Language and standard lib status?

[Benjamin Silbaugh]

What is the current development status of the Julia language and standard library? Has the project reached a sufficiently "stable" state that one could rely on it as a tool for constructing numerical simulations (e.g. FEM work). Has the Julia dev team made any commitments to maintain backwards compatibility with the current version (at least for the next year or so)?

I'm interested in experimenting with Julia; however, if the language and/or standard libs are still in a state of flux, then I'll hold off.

I apologize if I'm repeating a FAQ, but I couldn't find a straight answer online. On a related note, it might be helpful to add a "project status" section to the Julia website and/or README in the source.

Thanks!

[Harlan Harris]

On Thu, Apr 11, 2013 at 6:36 PM, Benjamin Silbaugh <ben.si...@gmail.com> wrote:

What is the current development status of the Julia language and standard library? Has the project reached a sufficiently "stable" state that one could rely on it as a tool for constructing numerical simulations (e.g. FEM work). Has the Julia dev team made any commitments to maintain backwards compatibility with the current version (at least for the next year or so)?

The development branch is continuing to evolve pretty rapidly. In the last week or two, among a number of other changes, they've added named/default function arguments, and changed module import syntax in a highly breaking manner. On the other hand, if you're OK with using the version of the language on the Downloads page, version 0.1.2, then that (more precisely, the 0.1 branch and its point releases), should be stable and backwards-compatible indefinitely. Note that the add-on Packages are considerably less stable than the core, and that you might have trouble using some of them with 0.1. However, if you're doing FEM and really just focusing on linear algebra and numerical simulation, that may be OK for you...

Side note: I see that you're in the DC area! (Rapportive is great...) There are a few of us in the region who are interested in Julia and who have posted on the mailing lists. Someday we'll drag/courteously invite Stefan down from NYC and make him/ask him to present at a Meetup...

 -Harlan

[Benjamin Silbaugh]

Thanks for the feedback.

Once Julia is further along, I think it would be interesting to see Julia applied to a problem in Computational Fluid Dynamics, or some other non-trivial numerical problem. Talking about the theoretical capabilities of Julia and showing micro-benchmarks is a sensible place to start; however, it still leaves a lot of room for skepticism. Creating a "real-world" scientific application would provide hard evidence for (or against) Julia.

Something to think about.

Keep up the great work!

[Tim Holy]

On Thursday, April 11, 2013 06:54:39 PM Benjamin Silbaugh wrote:
> Thanks for the feedback.
>
> Once Julia is further along, I think it would be interesting to see Julia
> applied to a problem in Computational Fluid Dynamics, or some other
> non-trivial numerical problem. Talking about the theoretical capabilities
> of Julia and showing micro-benchmarks is a sensible place to start;
> however, it still leaves a lot of room for skepticism. Creating a
> "real-world" scientific application would provide hard evidence for (or
> against) Julia.

There seem to be several of us running real-world applications now, and what
few reports are available indicate that Julia gives rather significant boosts
over Matlab/R, sometimes by even more than the benchmarks might suggest. That
was surprising to me, since I expected the gap to be largest for benchmarks,
which naturally tend to be targets for optimization. I think for my case and
one recently described by Doug Bates, one common factor was fairly sizable
(but not ridiculous) memory requirements; perhaps Julia's ability to manage
memory in a more fine-grained fashion pays major dividends for such problems.
These advantages are not well-demonstrated by the current benchmarks, so in
fact the benefits of using Julia for practical work are in some cases
_underestimated_ by the benchmarks.

That said, each person has his/her own sense for when to join a project, and
Julia is indeed in a state of fairly rapid flux. This speed of improvement is
overall a very good thing, but it's definitely a lot to keep up with.

When you decide to jump aboard, we'll look forward to your contributions.

--Tim

[Viral Shah]

I personally feel that Julia is good enough and stable enough for FEM applications and such. People in this thread have already talked about using julia for non-trivial applications.

The language has largely stabilized. New language features such as keyword arguments have been added, but existing syntax has remained stable for over a year now.The base APIs are reasonably stable, but a small fraction does undergo some churn. This may cause a bit of pain across releases, but the deprecation mechanism should alert you about these changes from release to release. Often this is a better function name, or a newer calling convention. Depending on the context, this may or may not be a concern.

That said, we will not be doing backwards compatibility or the equivalent of long term support until we hit 1.0. Also, Julia 0.2 is just around the corner.

-viral