Julia v0.6.4 has been released

Julia 0.6.4, the fourth (and hopefully final) release in the 0.6 series, is now available on https://julialang.org/downloads. Binaries are available for Linux (i686 and x86-64), macOS, and Windows (32- and 64-bit). A list of commits included in 0.6.4 is available here.

As with all patch releases, 0.6.4 contains a number of bugfixes, so we recommend that everyone currently using 0.6.0, 0.6.1, 0.6.2, or 0.6.3 upgrade to 0.6.4.



Hi @ararslan ,

I’m on MacOS and install Julia by Homebrew, and it seems 0.6.4 still unavailable on Homebrew. Could you please update it when you are free?


Homebrew isn’t an official distribution platform for Julia, and the Homebrew cask for Julia isn’t maintained by the Julia project, but rather by Homebrew. So typically updates there happen on their timescales rather than ours. (Also note that the Homebrew formula, i.e. not the cask, for Julia is entirely unmaintained.)

That said, the required change is incredibly minimal so I submitted a pull request this time: https://github.com/Homebrew/homebrew-cask/pull/49490.


Thanks for your explain and your work! It is confirmed that 0.6.4 is available on Homebrew cask now.


Does it have https://github.com/JuliaLang/julia/pull/27020 merged in, which is critical for us to even move from 0.4 to 0.6?

Well, its not even merged into master, how can it be backported?

I originally made the PR against 0.6 (and it works in 0.6), then I was told to rebase it on master so that it could get merged in there and then backported to 0.6, except that merging into master is blocked because unrelated tests fail and I’m not in a position to fix tests for code that I haven’t touched. So now this patch which is forcing us to stay on 0.4 isn’t going to get merged into 0.6 because of unrelated breakages in master. That’s a broken process and is going to keep us stuck on an unsupported version of julia.

Yes, this sounds like a broken review process for pull requests. :open_mouth:
So perhaps there can be a version 0.6.5 that includes this pull request?

It’s highly unlikely that there will be a 0.6.5.

Assuming that PR is ironed out before too long, you should be able to switch to 0.7 once that’s released.

The PR appears to fail on an LLVM assertion. Does the same patch pass all tests on Julia 0.6?

It did the last time I tried it which admittedly was a while ago.

I’ve tried switching to 0.7, there’s no (easy?) way to write 0.7 code that’s backwards compatible with 0.4. Since all of our customers will still have code for 0.4, they’d be unlikely to upgrade to a new version of our platform if it breaks all their existing notebooks, and I’d rather avoid having to maintain two versions of our platform.

I’m sorry that PR did not get finished and merged. We cannot, however—for obvious reasons—merge PRs that don’t pass CI. There was an announced window for testing for the 0.6.4 release which would have been the time to make some noise about the PR and get it into mergeable state.

Making PRs to master and back-porting them is not written in stone but is the way things are generally done for good reason: we don’t want people making fixes on old release branches which are then lost because they are never forward-ported to master; so instead you are asked to make fixes on master if they are still relevant there—which this one seems to be—and we will back-port them for you. In this case it seems unclear that the fix works cleanly on either master or 0.6. While I understand and sympathize with your frustration, I can’t see what could be done without a PR that works and passes CI, if not on master than at least on the release-0.6 branch. If you want to make a new PR against the release-0.6 branch and it passes CI there, that helps move things forward since it indicates that the failure is due to some difference since 0.6.


I am never sure what the social norms are about this in the Julia repo. I usually just ask or make a friendly ping if CI passes (or fails unrelatedly) and the PR is not merged for 3–7 days.

Pinging is generally fine—it’s easy for people to forget things so a reminder is good. One thing to consider is that if a PR has been bumped multiple times and no response or progress has been made, then why is it stuck? If you’ve asked for help or feedback and not gotten any, try to get past that hurdle first: If you’ve asked for review and no one has reviewed, perhaps no one knows who should do the review, so ask who should review since more people can answer that question than could do a review—and potential reviewers may not be aware of your PR.

1 Like

One can generally figure out who the potential reviewers are from the history of commits, but giving a rough list in the relevant section of CONTRIBUTING.md could be helpful.

One more thing: if you do get that PR working, we would be happy to make a 0.6.5 release that includes it. That’s the best we can do now that 0.6.4 is already out. Hope that helps!