I think the 265 code base estimate is a reasonable lower bound for the 150 respondants. An upper bound might be 4x higher or so, about 2k programs, for those 150 respondants. If you instead want a code base estimate for the total fortran programming community, beyond those 150, then some estimate is needed for the number of programmers. That would of course be many orders of magnitude larger than 150.
A search on GitHub leads to 45.3 k repos with Fortran as the primary language (Repository search results · GitHub). Of course, a lot of repos do not qualify as codebases and are merely tests and exercises. An interesting metric would be to know the fraction of codebases that are collaborative as compared to single-dev repos.
Taking the recent responses in order:
@johanna_p , thanks for your thoughts. I’ll respond to several items you raise:
-
Because the survey can be done anonymously, both in terms of the respondent not identifying themself and the respondent not identifying their codebase, it’s not straightforward to check on multiple people responding about the same codebase. We have some data, which we will report in a post in due course, to show that many codebases appear to be maintained by a single developer, for which of course there wouldn’t then be multiple respondents. There are a very small number of instances where two people have given us URLs to the same codebase, thus confirming some multiple reporting, and hence double-counting. This appears to occur infrequently so hopefully (
) wouldn’t meaningfully affect the results. -
Respondents shared URLs for about 70 codebases (we’ve removed duplicate URLs), of which the considerable majority are on GitHub, so we would be able to look at whether these projects are active.
-
We also want to better understand the number of closed source codebases. There are a small number of instances where respondents have provided a URL to closed-source codebase, explaining that, for example, the codebase would be accessible with an appropriate license.
-
I agree that it is extremely hard to estimate the number of Fortran codebases from our 150 responses. We need to start somewhere
. It’s also why we’re asking the community. -
In terms of PhD students, early career staff, and support staff: there is some data we will share on this in a subsequent post to this thread.
@RonShepard , thanks for your thoughts too. I’ll respond to a couple of items:
-
Your estimate of 4x for this sample (would that be >1K codebases?) is a helpful sense-check. Thank you.
-
The suggestion of estimating based on number of programmers is helpful.
-
Your suggestion that the number of Fortran programmes is “many orders of magnitude larger” is also a helpful sense-check.
-
and 8) are good examples of why we’re reaching out to the community.
@davidpfister , thanks for your thoughts as well.
- One of our colleagues is collecting data from GitHub on Fortran codebases. He presented on that briefly at FortranCon’25. I’ll ask him whether he can do some analysis on the characteristics of the GitHub codebases (e.g., collaborative vs single-dev projects, active/dormant, purpose of codebase) to see whether we can get something more meaningful from the ~43K codebases on GitHub.
Bummer! Missed this!
Sorry about that, @Hendrik . You’re very welcome to comment on the information we share here, or look at the open access repo (here) or message me and we can arrange a chat.
Continuing the discussion from Please participate: global survey on the state and impact of Fortran:
Type of codebase
We asked respondents what type of codebase they’d be responding about. A summary is provided below.
It’s clear that the majority of respondents focused on Fortran applications. As noted by @certik , earlier, the limited number of responses about tools may be explained by the fact that many tools for Fortran are not written in Fortran.
There’s a ratio of ~1:5 between libraries and applications. We wonder how that ratio compares to other languages, e.g., whether there be a similar ratio for Python or C/C++.
Sorry but the link address in “here” 10.0.20.161/zenodoXXXX is not accessible from the outside.
Does this work?
Yes it looks like it… but I was leaving my desk now. I will have a look at it :- ) Thanks!
1 Like
For example here there is something funky to think about. I have a couple of “main” codebases, which depend on smaller, utility libraries that I maintain and share across projects. So I have “main” → {“dep1”,“dep2”,“dep3”} but without being nitpicking, I’d say I work on multiple “main” codebases. However, the smaller deps could also be counted!
1 Like
That’s a helpful example, @jorgeg , thank you. Your comments have reminded me that we asked respondents about dependencies in the respective codebase. We have not yet looked at analysing those particular responses as we have a lot of data to work through.
Size of codebases, and some implications
We also asked respondents about the size of their codebase. We asked respondents to give us a qualitative response (e.g., “small”, “medium”, “large”) and then an estimate of the size of the codebase in terms of lines of code (which has its problems as a measure of size). We then constructed box plots of the responses, as below. (There is one potentially ‘spurious’ datapoint for the codebases described as “large”. Also, note the logarithmic scale on the y-axis.)
Amongst other uses, and benefits, these results help to convey to those outside the community the scale and complexity of Fortran codebases and therefore the extent of the challenge arising in continuing to develop such codebases (as well as the need for non-trivial resources to do so).
These results can also be used as evidence in the debate about investing in, or divesting from, Fortran: whichever way one turns and proceeds (e.g., to continue with Fortran or move off Fortran onto some other language), the way forward is likely to be enormously complex, effortful, costly, and risky.
I’m not sure whether the “works on, or with” is very clear either. I assume it doesn’t mean “use an application written in Fortran”, but how much do I need to do to be said to “work with” a codebase?Does it matter how often, or how recently, I used or contributed to these codebases?
My main codebase (Fortran, of course) links to several independent libraries by other teams, four of which are written in Fortran. It also ships with a post-processing tool written in Fortran. Depending on my interpretation of this question, I could pick any number from 1 to 6 just thinking of the main application we develop, without even thinking about the other applications to which we have contributed.
1 Like
Thanks for your comments, Phil (@pjh503 ). It’s very difficult to get the wording right on these questions. Your comment on the phrasing of questions, and your experience, e.g., of being involved in multiple Fortran codebases, are useful qualifiers for how the results should be interpreted.
Yes, it is definitely very tricky to get the wording of these things right. The survey is already very useful and interesting, it just means the analysis is a bit more nuanced.
1 Like