4 ms·

GitHub stars + issues + PRs is going to favour languages developed on GitHub rather than mirrored on GitHub. It also favours languages undergoing rapid developm

GitHub stars + issues + PRs is going to favour languages developed on GitHub rather than mirrored on GitHub. It also favours languages undergoing rapid development compared to those that move at a slower pace. Neither of these have anything to do with popularity, which means people actually using it. Here are some better proxies for popularity: - What people are searching for on search engines (https://trends.google.com/trends/explore?date=2012-07-31%202022-07-29&geo=GB&q=%2Fm%2F07sbkfb,%2Fm%2F0jgqg,%2Fm%2F01t6b,%2Fm%2F05z1_ https://trends.google.com/trends/explore?date=2012-07-31%202...) - How many questions they’re asking on StackOverflow (https://insights.stackoverflow.com/trends?tags=java%2Cc%2B%2B%2Cc%2Cpython https://insights.stackoverflow.com/trends?tags=java%2Cc%2B%2...) - Developer surveys (https://survey.stackoverflow.co/2022/#technology-most-popular-technologies https://survey.stackoverflow.co/2022/#technology-most-popula...). But as long as we’re not measuring popularity based on number of results returned by Google, we’re probably fine. (https://blog.nindalf.com/posts/stop-citing-tiobe/ https://blog.nindalf.com/posts/stop-citing-tiobe/)

toyg 4y ago
Actually, searches and SO questions would correlate with languages that are harder/less intuitive, rather than necessarily popular. It's like Distrowatch with Linux, which favours "flavour of the day" distributions people are looking for, rather than the well-established ones that actually dominate deployments - because nobody needs to search for those. Developer surveys are good but they typically self-select by audience. JetBrains surveys will over-represent core JB audiences (e.g. Java developers), SO other constituencies (newbies, MS...), etc etc.
- nindalf 4y ago
  Right, which is why I said they're proxies for popularity. We don't actually know what people are running on their computers so we use these proxies. They're not perfect, but they're a lot better than number of stars on the Github repo (TFA) or number of Google results (Tiobe). I also realise that StackOverflow has a bias towards newbies but that's not a bad thing. It just means that it's a forward looking prediction. The newbies of today are the grizzled veterans of 2030.
- hombre_fatal 4y ago
  > SO questions would correlate with languages that are harder/less intuitive, rather than necessarily popular People repeat this a lot because it sounds reasonable, but I don't see why it has to be the case. I would expect popularity to still dominate the SO question signal even if a handful of experts aren't asking questions on SO because they can RTFM. Newcomers to a language seem to ask a lot of questions no matter what, so bad SO signal seems to mean decreasing popularity. Also, the biggest questions on SO are super basic questions that not even the most intuitive, best documented language can escape. I don't see good evidence for the claim.
dustedcodes 4y ago
> What people are searching for This is an extremely bad indicator for anything. What people search for does often not correlate to what they choose to use in their projects in the end. Take crypto/web3/blockchain for example. Lots of searches, but not many people who build a blockchain. > How many questions they’re asking on StackOverflow Not necessarily a good indicator for popularity (although it will play a role). I'd say it's more of an indicator how much developers struggle with a particular language. > Developer surveys Better, but even those have a strong bias based on their historic audience. StackOverflow is overrepresented with developers from 1-2 languages who have been historically the first/biggest community there.
ehnto 4y ago
Not to mention, commercial code is not going to be well reflected on github. This might show what is popular in OSS but not necessarily what industry trends are happening in production use.
mseepgood 4y ago
> number of results returned by Google That would, to a large extent, measure the age of a language. The longer it's been around, the more documents on the web.
- nindalf 4y ago
  Sort of, but this metric can also be gamed. Look at this effort from 2008 to game the metric in favour of Delphi (http://delphi.org/2008/10/delphi-keeps-climbing/ http://delphi.org/2008/10/delphi-keeps-climbing/). That's why Delphi remains high in this useless, meaningless ranking even though no one uses Delphi.
nsajko 4y ago
> How many questions they’re asking on StackOverflow Apart from already mentioned issues, another reason why this doesn't make sense as a metric is that not all language communities gravitate towards Stack Overflow. Julia, for example, has its own Discourse instance: https://discourse.julialang.org https://discourse.julialang.org
withinboredom 4y ago
issues + PRs probably is a better metric of language stability/maturity, and github stars is a (obvious?) metric for how many people on github like projects. Nothing in there about popularity. At all.
Tozen 4y ago
> How many questions they’re asking on StackOverflow StackOverflow is no more valid a metric than GitHub. Off the top of my head, Pascal, and all its dialects and variants, comes to mind. Many Pascal users and those of the many dialects that it has, just about never use StackOverFlow. They can use websites that correspond to their dialect or variant (Delphi, Object Pascal, Free Pascal, Oxygene...), books (which there are many), etc... But, for the newer languages, I do think GitHub is much more representational. Not just the stars, but adding other GitHub metrics, can give a more wholesome picture. I do think that each language, along with the ecosystem and culture that surrounds it, has to be examined by using various metrics and different sources in order to get a more complete picture. GitHub metrics are telling us a lot, just not the entire picture.

toyg 4y ago

Actually, searches and SO questions would correlate with languages that are harder/less intuitive, rather than necessarily popular. It's like Distrowatch with Linux, which favours "flavour of the day" distributions people are looking for, rather than the well-established ones that actually dominate deployments - because nobody needs to search for those. Developer surveys are good but they typically self-select by audience. JetBrains surveys will over-represent core JB audiences (e.g. Java developers), SO other constituencies (newbies, MS...), etc etc.

nindalf 4y ago
Right, which is why I said they're proxies for popularity. We don't actually know what people are running on their computers so we use these proxies. They're not perfect, but they're a lot better than number of stars on the Github repo (TFA) or number of Google results (Tiobe). I also realise that StackOverflow has a bias towards newbies but that's not a bad thing. It just means that it's a forward looking prediction. The newbies of today are the grizzled veterans of 2030.
hombre_fatal 4y ago
> SO questions would correlate with languages that are harder/less intuitive, rather than necessarily popular People repeat this a lot because it sounds reasonable, but I don't see why it has to be the case. I would expect popularity to still dominate the SO question signal even if a handful of experts aren't asking questions on SO because they can RTFM. Newcomers to a language seem to ask a lot of questions no matter what, so bad SO signal seems to mean decreasing popularity. Also, the biggest questions on SO are super basic questions that not even the most intuitive, best documented language can escape. I don't see good evidence for the claim.

dustedcodes 4y ago

> What people are searching for This is an extremely bad indicator for anything. What people search for does often not correlate to what they choose to use in their projects in the end. Take crypto/web3/blockchain for example. Lots of searches, but not many people who build a blockchain. > How many questions they’re asking on StackOverflow Not necessarily a good indicator for popularity (although it will play a role). I'd say it's more of an indicator how much developers struggle with a particular language. > Developer surveys Better, but even those have a strong bias based on their historic audience. StackOverflow is overrepresented with developers from 1-2 languages who have been historically the first/biggest community there.

ehnto 4y ago

Not to mention, commercial code is not going to be well reflected on github. This might show what is popular in OSS but not necessarily what industry trends are happening in production use.

mseepgood 4y ago

> number of results returned by Google That would, to a large extent, measure the age of a language. The longer it's been around, the more documents on the web.

nindalf 4y ago
Sort of, but this metric can also be gamed. Look at this effort from 2008 to game the metric in favour of Delphi (http://delphi.org/2008/10/delphi-keeps-climbing/ http://delphi.org/2008/10/delphi-keeps-climbing/). That's why Delphi remains high in this useless, meaningless ranking even though no one uses Delphi.

nsajko 4y ago

> How many questions they’re asking on StackOverflow Apart from already mentioned issues, another reason why this doesn't make sense as a metric is that not all language communities gravitate towards Stack Overflow. Julia, for example, has its own Discourse instance: https://discourse.julialang.org https://discourse.julialang.org

withinboredom 4y ago

issues + PRs probably is a better metric of language stability/maturity, and github stars is a (obvious?) metric for how many people on github like projects. Nothing in there about popularity. At all.

Tozen 4y ago

> How many questions they’re asking on StackOverflow StackOverflow is no more valid a metric than GitHub. Off the top of my head, Pascal, and all its dialects and variants, comes to mind. Many Pascal users and those of the many dialects that it has, just about never use StackOverFlow. They can use websites that correspond to their dialect or variant (Delphi, Object Pascal, Free Pascal, Oxygene...), books (which there are many), etc... But, for the newer languages, I do think GitHub is much more representational. Not just the stars, but adding other GitHub metrics, can give a more wholesome picture. I do think that each language, along with the ecosystem and culture that surrounds it, has to be examined by using various metrics and different sources in order to get a more complete picture. GitHub metrics are telling us a lot, just not the entire picture.