way min works

we don't necessarily have to choose "the rest" as the collection

we could have just fed the entire collection

but that is specific to min perhaps, and the fact we chose "first" or "last" is an arbitrary choice

the specificity is something that irks me

that is the apply version

i think the point of transducers, besides efficiency, is decomplecting process from data

i feel its a good policy to make your algorithms no more specific than they have to be

how many of these things things are there?

they are all over

i have just 8 in this one function

and have you actually considered reducers?

if i use reducers i will double my memory footprint (at least)

b/c they can fold in parallel, giving you additional gains if you have a lot of values

ah you mean those reducers

where “lot” is probably at least 1k

yeah i'm considering that, but given the fact that at the "highest level" most of my transducers are composed entirely of maps

i could just use a pipeline

It seems like you may need to make a decision between the most performant solution and the most philosophically correct

in that case (first ...) (last ...)

or the destructered version would be best

I like the transduce personally :)

i'll just make sure to communicate the essence in the name

some-player-point, and other-player-points for instance

Silly question, but isn't there a name/concept in clojure for an item that's not a collection? Similar to Lisp atoms?

values, or symbols, depending on the context?

Well, independent of context, I think. The set includes numbers, strings(?), symbols, keywords, characters, (and what else?)

Oh, sorry. I now see what you meant.

atom?

It's fair to just use the lisp terminology, but of course most people will think clojure atoms are something else

ha right

Symbols can't be it. For example (symbol? :keyword) is false.

values is more like it? But collections are not values?

mmm I don't think there is something analogous to atom? you can maybe fabricate your own with symbol? keyword?...

Things that are not collections are typically called either symbols (+, defn, and) or values/literals (7, "foo", :keyword). I'm not sure there's an official universal term for everything, and if there is it's not widely used.

But there would be no reason for a predicate to exist to test it, because any possible parameter would return true.

@em also in Clojure these checks are based a lot on the underlying Java Interface (not saying this is good, but it is what it is) -> http://clojuredocs.org/clojure.core/coll_q

thanks. I'll roll my own atom?. (I'll call it something else to avoid confusion)

hrmph. apparently (reduce min (map first coll)) isn't the same as (transduce (map first) min (first coll) (rest coll))

for instance, if i use [[0 0] [0 1] [1 0] [1 1]] as the coll, i'll get a class cast exception

it should be (transduce (map first) min (first (first coll)) (rest coll))

the extra first is required because the "init" isn't passed through the transducer (the map to first in this case)

btw, (ffirst x) is (first (first x))

thanks. i feel like i'll have to approach the problem differently though

(map first) is a simple transducer

i have more complex ones, and I don't want to have to push a single value through it all the time

@bcbradley: that makes sense, the transducer will only be called on a collection, and init isn't always going to be the collection

actually, never mind, I think I see your point

here is a way to do what I want (apply min (map fun coll))

another way is (reduce min (map fun coll))

in order to improve performance, you can make one pass instead of two

(transduce fun min coll) comes to mind

however, min has no 0 arity, so an initial value must be provided

(transduce fun min init coll)

it turns out that init isn't pushed through fun

so imagine fun is (map first)

and coll is a mesh-- [[0 0] [0 1] [1 0] [1 1]]

you'd have to say (transduce (map first) min 0 [[0 0] [0 1] [1 0] [1 1]])

but how did you get that "zero"

you got it by choosing "any" point in the mesh

could be the first one for instance,

and then using (map first) on it

in other words, you really want the init to be pushed through the fun, but transduce isn't doing that.

i'd like to say (transduce (map first) min (first coll) (rest coll)) for instance

so how do you express performing a transducer on a single value?

another transduction.

transducers are functions of reducing functions, not individual values. this has less to do with transduce and more to do that you can't give the reducing function (min) a sensible init value

have you tried

(apply min-key first coll)

(apply min (map fun coll)) is what i've tried with apply

min-key and max-key are in core

if you really want to use transduce with this case:

(transduce (map first) min Long/MIN_VALUE coll)

that way there is something for the first element of coll to be compared to, and it will always win

that has been suggested

i don't like it because i may not necessarily know the type of the number in the mesh

It could be larger than 64 bits?

i suppose i could create my own transduce that wraps the standard transduce, converts the init into a single element collection, performs the transduction on it to generate an single element sequence that contains the init, and use first to pass it to the standard transduce

seems a bit roundabout

clone the first element of the collection as init?

(btw, why might you not necessarily know the type of the number?)

because i use fractions liberally in the code

i mix with floats, integers

gotcha. polymorphic math costs will dominate anything you're worried about wrt transduce

personally i wouldn't worry about it

are you looking for perf as #1 priority?

i'm using this specific pattern 8 times alone in just this one function. I'd say probably about 12% of all the lines of code i'm writing is this exact pattern

its important that it doesn't consume more memory than required

then I'd enable the boxed math warnings and loop/recur

chains of filters, maps, and reductions are out

actually nm, if you're doing polymorphic math you'll always box

Is it not feasible to standardize on one type of number?

E_EXHAUSTED_LUCK

i know it seems silly to have proper fractions in code that processes meshes

in this case its not graphics code, its collisions code

most of the collisions code i've seen written doesn't aim for strict reproducability

the tiny errors that occur in floating point ops generally doesn't matter in most applications because they don't accumulate

in video games specifically, they do

there is a concept in math called chaos, it means that tiny changes in the initial conditions result in enormous changes in final conditions

and that it isn't possible to ascertain the difference in the final based on the difference in the initial

i'm experimenting with exact math to make it possible to have an exact simulation

which means that it would be reproducible on any computer

the "gotcha" of floating point ops isn't that they error, its that the error between two computers may be different

so the simulation would diverge

authoritative servers aren't just for competitive integrity

they serve as an authority on the state of the game, so that diverging simulations know what the "true" state ought to be

if you've ever played a game and had a brief glitch where your avatar sort of tunneled through a wall , moved all wonky and then finally started acting normal

thats what happens when the client simulation and server simulation get vastly out of sync

and in that case, the server's simulation dominates

i want to avoid that so that I can have a wide peer to peer network infrastructure with no central server

i figured i could avoid the floating point chaos at first by asking for a majority consensus on the state

but that won't work because they diverge uniquely

so i am exerimenting with exact math

So if you use Long/MAX_VALUE, you can freely mix rationals and floats and the result will be correct.

i hate that i have to do it that way, but you are probably right