Community sizes

Recently I read quip somewhere that if some plane had crashed after a
Haskell(?) conference, it would have wiped out most of the language
community.

This made me wonder just how big the communities flocking around various
functional languages really are. If I go only by my own perception
(availability heuristic, anyone?), the Haskell community must be huge,
all that talk about monads and stuff, certainly larger than the C#
community which I barely notice. Now, that can't be true, I obviously
need a reality check.

So, are there any somewhat objective numbers, maybe based on conference
attendance, book sales, or job offers? It would be interesting, to have
an idea of the relative sizes of, say, Erlang, Haskell, (S)ML, OCaml,
Scheme. I'm not looking for a contest where quality equates to
popularity, contrariwise, for some this might even be an eye-opener and
prompt them to move on to a less crowded language.

Michael

--
Michael Schuerig
mailto:michael@schuerig.de
http://www.schuerig.de/michael/

On Aug 18, 12:13*pm, Michael Schuerig <mich...@schuerig.de> wrote:
> Recently I read quip somewhere that if some plane had crashed after a
> Haskell(?) conference, it would have wiped out most of the language
> community.
>
> This made me wonder just how big the communities flocking around various
> functional languages really are. If I go only by my own perception
> (availability heuristic, anyone?), the Haskell community must be huge,
> all that talk about monads and stuff, certainly larger than the C#
> community which I barely notice. Now, that can't be true, I obviously
> need a reality check.
>
> So, are there any somewhat objective numbers, maybe based on conference
> attendance, book sales, or job offers? It would be interesting, to have
> an idea of the relative sizes of, say, Erlang, Haskell, (S)ML, OCaml,
> Scheme. I'm not looking for a contest where quality equates to
> popularity, contrariwise, for some this might even be an eye-opener and
> prompt them to move on to a less crowded language.
>
> Michael
>
> --
> Michael Schuerig
> mailto:mich...@schuerig.dehttp://www.schuerig.de/michael/


See http://www.tiobe.com/index.php/conte...pci/index.html

Erlang is ranked 41, Haskell is 43, ML 44, Caml 50 (Ocaml is not
listed separately). The big surprise might be that Lisp/Scheme is
ranked 17. Unfortunately Scheme isn't ranked separately.

I have no idea how scientific this ranking is.

-Tom Gordon

On Mon, 18 Aug 2008 04:12:37 -0700 (PDT), tfgordon
<Thomas.Gordon@fokus.fraunhofer.de> wrote:

>On Aug 18, 12:13*pm, Michael Schuerig <mich...@schuerig.de> wrote:
>> Recently I read quip somewhere that if some plane had crashed after a
>> Haskell(?) conference, it would have wiped out most of the language
>> community.
>>
>> This made me wonder just how big the communities flocking around various
>> functional languages really are. If I go only by my own perception
>> (availability heuristic, anyone?), the Haskell community must be huge,
>> all that talk about monads and stuff, certainly larger than the C#
>> community which I barely notice. Now, that can't be true, I obviously
>> need a reality check.
>>
>> So, are there any somewhat objective numbers, maybe based on conference
>> attendance, book sales, or job offers? It would be interesting, to have
>> an idea of the relative sizes of, say, Erlang, Haskell, (S)ML, OCaml,
>> Scheme. I'm not looking for a contest where quality equates to
>> popularity, contrariwise, for some this might even be an eye-opener and
>> prompt them to move on to a less crowded language.
>>
>> Michael
>>
>> --
>> Michael Schuerig
>> mailto:mich...@schuerig.dehttp://www.schuerig.de/michael/
>
>
>See http://www.tiobe.com/index.php/conte...pci/index.html
>
>Erlang is ranked 41, Haskell is 43, ML 44, Caml 50 (Ocaml is not
>listed separately). The big surprise might be that Lisp/Scheme is
>ranked 17. Unfortunately Scheme isn't ranked separately.
>
>I have no idea how scientific this ranking is.
>
>-Tom Gordon

Going by book sales or usenet groups or Google searches can be
misleading. A lot of people who have passing interest may purchase an
intro book or ask questions of a group or do a flurry of searches and
then ultimately forget about it. They may also be searching for
something related to the language such as GC, type checking, pattern
matching, etc. and not actually be interested in the language itself.

Sales of textbooks (which tend to be expensive) and intermediate or
advanced programming books are better indicators of interest because
people who buy those really are using the language or are serious
about learning it and not just playing around. However, many
languages have no advanced books and many more have only one to a few
titles. Lisp, Scheme and ML are notable exceptions with each having a
good selection of titles to choose from.

George

Michael Schuerig wrote:
> Recently I read quip somewhere that if some plane had crashed after a
> Haskell(?) conference, it would have wiped out most of the language
> community.

I believe that was about the GHC developers rather than the entire Haskell
community.

> This made me wonder just how big the communities flocking around various
> functional languages really are. If I go only by my own perception
> (availability heuristic, anyone?), the Haskell community must be huge,
> all that talk about monads and stuff, certainly larger than the C#
> community which I barely notice. Now, that can't be true, I obviously
> need a reality check.
>
> So, are there any somewhat objective numbers, maybe based on conference
> attendance, book sales, or job offers? It would be interesting, to have
> an idea of the relative sizes of, say, Erlang, Haskell, (S)ML, OCaml,
> Scheme. I'm not looking for a contest where quality equates to
> popularity, contrariwise, for some this might even be an eye-opener and
> prompt them to move on to a less crowded language.

One of the best objective measurements I have found is the Debian and Ubuntu
package popularity contest results. This is an automated measurement of the
number of users who have packages (such as a specific compiler) installed.
However, it is obviously only an accurate reflection of usage on Linux and
not on Windows.

I have analysed these results by programming language before:

http://flyingfrogblog.blogspot.com/2...guages-on.html

Here are some recent results:

Debian Ubuntu Total
C#
mono-gmcs: 1,891 13,922 15,813

OCaml
ocaml-nox: 1,920 8,715 10,635

Erlang
erlang-base: 1,322 9,279 10,601

Haskell 9,794
ghc6: 1,214 5,392 6,606
hugs: 683 2,505 3,188

Haskell programmers will almost all have either hugs or ghc6 or both
installed, so there are between 6,606 and 9,794 of them.

You also mentioned the huge amount of talk about Haskell. You may wish to
distinguish between people who talk the talk and people who walk the walk.
I recently gathered statistics about this and found that open source
software written in OCaml is 30x more successful than Haskell in terms of
users garnered per programmer:

http://flyingfrogblog.blogspot.com/2...virginity.html

Finally, I should note that Don Stewart posted a rebuttal to my findings on
the Haskell-Cafe mailing list and then had me banned before I could reply:

http://www.haskell.org/pipermail/has...st/046129.html

My answers are probably obvious but I shall repeat them here
anyway:

.. By Don's metric of chat-room lurkers, Haskell is twice as popular as C#.
So that is clearly a broken metric.

.. Unused libraries are worthless.

.. Arch Linux only has a disproportionate number of Haskell packages because
Don Stewart generates them himself. They never get used. Indeed, Arch Linux
itself is virtually unused. Don demonstrated this nicely by packaging the
next 30 lines of code that were posted to the mailing list into yet another
unused Haskell library for Arch Linux:

http://www.haskell.org/pipermail/has...st/046170.html

.. Although the Haskell list has several times as much traffic as the OCaml
list, both lists have ~100 active contributors according to Google Groups:

http://groups.google.com/group/fa.haskell/about
http://groups.google.com/group/fa.caml/about

In essence, a tiny number of very vocal Haskell proponents try to make it
look as if Haskell has many real programmers and success stories but the
reality is quite the opposite.

--
Dr Jon D Harrop, Flying Frog Consultancy
http://www.ffconsultancy.com/products/?u

Jon Harrop skrev:
>
> One of the best objective measurements I have found is the Debian and Ubuntu
> package popularity contest results. This is an automated measurement of the
> number of users who have packages (such as a specific compiler) installed.
> However, it is obviously only an accurate reflection of usage on Linux and
> not on Windows.

At least as regards Erlang, I wouldn't put too much stock in this
statistic (even though I thought a surprisingly high number seem to
use packages). For various reasons, partly historical, most Erlang
developers install from source. That goes for the applications too.


BR,
Ulf W

Ulf Wiger wrote:
> Jon Harrop skrev:
>> One of the best objective measurements I have found is the Debian and
>> Ubuntu package popularity contest results. This is an automated
>> measurement of the number of users who have packages (such as a specific
>> compiler) installed. However, it is obviously only an accurate reflection
>> of usage on Linux and not on Windows.
>
> At least as regards Erlang, I wouldn't put too much stock in this
> statistic (even though I thought a surprisingly high number seem to
> use packages). For various reasons, partly historical, most Erlang
> developers install from source. That goes for the applications too.

Another related metric is Google Trends, which plots the proportion of
Google searches for a given search term:

http://www.google.com/trends?q=erlang%2Cocaml%2Cf%23

Finally, I forgot to mention book sales:

http://radar.oreilly.com/2007/05/sta...ok-mar-15.html

Note that Erlang was not represented and OCaml failed to include OCaml for
Scientists which is the best selling book on OCaml. Also, I suspect there
is a lot of noise on those results for non-mainstream languages as they are
only averaged over a quarter.

--
Dr Jon D Harrop, Flying Frog Consultancy
http://www.ffconsultancy.com/products/?u

> You also mentioned the huge amount of talk about Haskell. You may wish to
> distinguish between people who talk the talk and people who walk the walk.
> I recently gathered statistics about this and found that open source
> software written in OCaml is 30x more successful than Haskell in terms of
> users garnered per programmer:
>
> http://flyingfrogblog.blogspot.com/2...virginity.html
>

Your conclusions contrast sharply with the results on Ohloh, which is
also a flawed benchmark.

Haskell
http://www.ohloh.net/languages/38

OCaml:
http://www.ohloh.net/languages/61

> In essence, a tiny number of very vocal Haskell proponents try to make it
> look as if Haskell has many real programmers and success stories but the
> reality is quite the opposite.

Although I have seen your remarks about how Haskell projects have
higher rates of failure than projects using other languages I have
never seen any metrics to back up this claim. The Debian and Ubuntu
package popularity contest results although objective are limited in
their scope and thus inaccurate metrics for the relative success of
projects written in a particular programming language.

Parnell

parnell wrote:
>> You also mentioned the huge amount of talk about Haskell. You may wish to
>> distinguish between people who talk the talk and people who walk the
>> walk. I recently gathered statistics about this and found that open
>> source software written in OCaml is 30x more successful than Haskell in
>> terms of users garnered per programmer:
>>
>> http://flyingfrogblog.blogspot.com/2...virginity.html
>
> Your conclusions contrast sharply with the results on Ohloh, which is
> also a flawed benchmark.
>
> Haskell
> http://www.ohloh.net/languages/38
>
> OCaml:
> http://www.ohloh.net/languages/61

On the contrary, those findings combine with my own to show that the Haskell
community are generating huge quantities of unused (useless?) code.

For example, Don Stewart just announced the creation of 500 new Haskell
packages for the Arch Linux distribution alone:

http://cgi.cse.unsw.edu.au/~dons/blo.../08/21#the_500

>> In essence, a tiny number of very vocal Haskell proponents try to make it
>> look as if Haskell has many real programmers and success stories but the
>> reality is quite the opposite.
>
> Although I have seen your remarks about how Haskell projects have
> higher rates of failure than projects using other languages I have
> never seen any metrics to back up this claim.

On the contrary, the evidence you just cited shows that hundreds of Haskell
packages are being created and the evidence I cited shows that all but two
(Darcs and HPodder) remain unused.

Surely the only logical explanation is that these Haskell packages are
failures, presumably because they are useless.

> The Debian and Ubuntu
> package popularity contest results although objective are limited in
> their scope and thus inaccurate metrics for the relative success of
> projects written in a particular programming language.

They are certainly imperfect but they are among the best indicators I have
found. Google Trends is another good one. The Windows platform is the
obvious omission.

--
Dr Jon D Harrop, Flying Frog Consultancy
http://www.ffconsultancy.com/products/?u

Jon Harrop wrote:
> Haskell programmers will almost all have either hugs or ghc6 or both
> installed, so there are between 6,606 and 9,794 of them.

I am using ghc-6.8.3 which is not yet in my distro. Also, one should not
forget that there are people who do not take part in such surveys. That
means the real numbers are higher (it is unclear how much).

Cheers
Ben

> On the contrary, the evidence you just cited shows that hundreds of Haskell
> packages are being created and the evidence I cited shows that all but two
> (Darcs and HPodder) remain unused.
>
> Surely the only logical explanation is that these Haskell packages are
> failures, presumably because they are useless.

The evidence you cited is based on a flawed premise: that all installs
of the libraries in question will be recorded by the Debian and Ubuntu
package popularity contest. Because that is not true then your
conclusion is flawed.

> They are certainly imperfect but they are among the best indicators I have
> found. Google Trends is another good one. The Windows platform is the
> obvious omission.

Google fight is more fun:
http://www.googlefight.com/index.php...word2=Haskell+