non/spire

This room has moved: https://gitter.im/typelevel/spire

Denis Rosset

@denisrosset

Moreover, Spire depends on cats-kernel, which evolves much slower.

Grigory

@pomadchin

Actually we exeprience a problem with Order type class

it looks like cats are not binary compatible across versions

java.lang.NoSuchMethodError: spire.std.package$any$.OrderProduct2(Lcats/kernel/Order;Lcats/kernel/Order;)Lcats/kernel/Order;
                      at geotrellis.raster.render.BreakMap.geotrellis$raster$render$BreakMap$$vmTree$lzycompute(BreakMap.scala:83)

It actually happens there: https://github.com/locationtech/geotrellis/blob/f8af6e6d72749b8fb175a0c44cbc7d027a940686/raster/src/main/scala/geotrellis/raster/render/BreakMap.scala#L83

Btw FYI, we had a huge problems with updating FS2 up to 1.0 and still having a dependency of cats core lower than 1.4 :/ that's probably also caused by changes in cats-effects and a lil bit a differnet story, but it looks like they don't follow semver really well.

Grigory

@pomadchin

Hm.... but i see you point... mb we have smth weird in our deps internally

I'll doublecheck that

or with imports and it weirdly works in some cases. anyway, thanks for your response. Mb we bumped up a version and smth went wrong. Wonderin why tests are passing.

Denis Rosset

@denisrosset

The missing method is actually in Spire; check that you don't have multiple versions present

Note that Spire does not have binary compatibilities as we are not yet at the 1.0 release

Grigory

@pomadchin

no worries there and sorry for a lil bit messy messages; i haven't investigated the problem well enough, agree with you, sounds like you're right

Denis Rosset

@denisrosset

good luck!

Grigory

@pomadchin

just evening... thank you!

Grigory

@pomadchin

all right, it is a spark classpath; sry for bringing some messy messages there @denisrosset

it looks like spark 2.x has spire 0.13 in its classpath by default

Denis Rosset

@denisrosset

No prob, thanks for clarifying!

Luciano

@lJoublanc

Can somebody confirm if the ULong AnyVal implementation has full 64-bit precision?

Denis Rosset

@denisrosset

yes, that's the point! but it boxes in an array, beware

Jakub Kozłowski

@kubukoz

hi, is there a way to get all values inside an Interval?

actually, that makes no sense. Forget it ;P

Denis Rosset

@denisrosset

depends on A !

Jakub Kozłowski

@kubukoz

I was thinking LocalDateTime

turned out I needed a Set instead of an Interval, though (for subtraction and intersection)

Denis Rosset

@denisrosset

@kubukoz There are sets/seqs of intervals in spire-extras, although I've never used them

Jakub Kozłowski

@kubukoz

I'll take a look

Luka Jacobowitz

@LukaJCB

What’s the best way to get a serializable reprsentation of polynomials? I was thinking a list of terms, but that seems to be giving me some weird index out of boundexceptions

Luciano

@lJoublanc

Are you just using java serialization? You'll need both the coefficients and exponents (or alternatively 0 terms for exponents that you skip).

Luka Jacobowitz

@LukaJCB

I'm trying to write a circe codec

I turn it into a list of terms and serialize that

And then I decode in the same way

Luciano

@lJoublanc

I was actually thinking about this the other day ... you know scala 2.13 has support for opaque types, that I believe allows creating new types that are unboxed arrays essentially.

If there was an implementation using this as the data type (currently I don't think spire's polynomial is implemente like this), then that would be really efficient to serialize. It would just be an array with each element being the coefficient (with zeros for those powers that are missing).

So essentially no need to serialize at all, as the scala representation could match the binary representation.
It would also allow really efficient vectorized computation.

Luka Jacobowitz

@LukaJCB

Any idea why using PolySparse.apply would give me an index out of bounds exception? 🤔

Luka Jacobowitz

@LukaJCB

I’ve minimized the problem

apparently the Eq instance throws exceptions when comparing

Luka Jacobowitz

@LukaJCB

non/spire#756

Luka Jacobowitz

@LukaJCB

On another note, is there really no way to get an Algebraic back from an Expression?

Denis Rosset

@denisrosset

No idea on that, I don't know that part of Spire so well. Ask @tixxit maybe?

Luka Jacobowitz

@LukaJCB

I actually got it fixed, thanks :)

Ichoran

@Ichoran

Slightly off topic, but this is the right kind of crowd for this type of stuff: if any of you have ideas about how to get less fraught behavior out of BigDecimal I'd like to hear about it on scala/bug#11152

Mark Tomko

@mtomko

Perhaps we're doing it wrong, but we just ran into an issue building an assembly jar for a project that has spire in its transitive dependencies - it seems that the subprojects of spire each get built with a buildinfo.BuildInfo$ object, from the sbt-buildinfo plugin. On assembly, these are causing a conflict. I can configure the assembly to discard them but that's a little bit scary. Would it be okay if I made a PR that tried to do one of the following:

put a buildinfo class in a module all to itself that the other modules depend on

put the buildinfo objects for each subproject in their own packages

sorry, my enumeration failed there

Jakub Kozłowski

@kubukoz

looks like it wasn't a good idea to put buildinfo into the suggested package

I'd just use a different package

Mark Tomko

@mtomko

that makes the most sense to me

I can take a quick look at it later today and make a PR for it.

Jakub Kozłowski

@kubukoz

it'd be lovely if all BuildInfo stuff was inlined at compile time though...

maybe there's a way to do that in sbt-buildinfo and fix everything in one go

Where communities thrive

People

Repo info

Activity