But you are right, my example is a little bit irrelevant. SAP would be better one.
Seriously, nowadays tools are so good, that you could really write in any popular language you like (in fact - in a few at once, depending of which one fits better for a sub-task).
The internet services, for some reasons, are not (mostly not) implemented in Java. This site, for example, is powered by author's very own dialect of Lisp. Slashdot has been written in Perl, Reddit is mostly Python.
The real problem is still software complexity and lack of programming/engineering (not coding) skills. Problem is with modeling and [de]composition, not with coding. Coders are cheap.
The business of [outsourced] Java sweatshops of unskilled coders "getting shit done" by quickly picking up methods out of IDE's context help without any understanding of underlying principles is like a whole sector of economy. Products they made in the most cases are exactly what they themselves call it - "shit done in exchange of salary".
But this subject is irrelevant here.
They also do some Common Lisp on the airline search division.
btw, Common Lisp is a "strong typed" language, and many large enough code-bases were here. Lisp machine ecosystem is a good example.
A lot of banking software has been written in 4GL and stored procedures (SQL). When Oracle stuffed its products with Java it was not necessarily an improvement, rather things became more difficult and messy than before.
I'm not sure how widely Go is being chosen (over say Java) for new projects inside Google. Maybe some Googler could clue us in?
> btw, Common Lisp is a "strong typed" language,
Can you point me to a code example? The ones I did find didn't have any type information in them.
> A lot of banking software has been written in 4GL and stored procedures (SQL). When Oracle stuffed its products with Java it was not necessarily an improvement, rather things became more difficult and messy than before.
Banking was just an example, and I'm sure they are plenty horror stories there involving Java (esp. when used in some awkward way). What I wanted to say is that IMO the effort to build and maintain large code base in for example Python would typically be much greater than in Java. For small code bases, I'm not sure who wins - the development phase will obviously be faster in Python (which is the only thing that matters for startups - hence python/ruby/... popularity here), but maintenance will still be a pain - after the initial devs leave, hires will need to learn that Python codebase, and lack of type information will make it harder.
* (type-of 1)
BIT
* (type-of 123)
(INTEGER 0 4611686018427387903)
* (type-of "hi")
(SIMPLE-ARRAY CHARACTER (2))
* (type-of 'hi)
SYMBOL
*
And there is also typecase macro which is basically cond on type-of - the way to do dispatch on a type.This technique of extending a language with new special forms using macros is the essence of Lisps.
What Lisp is lacking though is type declarations for variables, method parameters and returned values. These make codebases immensely more readable.
A usual type declaration looks like this:
(defun month-name (month)
(declare (type (integer 1 12) month))
(the string
(ecase month
(1 "Jan") (2 "Feb") (3 "Mar") (4 "Apr") (5 "May") (6 "Jun")
(7 "Jul") (8 "Aug") (9 "Sep") (10 "Oct") (11 "Nov") (12 "Dez"))))
You can also declare the type of the function, outside of the function: (declaim (ftype (function ((integer 1 12)) ; the parameter list
string) ; the return value
month-name)) ; the function name
CLOS (the Common Lisp Object System) was added a few years later. CL-USER 1 > (defmethod add ((a integer) (s string))
(+ a (parse-integer s)))
#<STANDARD-METHOD ADD NIL (INTEGER STRING) 402005B1C3>
CL-USER 2 > (add 12 "23")
35
What do you think the words 'integer' and 'string" are about?Quite widely, for new projects.
In that regard it has been fantastically successful.
In my opinion, Java's "success" is a like a "religion". It is relatively easy to make money on "believers".
Nevertheless, facts are that in telecoms we have "success" of Erlang, in Android we have "success" of NDK, on server side there is everything mixed, and if someone would collect statistics of failed projects and project went hugely over time and budgets Java would be on the first place. And here, in my opinion, it would be a more of causation, not mere correlation.
Let's stop here.
> In that regard it has been fantastically successful.
It allows median programmers to work for years on large systems, yes, and usually to maintain them as well. That indeed has been my experience.
However, in my experience, it is not those median programmers that actually built those systems, and wouldn't have been able to. Much better skilled individuals did.
In my experience, large systems built by mostly median programmers tend to fail (or never even reach production status), regardless of the implementation language. Which is to be expected, since system complexity is often inherent. However, it seems like the idea that "all programmers are just replaceable cogs" and thus that's not the case is way more prevalent around Java (and to some extent .net) than around other ecosystems.
Java has been fantastically successful in the sense of keeping armies of programmers employed maintaining large systems, and at keeping those systems running. Personally, I suspect that most of those cases could have been fantastically more successful if they didn't use the "replaceable cog" model that for many projects is Java's core feature - but of course I cannot offer any proof.