They’re taking what they can. The language isn’t getting the adoption that it would have if it was as good as the small community around it claims that it is. It just comes across as a cult.
I really like Julia. I wouldn't say I come off like a cultist, but I do think they spend an annoying amount of time on PR for themselves.
I think Julia is stuck in a spot where they are better than say R/Python for scientific computing on a fundamental language structure level and package manager level, but that it's not good enough to make up for the fact that those languages have a more robust ecosystem.
I think many people underestimate the inertia present in a large community, particularly one like scientific computing. I think Julia is much-much better than R/Python at basically everything important (as a language and platform). I certainly wouldn't call the python ecosystem robust. It's decades upon decades of ugly hacks and hacks upon hacks. The pandas internals are a Kafkaesque nightmare. But R/Python is where millions of people work daily and produce a huge amount of mindshare and libraries. You'd need an incredible amount of resources to replicate that in any other space.
I'd argue though that this mindshare is not there due to Python/R as a language (and platform) being better, but simply because there were no better alternatives 10-15 years ago and by now the sheer inertia makes it impossible to stop.
You'd need to convince a sufficient portion of people to move to Julia at more or less the same time. Few people want to be first movers. These tend to be the ones who actually care about the qualities of the platform, not just "get the work done and clock out".
If you are a data scientist, you won't be paid for moving to Julia. You'll be paid for coming up with working models. And you take a serious risk by moving to a new platform with little adoption. The platform might die, taking your tools and processes with it. You won't be able to rely on your colleagues advice about technical issues. You can run into bugs more frequently simply because fewer eyes have looked at the ecosystem.
The end result is that few people take the plunge.
I would outright disagree with this - data science and ML is a subfield of software where you have a much higher ratio of greenfield projects over the past decade, and you also have many people who are moving to the field or are starting out in it. It also attracts the kinds of people who will be willing to use cool tech just for the sake of it. If that's not conducive to picking a new programming language with lots of promises, I don't know what is. And they still aren't picking Julia.
What do you disagree with if I may ask? Most of the people I work with really like Julia, we have dabbled with it and would like to see it succeed python/matlab/r.
Why aren't we moving, you ask? Because moving to another ecosystem does not put bread on the table. I cannot go to clients and say: this past 6 months we made no improvements to our strategies, but look, we migrated to a new programming language that is used by a fraction of a percent of our peer group.
The field attracts clever people and there are many greenfield projects. But just because we start a new project, it doesn't mean that doing it in a tiny ecosystem is sensible. Some firms can do it. Jane Street has the means to basically be "OCAML The Systematic Trading Language". This is not a luxury afforded to most.
And thus we get a chicken and egg problem, where nobody wants to be the sole first mover as there's little advantage to it. At the same time, we all see that everyone would be much better off if we moved.
I've been using R since beta and Julia is the first thing that I've seen since then in numerical computing that reminds me of that time.
Julia still has a lot of empty library space compared to R or Python, and isn't perfect, but my guess is it will catch up. R did when it was being compared to SAS, fortran, C/C++, lisp, and so forth.
I'll be honest and say that I wish something else more general-purpose (to the point of say, having a bootstrapped compiler) would be in its spot but I can't really complain right now. Maybe something else will catch up.
I think for me personally is that R and python is a bit in denial about its performance limitations when it comes to hard problems. You pretty much have to drop down into C/C++ to address them, and for certain things, julia really does do many times better time-wise, without the weeds of C/C++. I think there's some people (myself probably included) that are tired of being forced to choose between the C/C++ and R/python worlds. I think there is a bit of overhyping and/or cult-like nature of julia but I also think some of it is trying to convince people that you don't have to choose between expressiveness and performance.
Since people have been able to choose what tools to use.
It might not be a perfect relationship - there will be lagging effects and other factors, it will vary based on the specific task, and some of it will be subjective - but you can’t pretend that there is no relationship at all.
If programming is your job you seldom have the choice of the programming language.
There are lits of external factors like costs, available support, availability of compilers for certain hardware, support of the existing tool chain, learning curve, career chances etc.
Python and JavaScript aren't the best but sufficient enough
> If programming is your job you seldom have the choice of the programming language.
This isn't true. Sure, you can't unilaterally pick whatever language you want when you work with other people. But for every piece of software ever built somebody had to decide what to build it with. And even though Julia has existed for 10 years, almost nobody is picking it over the alternatives.
Julia's first light may have been 10 years ago, but the first point that it really met a substantial fraction of promise was much more recently. Until 2.0 was released around 2000, Python really was pretty niche and even its popularity was even waning (according to TIOBE) until about 2014.
Java started with a first implementation in 1991, but I didn't find it usable for serious production until about 1999-2000. That isn't quite 10 years.
10 years from inception to mature-enough-for-production seems to be kind of the rule. Building support systems and a community is hard, hard work.
