Julia v1.5.0 has been released

anon37204545 · August 4, 2020, 7:02pm

Debian / Ubuntu is still on 1.4.1, so I guess not soon.

vbocan · August 4, 2020, 8:32pm

When I download the Windows 64-bit installer on Windows 10, it gets flagged as dangerous.

Version 1.4.2 was fine.

kristoffer.carlsson · August 4, 2020, 8:51pm

It just says it isn’t commonly downloaded. Which makes sense since it is a new release.

lewis · August 4, 2020, 9:43pm

Very nice.

The usual house-cleaning with a new version, but very little house-cleaning related to Julia 1.5 itself. It keeps getting easier.

StefanKarpinski · August 6, 2020, 2:52pm

4 posts were split to a new topic: Stable hashing across Julia versions

zekeriya.sari · August 6, 2020, 10:29am

The repl terminates immediately after being launched in julia-v1.5.0. Here is the platform info

Platform Info:
  OS: Linux (x86_64-pc-linux-gnu)
  CPU: Intel(R) Core(TM) i7-6700 CPU @ 3.40GHz
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-8.0.1 (ORCJIT, skylake)

There is nothing wrong in the previous julia versions.

simeonschaub · August 6, 2020, 10:38am

Probably Does using Revise cause julia 1.5.0-rc1 to crash?. (You need to start Julia as julia --startup=no and then update Revise.)

zekeriya.sari · August 6, 2020, 10:50am

That was exactly the case. The solution you suggested worked. Thank you @simeonschaub

FHell · August 6, 2020, 12:35pm

Excellent work, the new behaviour of immutable structs is a major enabler for us and we are seeing some performance wins already. I did, however, also find one performance regression:

Julia 1.4: 1.583 μs
Julia 1.5: 2.607 μs

in

using LightGraphs
using BenchmarkTools
using Random
Random.seed!(3)

### Defining the graph

N = 100 # nodes
g = barabasi_albert(N, 5)
B = incidence_matrix(g, oriented=true)
B_t = transpose(B)
x0 = rand(N)

@btime B_t * x0

instead using

@btime B * sin.(B_t * x0)

has:
Julia 1.4: 5.638 μs (6 allocations: 8.69 KiB)
Julia 1.5: 8.850 μs (6 allocations: 8.69 KiB)

Edit: Filled an issue: https://github.com/JuliaLang/julia/issues/36941

dpsanders · August 6, 2020, 4:33pm

Do you still find that if you interpolate correctly with $?

FHell · August 6, 2020, 4:50pm

Yes:

hellmann@clara:~/Nextcloud/Playground/NDBench$ julia bench.jl
 Activating environment at `~/Nextcloud/Playground/NDBench/Project.toml`
  3.009 μs (1 allocation: 3.88 KiB)
  8.658 μs (4 allocations: 9.50 KiB)
hellmann@clara:~/Nextcloud/Playground/NDBench$ ~/julia-14/bin/julia bench.jl
 Activating environment at `~/Nextcloud/Playground/NDBench/Project.toml`
  1.753 μs (1 allocation: 3.88 KiB)
  6.187 μs (4 allocations: 9.50 KiB)

dpsanders · August 6, 2020, 5:07pm

I confirm that I see a performance regression for this code on Mac OS.

Palli · August 13, 2020, 2:34pm

is on “Latest/stable 1.0.4”, which isn’t even true for latest LTS. I’m not sure where/how to contact anyone to update this, and the unofficial snap is not limited to 32-bit like the official one. And since Jeff recommended snap at some point it should happen. [EDIT: I’m not saying Jeff recommended 32-bit Julia snap, I don’t recall, there may have been no Julia snap at that point, he was recommending snap for distributing Julia.]

The unofficial better one is at: GitHub - mrcinv/julia-snap: Snapcraft configuration for julia

jling · August 13, 2020, 2:41pm

wait, you telling us Ubuntu has 32-bit Julia in the official snap store?

Palli · August 13, 2020, 2:56pm

Yes, exactly (I just checked again be removing and reinstalling it and see “OS: Linux (i686-pc-linux-gnu)”, unlike for the unofficial snap), and while that works for many (and only option if you don’t have 64-bit), I had to debug that misguided choice when calling to 64-bit libraries (was using PyCall). I thought it might have been intentional to cover 32-bit and 64-bit platforms with one binary, but see now you’re not limited to just one arch.

Here’s the unofficial snap, up-to-date for Julia 1.5 (supporting 32- and 64-bit, also ARM):
https://github.com/mrcinv/julia-snap

I see the file the official one needs (just don’t know where to make a PR for it for the official snap): https://github.com/mrcinv/julia-snap/blob/master/snap/snapcraft.yaml

non-Jedi · August 13, 2020, 4:16pm

There don’t appear to be any plans to keep the official Julia snap up to date:

https://github.com/JuliaLang/julia/issues/32639#issuecomment-515080786

tbeason · August 13, 2020, 4:17pm

I really think jill doesn’t get enough love in these threads. If you are on Linux, do yourself a favor and use jill.

https://github.com/johnnychen94/jill.py

tamasgal · August 15, 2020, 5:38pm

Interesting that Microsoft seems to track download URLs to warn the user.

tamasgal · August 15, 2020, 5:42pm

I personally prefer environment modules. I only have to maintain one system to install things which are either not present in the distributions repository or too outdated for my taste.

But jill seems also be quite useful.

tbeason · August 15, 2020, 7:21pm

Oh that’s pretty cool. I guess my HPC cluster uses that or something like it. I never thought about doing it on my own system.

If I used anything other than the latest Julia release that might be useful.