Why I like D

[Bill Baxter]

Sounds pretty close to my thoughts on the subject.

A couple of years ago I decided that I was really tired of fighting
with with C++ (mostly the memory micro-management, the horrible
template mess, and totally unreadable code in standard libs like Boost
and STL).  So I decided to find something easier and more fun to use.
I decided that I would go with 1 dynamic language and 1 static
compiled language.   After looking around I decided on Python for the
dynamic language, (primarily because of the active NumPy/SciPy
community), and for the static language I chose OCaml because it
sounded like a nice hybrid of functional with OOP, and appeared to be
used by various people who care about having their numerical code run
fast.

But then I sat down to try to figure out how to write a simple OpenGL
program in OCaml.  Hmm.  Couldn't find anything about it.  From a
quick google just now, I see someone made some GL bindings late in
2007, but that was too late for me.  Anyway it wasn't just the lack of
GL bindings.  It was sitting down and realizing that with OCaml I was
basically going to have to re-learn how to do everything I already
knew how to do, but using Monads or whatever.

Python didn't put any such roadblocks in my way so I got started using
it right away with no problems.  No problems, that is, until I started
trying to write some very loop-intensive mesh processing code.  And it
was just soooooo slooooooow.  And I began spending lots of my time
trying to figure out how to turn simple nested loops into one-line
vector expressions just so that Numpy would run them at reasonable
speed.  And I also got increasingly annoyed by the silly runtime
errors that any decent compiler would tell me about.    I got a Python
IDE which I liked, but I realized that the startup time for running
python programs in that IDE was about as long as compile times for
equivalently sized programs in C++, so where's this supposed advantage
from not having to compile?

The same time I had been eyeing D, and occasionally checking out the
webnews but I had the concept that it was hard to integrate with other
code because I would need special versions of all my libraries (DMD
COFF libs vs the MS libs I have lying around already).  That put me
off a bit.   But as I got into Python more I started learning how to
write my own native Python modules, and started getting the hang of
it.  At some point I realized, hey, getting my C code to work with D
is kind of like writing a native Python module, but a *whole lot
easier*!  All I have to do is re-write function signatures, instead of
doing a bunch of funky PyTuple_FromArgs() PyIncRef() nonsense!  Funny
how perspective works.  From a C++ coders perspective the hoops D was
going to make me go through looked very annoying, but with Python
goggles on, talking to C code from D looks like a walk in the park.

So I decided to give D a serious test drive.   And it's turned out
very well.  D is a lot of fun to develop with.

Even when I think about the downsides of D -- having to write or port
a lot of my own libraries, working around compiler bugs, not having a
super-duper GUI debugger (and when I started no real debugger at all)
-- all these things are much less annoying than trying to debug some
intricate memory allocation issue in C++ or trying to figure out why
some complex template lib like boost::serialize is not working.  So
maybe overall I don't develop much faster, but I sure have a lot more
fun doing it.  Especially the first one -- having to write or port
your own libs -- it may be a lot of extra work you wouldn't have to do
in C++, but it's kinda fun, and when you're done you know your lib
inside and out and know you can fix any problem that arises.  Heck,
even with C++ many developers will rewrite even when they could reuse
just because they prefer that level of ownership and total control.

Now I just wish the BigMegaCorps of the world would take notice and
write us some nice VisualStudio-caliber integrated gui IDE/debuggers.

--bb

On Fri, Dec 5, 2008 at 4:20 AM, bearophile <bearophileHUGS at lycos.com> wrote:
> Few moths ago someone here has implicitaly asked why I like D, now I can answer. Feel free to skip this boring wall of text.
>
> I don't like the syntax and rules of C++, to me they look ugly, confusing and too much complex, and when I can I avoid this language. It makes me waste too much time, and I don't like to learn all its special cases.
>
> I like Python because it's small enough to fit into my limited brain (few parts of D seem to not fit in my brain well, and I may suggest to improve them to make them more intuitive and simpler), because it's designed for the programmer and not for the CPU, because it helps me avoid many kinds of bugs, because its syntax is clean, readable and easy to remember. It also has tons of std lib and third-part modules, that allow me to plot things, process images and sounds, create GUIs, and lots of other things.
>
> But for some of my purposes Python is too much slow, and I have not appreciated all the tons of solutions used by Python programmers to solve this problem. I have even given a big hand to help the development of a Python compiler (ShedSkin) but I now think it's a failed experiment. The only solution I like to use is Psyco, but often it doesn't lead to fast enough programs.
>
> In some of my programs I need more than just speed: my problem may require lot of memory (and even if it's not a lot of memory, compacting the memory usage usually leads to better performance), and genrally I just want the freedom of managing memory as I like, so I need something like structs, unions, pointets and all that "unsafe" stuff. The D GC allocates blocks of memory as powers of 2 (under a certain size) and I don't like that much, but there's the std.c.stdlib.malloc too anyway, for special situations.
>
> Speed coming to being closer to the metal and freedom of memory usage allow me to solve problems that you can't solve in a acceptable time with Python. For a silly recent example, it has allowed me to compute the next number of this sequence, with a program 40 times faster than the original Python one:
> http://www.research.att.com/~njas/sequences/A119770
> Of course you can wait a day to receive the answer from the Python program, but if the problem space gets ten times bigger I can still use the D program...
> Beside this one, I can offer several others less silly examples where the speed of the D program was essential, or where for example I have used almost 2 GB of RAM in an effective way (where using a dynamic language was hopeless), in bioinformatics problems, graph-related problems, etc.
>
> Of course I can use Delphi, C, C++ or CommonLisp to write such programs, but I have seen that I am much slower in writing C code, compared to D code, even if I am using C for much more time. D helps me avoid problems and bugs, and just the built-in arrays are a huge time saver, expecially when you want to manage 2D arrays (so D allows me to write low-level code only where I want, to speed up hotspots and not to write all the program in low-level style as C programs generally require). Delphi language is much less modern and offers me less nice things.
>
> Using my dlibs I need only a short time and almost no pain to translate Python code to D, or to write high-level D code in the first place (translating it to C/C++ is surely much more painful, even if someone write something as my libs for C++), but with D I can also write in a style very close to C, this is useful to translate slowly higher level code to actual C programs (so I use D as an intermediate language progressively closer to C or even to asm). I can also inline asm code, and learn more assembly, more things about the CPU and write fast SSE code.
>
> So I like D because it's a multi-level language (vaguely as C++), because it helps me avoid lot of bugs I put into C programs, while being still simple enough for my limited mind to learn and manage, unlike C++.
>
> I have also appreciatd D because it has given me the chance to re-learn several things of computer science and how a low-level language works. There are several things I don't know still that are often discussed in this newsgroup, so I have more to learn.
>
> Using D sometimes gives me the same fun people receive from solving puzzles: the complex things I have used for example to create the Record struct of my dlibs (that mimics a little the tuples of Python) has required me about one year of work, and it was better than a complex puzzle. Now for example such Record support opCmp and hashing, even if they recursively contain vanilla structs :-)
>
> Haskell language will surely teach me several other things that D (and other languages I know) doesn't know about. But even very common languages like Java offer a lot of things to learn that are not much present in the D programming.
>
> D currently has tons of problems and weak spots, first of all the way it is developed, and I really hope to see it fix some of such problems, but for my purposes it's better than most of the alternatives.
>
> Bye,
> bearophile
>