Visualizing Language Migration Over Time
Visualizing Language Migration Over Time
Written by Janet Swift   
Friday, 14 July 2017

It's not unusual for experienced programmers to switch from one language to another. This could be to handle the requirements of different projects or just to try out new options. Whatever the reason there's quite a lot of migration, both temporary and permanent.

Given I Programmer's Interest In all computer languages, a recent post by Waren Long, a Machine Learning Intern, on the source{d} blog tackled a very relevant topic - developers changing the languages they code in over a time period starting in 2000 and spanning 16 years. The approach used was as fascinating as the result themselves and Long provides a lot of detailed math and statistics that is well worth reading about from a data science point of view. The scripts used, for the analysis and the blog post itself are all open source and available.

The inspiration for this study of GitHub code was a blog post in March this year from Erik Bernhardsson who, in The eigenvector of “Why we moved from language X to language Y” tackled the question:

Is it possible to generate a N * N contingency table of moving from language X to language Y?

Bernhardsson's analysis was on Google queries related to changing languages and covered 25 languages. The one glaring omission from the list of languages was JavaScript which Erik explained with two reasons:

“(a) if you are doing it on the frontend, you are kind of stuck with it anyway, so there’s no moving involved (except if you do crazy stuff like transpiling, but that’s really not super common) (b) everyone refers to Javascript on the backend as ‘Node’”. Our data retrieval pipeline could not distinguish regular JS from Node and thus we had to exclude it completely.


Whereas Bernhardsson's analysis reflected hypothetical language migration and included those who were considering or investigating making a switch, Long's, which is based on GitHub source code rather than Google searches  gives information about the proportions who actually made a switch, or didn't. The dataset the source{d} had at its disposal was: 

  • 4.5 Million GitHub users
  • 393 different languages
  • 10 TB of source code in total

Some preliminary analysis was done to eliminate "Hello world" GitHub repositories from the dataset and then a transition matrix was computed between consecutive years for GitHub users and summed over users and over the last 16 years. The results were plotted on a grid using Bernhardsson's script, which makes it easy to get an overview of the differences, and similarities between the two results.

 

erikcontingency

 

flowtransmatrix

 

The two grids list the languages in alphabetical order and empty rows and columns in the source{d} matrix are those for Cobol, Kotlin and Lisp where were not found in the GitHub data. Although the numbers in the two grids are very different, the shading is based on a logarithmic scaling from 1 to the maximum value - so represent density.

The other big difference is that, whereas the diagonal in the contingency table is blank (you can't consider switching from language X to language X), in the flow transition matrix it isn't. In fact it always contains the most dense shade in both its row and column and represents those who don't switch language and use the same one from year to year.

In his comparison of the popularity of languages across the two analyses, Long writes:

Python (16.1 %) appears to be the most attractive language, followed closely by Java (15.3 %). It’s especially interesting since only 11.3 % of all source code on GitHub is written in Python.

In Erik’s ranking, Go was the big winner with 16.4 %. Since Erik based his approach on Google queries, it seems that the buzz around Go, which makes people wonder explicitly in blogs if they should move to this language, takes a bit of time to produce projects effectively written in Go on GitHub.

Furthermore, C (9.2 %) is doing well in accordance with Erik’s grading of 14.3 %, though it is due to the amount of projects coded in C on GitHub.

Although there are ten times more lines of code on GitHub in PHP than in Ruby, they have the same stationary distribution.

Go (3.2 %) appears on the 9th position which is largely honorable given the small proportion (0.9 %) of Go projects which are hosted on GitHub. For example the same proportion of projects are written in Perl, but this language doesn’t really stir up passion (2 % popularity).

Popularity ranking, with most popular language at the bottom, is used for this visualization.

The following transition matrix shows the proportions of GitHub users going from language X to language Y and vice versa. So for example it shows that 40% of Scala users switch to Java, whereas 4% of Java users switch to Scala. If you sum the proportions they will fall short of 100%. The shortfall is the proportion who stick with their language year on year.

propmovematrix

 

Picking out noteworthy points, Long comments: 

  • Developers coding in one of the 5 most popular languages (Java, C, C++, PHP, Ruby) are most likely to switch to Python with approx. 22% chance on average.

  • Besides, according to Erik’s matrix, people switch from Objective-C to Swift and back with greater probabilities - 24% and 19% accordingly.

  • Similarly, a Visual Basic developer has more chance (24%) to move to C#while Erik’s is almost sure in this transition with 92% chance.

  • Users of Clojure, C# and, above all, Scala would rather switch to Java with respectively 22, 29 and 40% chance.

  • People using numerical and statistical environments such as Fortran (36 %), Matlab (33 %) or R (40 %) are most likely to switch to Python in contrast to Erik’s matrix which predicts C as their future language.

  • One common point I found with Erik’s results about Go is that it attracts people who gave up studying Rust.

Long picks out four matrices from different timeline intervals which he notes show the same language profile every year, i.e. the deepest shades in the same positions. Here are two from a decade apart, 2005-2006 (with many fewer languages) and 2015-2016. 

languagematrix2

 

languagematrix

 

Finally to put all the data about language useage together Long produces this chronological sequence in which the thickness of each band corresponds to the value in the dominant eigenvector.  

languagepop

Long comments:

  • The first two languages, Python and Java have the same profile. They have been taking the place of C for 15 years. Indeed, the aggregation of these first 3 layers gives a straight one.

  • The attractiveness of C++ dropped prominently in 2008 when languages like Java or Ruby started growing rapidly. Nevertheless, it has been sustaining its popularity ever since this period.

  • I definitely support Erik’s conclusion that Perl is dying.

  • Apple presented Swift on WWDC’2014 and it was supposed to replace Obj-C. So Obj-C adoption should start to decrease after that event, but the sum of both languages should remain the same. Looking at the figure, this hypothesis turns out to be right.

  • Ruby appears to have had 6 years of glory starting from 2007. It might be explained with the launch of the web framework, Ruby on Rails (RoR), which reached a milestone when Apple announced that it would ship it with Mac OS X v10.5 “Leopard” - released in October.

  • Regarding Go, the popularity stays relatively low. However, the dynamics is clearly positive.

To understand the final comment you need to know that Go emerged as the front runner in Bernhardsson's analysis, to which he commented: 

Surprisingly, (to me, at least) Go is the big winner here. There’s a ton of search results for people moving from X to Go. I’m not even sure how I feel about it (I have mixed feelings about Go) but I guess my infallible analysis points to the inevitable conclusion that Go is something worth watching.

Go is a language there is a lot of hypothetical interest in. Whether it really is the language of the future can be the subject for a repeat analysis at some point in the future. 

  

More Information

Analyzing GitHub, how developers change programming languages over time

The eigenvector of "Why we moved from language X to language Y"

Related Articles

Go Language Of The Year With Dart Catching Up

Most Popular Computer Languages 2015

JavaScript Is The Language Of 2014 

Programming Languages An Infographic

To be informed about new articles on I Programmer, sign up for our weekly newsletter, subscribe to the RSS feed and follow us on, Twitter, FacebookGoogle+ or Linkedin.

 

Banner


Q.bo One - Second Generation Robot For Developers, Hobbyists and Kids
24/06/2017

A new robot from the TheCorpora Robotic Company has been launched on Indiegogo, where you can order them, in kit form and fully built, for delivery in December. Q.bo One is an open source standard rob [ ... ]



The Future Development Of Kotlin
28/06/2017

With Google's promotion of Kotlin to a supported Android language, the way that it is going to develop is much more important. We now have the results of the JetBrain's survey of what programmers want [ ... ]


More News

 
 

 

blog comments powered by Disqus

Last Updated ( Friday, 14 July 2017 )
 
 

   
Banner
Banner
RSS feed of news items only
I Programmer News
Copyright © 2017 i-programmer.info. All Rights Reserved.
Joomla! is Free Software released under the GNU/GPL License.