Julia ranks pretty good on a recent ranking: https://langpulse.top/
According to the link, it’s more popular than MATLAB and Fortran combined, judged from the number of public repos with recent pushes, FWIW.
When I look at those charts the first question that comes to mind is, “What happened on May 18?”
All the languages had what looks like a step-function change right after May 17. This is peculiar.
One possibility (and it is only a guess) is that the ranking algorithm is tweaked at that time affecting all rankings. I’ve seen this same phenomena with PYPL. This really shows that you should not take short term changes in the rankings too seriously. As a case in point, if you look at the TIOBE graph at top you will see one outlier data point in late 2023 where the Julia ranking was about 0.85 which was enough to put Julia in the top 20. Guess what? there were a bunch of news posts about how Julia had cracked the top 20. Pure garbage reporting.
That’s sort-of what I was getting at. If you see that large of an unexplained artifact in the data it’s a sign that your data sources are not of the highest quality. I wouldn’t pay too much attention to such results.
Correct. But my gut feeling is that there is information to be gleaned in the long term (like a decade) trends especially if different rankings are compared and combined.
Language popularity matters considerably less than it used to. Of course network effects are still important, but LLM’s can help fill out the “boring” parts of the ecosystem which had less traction previously. Many of the things I wanted to do with Julia but couldn’t due to lack of support for this sort of boring stuff (HTTP/2, gRPC, pretrained vision backbones) flipped from blocking to reasonably achievable with a bit of a effort. Today it’s more important how a language and surrounding ecosystem stands out / is different, and Julia has no shortage of these sorts of things.
While true, another aspect that concerns me is that newcomers may not be as attracted to the expressivity of a language (for a human programmer) as they were in the pre-LLM times.
OTOH, newcomers may be less dissuaded by unfamiliarity. I find myself using (relatively) less python now that claude can write Rust for me
Yes, https://langrank.hexq.ru/ =)
I think it’s possible to combine several orthogonal rankings to obtain less noisy results. Ideally, we should apply a multidimensional Kalman filter to extract the signal.