There are two issues here:

1. Literal speed – the benchmarks indicate Julia being 2x slower than fast
languages. They benchmarks, however, are including Julia startup time and
JIT compilation. So being 2x slower than C is pretty damned good.

2. Development speed – "Julia instead is slowing down" is not, unless I'm
seriously misreading it, about performance regressions, but about
development velocity and rate of adoption. There is certainly less hype
around Julia right now than a year or two ago, judging e.g. by number of
Julia-related posts on HN per unit of time. On the other hand, there are
far more serious users of the language (the NY Fed, for one obvious
example), and the growth of the ecosystem in terms of number of packages
continues in the same linear fashion <http://pkg.julialang.org/pulse.html>
as it has since we introduced a package manager.

In terms of development of the language itself, I can see how there could
be an impression that it's slowed down since the features being tackled
these days are much larger and more difficult. As a result, the time
between them is correspondingly longer. But 0.4 introduced function call
overloading, docstrings, generated functions – a feature that no other
mainstream language seems to have, which is nearly a superpower for writing
general and efficient numerical libraries – package precompilation, and a
new, much faster garbage collector, just to name a few.

The 0.5 release is going to introduce many even more impressive features,
including: a number of important changes to the behaviors of arrays and
strings, a revolutionary improvement in the performance of higher order
programming – you'll have to try it to really understand what a big deal
this is – real multithreading as an experimental feature, and much better
support for static whole-program compilation. These are huge features that
don't exist in any other mainstream dynamic language and it's vital to
tackle them in Julia while the language is still malleable enough that we
can. Think about how many early decisions in other languages ended up
making features like multithreading and static compilation impossible. In a
year, radically redesigning the way closures and higher-order functions
work would be far more difficult, and maybe impossible.

Perhaps Julia is in a lull in the hype curve at the moment. But if so, I
think that's ok – we're doing crucial work in preparation for Julia 1.0 –
work that will let us race far ahead into the future.



On Tue, Jan 26, 2016 at 5:47 PM, Ismael Venegas Castelló <
[email protected]> wrote:

> I was also saddened to see this observation:
>
> *Julia instead is slowing down and it disappeared from my radar during the
> last 9 month. It is still young and promising but it slowed down a bit this
> year so I don’t think it deserve the Top 5 for now.*
>
> http://www.davideaversa.it/2015/12/the-most-promising-languages-of-2016
>
> *@ylluminate*: that is just one opinion! ;)
>
> El martes, 26 de enero de 2016, 16:36:22 (UTC-6), Ismael Venegas Castelló
> escribió:
>>
>> Continues:
>> https://github.com/JuliaLang/julia/issues/14808#event-527206420
>>
>

Reply via email to