Talk:Julia (programming language)

Checked overflow support on all integer operations, allowed now. Reworked some more of this text that I deleted earlier.

In Julia all data types are supported in a completely general way (not needing manifest typing, that is optional); even those commonly implemented in processors, such as integers (signed and unsigned) from 8-bit to 64-bit, defaulting to signed 32-bit or 64-bit depending on CPU architecture (and additionally 128-bit and arbitrary-precision arithmetic "BigInt", which are slower as not supported in hardware by CPUs); all common/fast floating-point types up to 64-bit/"double" (and a package bigger using the double-double arithmetic trick), even decimal floating-point type (not supported in hardware by x86/ARM CPUs etc. requiring software emulation) in a package Decimals.jl. In general, all types, whether built-in or not, are treated the same, with "no abstraction penalty" (while software emulation, would necessarily slow down, when CPUs support a type if Julia wouldn't). While CPU registers bigger than 64 bits are available in CPUs, such as in x86, they are SIMD-style vector registers of many 32-bit values (or smaller), also supported in Julia at high speed (might in some cases need the @simd macro).

Complex numbers are supported by the language.

Julia uses efficient "machine integers" (e.g. modular numbers), not checking overflows (does for floating point), by default, while separate functions allow that. "integer arithmetic overflow" is discussed for addition by version 1.0 and at JuliaCon 2014, but may or may not end up being the default. The performance section of Julia's manual justifies non-checked machine integer use (that is faster), compared to saturated (that e.g. MATLAB uses), or exceptions on overflow. Julia however has "checked integer conversions" in 0.4.0-rc3. comp.arch (talk) 09:07, 9 December 2015‎ (UTC)

References

Criticism

Currently the article says nothing bad about Julia. Its tone is pretty dignified --- not quite cheerleading/advocacy --- but it still reads as if Julia solves all problems with no drawbacks. Some balance would improve the article. For example, this user survey suggests some room for improvement (although a secondary source would be better). Mgnbar (talk) 19:46, 8 February 2020 (UTC)

The drawbacks of Julia are pretty obvious, they are mostly inherited from drawbacks of its paradigms and design choices. For example, there is a compile time attached to pretty much everything you do in Julia for the first time, because it's a compiled language. So the fact that we refer to ahead-of-time compilation is a big hint towards that. We could create a section and populate it with these drawbacks, however. For reference, Python doesn't have a section like that either, nor is the text particularly hinting at its disadvantages. Bruno H Vieira (talk) 14:50, 5 April 2020 (UTC)

As you say Python has no drawback section, and its pros and cons are somewhat opposite to Julia. Julia however also has an interpreter, a non-default and undocumented... --compile=min option to enable it so you can get roughly Python behavior, avoiding that "compile" overhead. Language (by default) can have only either or, and both have pros and cons. There is a third hybrid option "tiered compilation"[1] (also with pros and cons, while trying to get most of the pros minimizing the cons) in C#, not yet in Julia. comp.arch (talk)

Following up on this. The article seems to read as an advertisement right now, most egregiously/noticeably the Notable uses section. Would be great to have another editor check this. For starters, I would add a criticism section, note drawbacks in various sections, and "neutralize" the tone in several parts of the article. — Preceding unsigned comment added by 24.238.13.130 (talk) 21:22, 1 July 2020 (UTC)

Just to grab a few pieces which might be contributing to an overly positive message:

• "While it is a general purpose language and can be used to write any application, many of its features are well-suited...", which is indeed sourced, but may still be too opinionated. Alternatives could be more in the vein of "is designed for". One of the sources here is "Out in the Open: Man Creates One Programming Language to Rule Them All", so it may be possible that finding a wider variety of sources could be beneficial here?
• "Julia is in practice interoperable with many languages" (unsourced)
• Lots of "also"s, or things which do not technically contribute but may cause a biased tone overall: "Julia 1.4 allowed better syntax for..." that could be replaced with a less opinionated or more specific wording.

Besides the notable users, The "Current and future platforms" section also feels a bit like an advertisement or maybe just confused. It compares Julia's JIT to other languages and even brings up the support tiers. It goes into a very high depth here, although I think we should expect more detail for a numerical language than most conventional languages. Yafwa (talk) 04:33, 5 July 2020 (UTC)

On "Julia is in practice interoperable with many languages", I believe I wrote it, and a lot of other, some you point out, so feel free to amend. Where do you draw the line with "many" (you can even use Julia from within Excel[2] but that's proprietary, and otherwise I only list no-cost solutions)? You can call to/from Python, R, C, C++, Rust, Zig, (Swift I guess), and of the top 20 language on July 2020 TIOBE index list, I only see Scratch as problematic. You can call C#[3] (while not much done), i.e. .NET and thus I believe Visual Basic .NET (wouldn't cover Classic, but I'm unsure if important or how easy to do differently). For some languages you can only (in practice, currently, as you can interoperate with any language theoretically both ways) call from Julia, e.g. now to MATLAB, Java (not back as far as I know, except with callbacks). You can call to JavaScript, meaning Node.js (could call from but that package is outdated), or use in a web browser or through WebAssembly. I've called Octave from Julia, but that was though Python. You can call all languages with a C API (i.e. without garbage collection), why I listed Zig without knowing anyone actually having done that.. You can call from Ruby[4] and more, so in some sense those are interoperable but I've not seen a package to call in the other direction. "we send an integer value from Julia to Go and back via a C shared library (written in Go)."[5] You could call to/from PHP and Perl [6] but that package to do it is outdated, only needs to be revived for Julia 1.0+, and you can still shell out and call those language, or from that way or differently (web services). comp.arch (talk) 13:47, 8 July 2020 (UTC)

Thanks for the response. I'll put the caveat that I'm relatively new to Wikipedia, so I might not be perfect on the standards. The main issue in my mind is one of sourcing/notability. There's no sourcing on that statement, and it's something of an opinionated, abstract statement on its own. So the answer to where we draw the line on "many" is that we should avoid making judgments like that outright. While it may even be true that there are quite a few languages, making that sort of judgment could be original research. I can say that I personally cannot determine from this that Julia has necessarily more than other languages; most languages have a C FFI, and in many cases bindings to other languages.

I'm thinking a wording more like the following would work: "Julia has a foreign function interface to C, with libraries available to interact with other languages[...refs]". That sentence would avoid trying to judge how practical or complete the interoperability is. And of course, if there are other reasons that might be notable, we can mention those. Is interop a priority for the Julia language, or something which makes it stand out amongst languages in the same class? If some sources have identified Julia as being particularly strong in interop, it would also be notable. Yafwa (talk) 06:31, 11 July 2020 (UTC)

I just thought "many" would be uncontroversial at the time, when I put it in. To clarify, if you can call another language at all, you can usually call C, and then indirectly to a lot of other. In theory, all (those that can be embedded). Also you can interact with a web service (or ZMQ), so that way in theory also all languages, but often not what's meant. Yes, I think it's fair to say that interop is a priority for Julia, why it has the ccall keyword (and now macro too), it is unusual (no boilerplate, for many other languages it's the opposite). At https://github.com/JuliaInterop/ you can e.g. find MathLink.jl to call proprietary Wolfram Mathematica™, and I just discovered OctCall.jl, but I was calling Octave before, through Python (not a big hoop to jump through, but without would be slightly easier). I did not know about DotNET.jl Julia [heart .NET] package. You can call Julia from D, but as I know the guy personally who made it work, I only list from people I'm not familiar with, non-Julia websites. From different people in that thread: "Julia has 0-copy numpy array interop through PyCall.jl if that's what you're asking."[7] "There are only 2 languages with really good Python interop [..]One of them is Julia! [..] The other is Swift [..]And here is the kicker: Packages is Julia are far more interoperable than in any other eco system I have seen. Than means 10 packages in Julia can quickly do more than 50 packages in other language."[8] "Julia's `ccall` is great in terms of overhead[0], so calling Rust shared libraries is not a problem."[9]. A tutorial I found on interop and a meetup. And:

“When you see such a tweet from a renowned author about Python and Julialang interoperability then you feel good about investing time learning in #JuliaLang ”

[tweet]

"This is 100% Julia code, If you replace the first 3 lines with [those importing tensorflow and keras)], you can run the exact same code in Python. Talk about excellent interoperability!".[10]

I clicked on the tweet author's name and saw "Author of the book Hands-On #MachineLearning with #ScikitLearn, #Keras and #TensorFlow. Former PM of #YouTube video classification. Founder of telco operator." comp.arch (talk) 18:44, 14 July 2020 (UTC)

Following up on this, I still think the article reads like an advertisement, so I will be adding a cleanup template. Very Fantastic Dude (talk) 21:49, 12 March 2022 (UTC)

Very Fantastic Dude can you point at one, two (or more) issues you have with the article? I can't cleanup otherwise, or choose to disagree. That said, I'm not sure the advert template applies even if you feel it promotional, since Julia isn't a "product", in the sense of being sold for money, and the template is explained as "When to use: Add this to articles [when in] whole or part they are advertisements masquerading as articles." with a link to "Articles considered advertisements include those that are solicitations for a business, product or service, or are public relations pieces designed to promote a company or individual." Until we figure out/agree if the template applies at all, you could also move it to one section (there's a variant for sections) if you had only one in mind and/or use a different template that might apply. comp.arch (talk) 22:18, 12 March 2022 (UTC)

@Very Fantastic Dude: @Comp.arch: What do you think? I think it is time to remove this template. It has been there for 3 months and the issue has as far as I know not been very active. The only thing that I could maybe think of is moving the history section down and starting with the Language Features, because this is the section where the archievements of the language are stated (in particular the section on Notable uses). Starting with the features may set the tone of the article differently. However other sites about programming languages (eg. Python (programming language) start with the history as well. I personaly am not under the impression that this article reads like and advertisement, nor do I think that this template fits, because it does not hit any of the points in WP:PROMOTION.

Because the discussion seems to be dormant, I will be removing this template (see point 7 of WP:WTRMT). If someone does not agree with this, please feel free to put it there again and write on the talk page.

Happy editing TheFibonacciEffect (talk) 23:03, 18 June 2022 (UTC)

Vim and Emacs and other cruft

I think it's pretty silly that the article notes that one can use "Vim and Emacs" to work with Julia :-) In case it isn't clear, one could say the same thing about any textual programming language, and (for that matter) all written languages!

I think this is emblematic of the cruft in the article. 73.252.146.143 (talk) 21:16, 7 March 2020 (UTC)

Agreed, we could remove it. You can write Julia on paper as well, paper just lack the interpreters :). So in my opinion it makes sense to keep dedicated tools and extensions such as the one in VSCode, Jupyter and Juno. Bruno H Vieira (talk) 14:54, 5 April 2020 (UTC)

Usage section

I tentatively added a Julia (programming language)#Usage section, mimicking the one in Python (programming language). The idea is that we can refer to notable libraries and other usages of Julia here. There are a few red links already, but it's missing crucial stuff such as DifferentialEquations.jl et cetera Bruno H Vieira (talk) 14:58, 5 April 2020 (UTC)

I undid your revision @MrOllie:. Would you care to explain the rationale behind the removal of the Julia (programming language)#Usage section? Bruno H Vieira (talk) 14:57, 22 April 2020 (UTC)

It's a list of apparently randomly selected libraries. Wikipedia isn't a howto or link directory so such lists are off-mission for us. - MrOllie (talk) 15:02, 22 April 2020 (UTC)

@MrOllie: It's not a list of randomly selected libraries, but it's under construction. I'm adding the most prominent ones first. I added that section directly inspired by the Python#Uses section, so the precedent is out there. Bruno H Vieira (talk) 15:06, 22 April 2020 (UTC)

The Python section is based on independent reliable sources and mentions libraries that are called out as significant by such sources. Your section was a list of libraries with no apparent selection criteria. - MrOllie (talk) 15:20, 22 April 2020 (UTC)

Several entries in the Python section are not referenced, that's where I got the precedent to add the section to this article. If the entries over there were removed as well then I would understand it, but it's quite an established part of the article. Bruno H Vieira (talk) 15:49, 22 April 2020 (UTC)

I'm really trying to understand how to make this section right, don't get me wrong. Bruno H Vieira (talk) 15:51, 22 April 2020 (UTC)

Bruno H Vieira, I share your passion for for Julia and its libraries (at least DifferentialEquations.jl is state-of-the-art, and good sources should be possible to find). I was maybe to enthusiastic adding libraries at the time, and gave up on editing the article in 2015 and ended up deleting about half of it then. comp.arch (talk) 11:28, 30 September 2020 (UTC)

Is it time for some separate Julia packages to get their own page?

I have Flux.jl and/or Knet.jl in mind, and JuMP was first state-of-the-art Julia package, there's also e.g. SciML/DifferentialEquations.jl (and other in that ecosystem) and a ton more. Maybe Gen.jl or similar, I just do not follow such packages too much.

Project Jupyter has a page (in part because of Julia, why the prefix Ju-, it was formerly IPython), Pluto.jl seems to improve on it (is Julia-only, I'm pretty sure). comp.arch (talk) 13:21, 28 September 2020 (UTC)

I know you know it, but just so others know the precedent: Flux has one already :) Flux (machine-learning framework). Other notable Julia packages will get their too in the due time. I am enthusiastically in favor of DifferentialEquations.jl, but Gen.jl and Turing.jl are quite notable too. They were even referenced in other articles IIRC Bruno H Vieira (talk) 00:06, 9 March 2021 (UTC)

Strong Typing

I removed the claim that Julia is a "strongly typed language," as the citation given was poor and ambiguous. The citation linked to a forum thread, wherein a user claimed that Julia is "strongly typed," citing Wikipedia to do so, and then another user below them claims the opposite, also citing Wikipedia. A forum thread is already not a reliable source of information, but add to that that the cited thread leaves the question of whether or not Julia is strongly typed as inconclusive at best, and it seemed appropriate to not only remove the citation, but the potentially erroneous claim as well.

If this edit has been made in error, feel free to revert and add in a less ambiguous source. drgribb (talk) 20:30, 27 January 2022 (UTC)

Hi! Pondering on the correctness of typesystem statements here, it might be also useful to clarify the claim on parametric types in the table below: "Parametric types" header links to parametric polymorphism page which actually provides a definition of parametric polymorphism that conflicts with the definition of Julia, which instead more or less precisely satisfies the "ad-hoc polymorphism" definition. I believe the correct name for what describes the table column there is "indexed types" as in here https://en.wikipedia.org/wiki/Generic_programming . Exaexa (talk) 07:27, 17 May 2022 (UTC)