How programmers transition between languages

[Russel Winder]

On Sun, 2018-01-28 at 18:54 +0000, Laeeth Isharc via Digitalmars-d
wrote:
[…]
> That's what you would expect, because D is a very ambitious 
> language, which means its natural user base is much more spread 
> out and less highly concentrated.  And beyond that, most code is 
> enterprise code that's closed source, and whilst the web guys 
> talk a lot and influence the culture, enterprise guys talk much 
> less and just do their thing quietly.  Even in our world, how 
> often do you see the people using D get involved in forum 
> discussions?  Sociomantic, Weka, Ebay, and so on.  (Or Microsoft 
> - did you know that D was used in their COM team?  They didn't 
> exactly send out a press release...)  A little bit, but only a 
> little in relation to their use of the language.  If you're 
> trying to accomplish something in a representative enterprise 
> context with lean resources, you don't have much time to talk 
> about what you are doing.
> 
> If you want to draw people to the language (and, honestly, I 
> wonder why it matters so much to many here - it's clearly taking 
> hold, has momentum and will continue to grow for decades; an 
> acorn will become an oak tree, and fretting about how much it's 
> grown in the past year might be missing the point, so long as 
> it's healthy enough), why not just focus on both improving the 
> language itself (pull requests, documentation) and on 
> accomplishing something useful and worth doing with it?

Go was ambitious, but grew more like a Cypress than an Oak. Many
reasons, we have mulled over this a number of time already. As I
remember the main ones:

0. Hype.
1. Timing.
2. Marketing.
3. Simple language that Google interns can use and not make errors.
4. Documentation.
5. Good standard library.
6. Ability to work with Git, Mercurial, and Bazaar repositories as
dependencies.
7. Documentation.
8. Funded support.
9. It replaces C. But then so does Rust.

Of course there were problems:

0. No way of doing complex repeatable builds.
1. The Web absorbs all.

There is a lot of proprietary as well as FOSS Go code. Ditto Java, C++.
But FOSS is where the marketing is and where many of the proprietary
decisions on programming language are made.

One thing that Go got almost right was the way of using FOSS packages
and libraries. Rust, via Cargo, did a much better job. Go has a small
standard library and allows use of any DVCS, there is no central
contributed system. Rust/Cargo has a small standard library, a central
contributed library, and the ability to use arbitrary DVCS. Rust/Cargo
wins on this hands down. I suggest Dub needs more work to be like
Cargo. This and documentation strike me as the most important things
for the evolution of D.

> Of course there are the usual trolls who don't seem to write much 
> D, but seem to be drawn like vampires to the energy of those who 
> do.  Sad.

:-)

> On Sunday, 28 January 2018 at 17:23:12 UTC, Paulo Pinto wrote:
> > This has been mentioned multiple times, D really needs some 
> > kind of killer application.
> 
> Why?

s/killer// and you get a better measure. D needs more applications out
there to act as exemplars. Currently we have Tilix and…

> It's a generalist language for getting stuff done in an age where 
> people have succumbed so much to Stockholm Syndrome that they 
> think it's a positive thing in a language that you can only use 
> it to do something special.  Yet trends in performance and 
> performance demands point to the rising importance of efficiency 
> (and I suspect there will be a return to the recognition of the 
> importance of being a generalist - in programming, as in other 
> fields).  There was a tweet by the author of Musl libc observing 
> that software today runs slower than software twenty years ago, 
> and linking the bloat to the insane pressure to maximise CPU 
> performance over all else.  The era of that kind of ruthless 
> optimization is over because it's not the only thing that 
> matters, and we start to see the price of it.  And generalism - 
> in a dynamic business environment, there's considerable value to 
> have capabilities that aren't adapted to particular narrow skills 
> when what you need is always changing and may be unknown even to 
> you.

I am less convinced by this argument. Go, Rust, and especially Java
have shown the power of tribalism and belonging to the one true tribe
eschewing all others. Java is a superb example of this: the JVM is now
a polyglot platform, and yet Java programmers, especially enterprise
ones, will only contemplate using Java and refuse to be educated about
Kotlin, Ceylon, Groovy, JRuby, etc. However when a feature inspired
(many years later) by the innovations in other JVM-based languages gets
shoehorned into Java then, eventually, the Java folk are prepared,
reluctantly, to learn about it. And maybe a few years later actually
use it. 

> My generation was privileged because very quickly if you wanted 
> to get anything interesting done you had to learn assembly 
> language (maybe write your own assembler or disassembler), had to 
> learn a bit about hardware, and could never pretend the CPU was 
> this perfect platonic abstraction.  And for a while that changed, 
> but I think the past is returning again, as it often does.

FORTRAN, the one true way. Who needs the intricacies of assembly
language when you have FORTRAN. :-)

> So I see a value in hiring hacker / generalist types who can 
> figure things out.  For example:
> 

All companies should. Expertise in programming is not length of time on
one language, but the number of different computational models you can
work with to a reasonable quality of code.

Obviously familiarity with a set of needed API can be a help to speed
of getting a new person productive in a given project, but that is
learnable.


> Back in 2007, most finance types would have said how completely 
> impracticable and unreasonable.  But I say, with GK Chesterton, 
> that "all progress depends on the unreasonable man".  And someone 
> like that doesn't succumb to helplessness once they are outside 
> of their shiny IDE, knows that in the end everything is just 
> code, and you can change it if you want to, and there is 
> strategic value from building organisational capabilities from 
> hiring such people.  Usually I'm a couple of years ahead, and I 
> think others will follow.  If you hold a contrarian point of 
> view, you know you're right when surprises start coming in your 
> direction, and people still can't see it.  And I think that's 
> been the case since 2014.
> 
> Anyway - so D is a general purpose language, and I think we are 
> likely seeing a nascent return in recognizing the value of 
> generalist tools and people.

It never really went away, we just had a period of PHP, JavaScript,
HTML, and CSS being all there was as far as people with money to fund
stuff were concerned.
 
> > On my line of work having Go on the skills list is slowly 
> > becoming a requirement, due to Docker and Kubernetes adoption 
> > on cloud infrastructures.
> 
> That's great.  Walter says that good code should look good on the 
> page, and Go code looks nice enough.  It's got nice network and 
> infra libraries, as you say.  But why would the adoption of Go be 
> bad for D?  I think it's great for D because after a period of 
> stagnation it gets people open to newer languages, and on the 
> other hand the gap between the spirit of Go and D isn't that far 
> (GC, clean code, native target) even if they don't have generics. 
>   It's a big world - both D and Go can succeed, and the success of 
> one isn't bought at the cost of the other.

And Go has channels and task pools. 

Whilst David Simcha did a lot of work on tasks and task pools for
std.parallelism, it only really addresses data parallelism. I think
there needs to be some work within std.concurrency and std.parallelism
to have a single task structure and "thread pool" system. That Go had
this from the outset is a Good Thing™. Go does this at the language
level. D does it at the standard library level. Rust has gone one
further and has a low-level thread and channel system in the language
and standard library, everything else is in a separate library. This
raises the question of whether Phobos should be broken up.

Python originally had the "batteries included" view of it's standard
library, and continues to suffer problem because of trying to retain
this. PyPI and pip have removed the need for much of the standard
library. I am beginning to think that D needs a smaller standard
library (Phobos) and a better package system (Dub) taking lessons from
Rust and Cargo.

> > Just wondering if mir or easier GPGPU programming could be that 
> > killer application.
> 
> We sponsor mir algorithm (some of the routines within were 
> developed for us, and we were happy to open source them), and we 
> are rewriting our core analytics - used across the firm in a 
> $4.1bn hedge fund in D from C++ before that.  What alternative 
> really exists for what we are doing there?  And C++ vs D, it's 
> not even a fair fight if you care about productivity, plasticity 
> of the code, and generating wrappers for other languages that you 
> can still understand whilst maintaining decent performance.  At 
> the same time, we're not a D shop - a diversity of languages is 
> not a bad thing, provided you have some way for them to work 
> together.  Code reuse is very difficult, but the UNIX way does 
> work.  On the other hand, if you want to connect components, how 
> are you to do that?  Well, D is pretty nice for writing DSLs that 
> can connect to code written in other languages, and where 
> expressions can be evaluated from other languages.
> 
> A specialist language adapted to a particular domain or set of 
> domains - yes, that benefits from a killer app.  But for a 
> generalist language that's useful for getting stuff done - why 
> would there be a single killer app?  That doesn't make sense to 
> me.  There should be multiple successes across different domains, 
> and that's what we are beginning to see.  Just bear in mind that 
> the web and tech guys talk a lot, but most programmers don't work 
> in those industries.  It would be a mistake to conflate salience 
> with economic importance, I think.

I agree we do not need "a killer app", but we do "exemplar apps", Tilix
is one, but more are needed. Also better documentation. I still find
the documentation for Phobos impenetrable most of the time. And the
documentation for GtkD and GStreamerD tells us about the C API not the
D binding. It clearly needs effort over and above what is being
provided, and there is no resource, so in all likelihood nothing will
change.


-- 
Russel.
===========================================
Dr Russel Winder      t: +44 20 7585 2200
41 Buckmaster Road    m: +44 7770 465 077
London SW11 1EN, UK   w: www.russel.org.uk
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 833 bytes
Desc: This is a digitally signed message part
URL: <http://lists.puremagic.com/pipermail/digitalmars-d/attachments/20180129/d33de550/attachment.sig>