Julia v1.3.1 has been released

Announcements
1

Julia version 1.3.1, the first patch release in the 1.3 series of releases, is now available. You can get binaries for Linux (i686, x86-64, AArch64, ARMv7), FreeBSD (x86-64), macOS, and Windows (32-, 64-bit) at Download Julia.

As a patch release, 1.3.1 contains no new features or breaking changes, only bug fixes and performance improvements. You can see a list of commits included since 1.3.0 here. We recommend anyone currently using 1.3.0 upgrade to 1.3.1.

Note that 1.3 on Travis, AppVeyor (with Appveyor.jl), and Cirrus (with CirrusCI.jl) now refers to 1.3.1.

Enjoy!

38 Likes
2

Weird offtopic question. Will there every be a way to upgrade julia from the package manager? :slight_smile:

11 Likes
3

Thank you for this and all your efforts.
I see the list of commits and their comments,
and I’m sure everyone is tracking their individually important PR / bug fix etc.

But for the busy “elsewhere” it difficult to extract the META improvement for each
version.minor.minor so would it be possible to also publish :

  • A 2-3 sentence summary about what’s better with this version.

Add regular backtest results to Building a Julia Project - Travis CI for

  • Performance / Scaling
  • Model zoo MWE test results of say MLJ , Flux, ForwardDiff, Zygote etc. ?

Also sure that the group will have other ideas about what to track and report about
in each version at the final version build.

9 Likes
4

Since this is a patch update, it’s just a collection of miscellaneous bug/perf fixes; the commit list is about as good a summary as one can give. You should update in any case! It’s possible that there is a bug in one of the fixes, but if so, it isn’t currently known, so there’s no warning we could give.

4 Likes
5

When will it be available as a Julia Pro download?

6

I’ll just leave a friendly reminder here about the wonderful world of jill to install Julia on Linux/MacOS!

9 Likes
7

nice link tbeason!

1 Like
8

One could imagine doing this through the BinaryBuilder infrastructure, but the main non-trivial issue is that Julia doesn’t work as a cross compiler.

2 Likes
9

We are still working out whether JuliaPro should do 1.3 or directly go to 1.4.

5 Likes
10

For what it’s worth, I think it would seem a little strange for JuliaPro to skip an entire minor release of open-source Julia.

7 Likes
11

Binaries are now available for ARMv7 and are linked on the website along with the others. Apologies for the delay.

6 Likes
12

Hello. I installed Julia v1.3.1 on all of my machines, and ran the Basebenchmarks.jl. Then, only on my macOS machines (MacBook Pro and iMac; macOS Mojave 10.14.6) generated the following error:

using BenchmarkTools, BaseBenchmarks
BaseBenchmarks.loadall!() # load all benchmarks
results = run(BaseBenchmarks.SUITE; verbose = true) # run all benchmarks
... # many benchmarks went through ... then the following happened...
(22/55) benchmarking ("exponent", "subnorm", "Float32")...
ERROR: DomainError with 0.0:
Cannot be subnormal converted to 0.

Neither my Windows 10 (v 1909) nor Linux machine (Ubuntu 18.04 LTS) generated the same error. So, I’m wondering whether something wrong for the macOS version of Julia. I posted the details at benchmarking ("exponent", "subnorm", "Float32") produced the DomainError when ApproxFun package is also used · Issue #253 · JuliaCI/BaseBenchmarks.jl · GitHub. Thanks for your help!

3 Likes
13

I need to update what I posted earlier. In fact, the same problem occur not only on macOS, but also Windows 10 and Ubuntu Linux machines if I also run using ApproxFun, i.e.,

using ApproxFun, BenchmarkTools, BaseBenchmarks
BaseBenchmarks.load!("scalar")
results = run(BaseBenchmarks.SUITE["scalar"]["floatexp"]; verbose = true)
...
(22/55) benchmarking ("exponent", "subnorm", "Float32")...
ERROR: DomainError with 0.0:
Cannot be subnormal converted to 0.

So far, I haven’t encountered the same problem with the packages I typically use other than ApproxFun. I’ll also post the issue on ApproxFun github site.

14

This issue was resolved. It was neither the problem of Julia v1.3.1 nor the BenchmarkTools/BaseBenchmarks. It turned out that the library in FastTransforms.jl that was complied by gcc with -Ofast flag. The maintainer of that package replaced this flag by -O3, which creates no further issue of this. See benchmarking ("exponent", "subnorm", "Float32") produced the DomainError when ApproxFun package is also used · Issue #253 · JuliaCI/BaseBenchmarks.jl · GitHub for many unexpected turns until this resolution was reached.

1 Like
15

My Mac does stop me to run Julia 1.3.1 on atom since I upgraded it to catalina, so IOS tries to protect me from damaging software - does anyone know how to fix it ?

16

Go to Applications, right click on Julia and press Open and then ok.

17

Fedora is still using Julia 1.2 for default Julia installation.

18

I don’t think complaining here would help solving the issue, as no one here has any power to address it. You should file an issue in Fedora.

3 Likes
19

You can sudo dnf copr enable nalimilan/julia to add a with great expertise maintained Fedora repository of Julia. With this dnf update should include the latest release, usually within a few days.

3 Likes
20

will do.