When starting a new Haskell project which needs language features that are offered by both new and older extensions, which should I use? The new one or the old one? For example:
GADTs supersede ExistentialQuantification.
TypeFamilies in many cases supersede FunctionalDependencies.
etc. (if you have more examples feel free to add here)
I incline towards the newer ones as they feel cleaner and they add more features to the language in a unified manner. However, the older ones are supported by more compilers, they're used in existing libraries and perhaps also better known among programmers.
Update: As mentioned in answers/comments, the "supersede" relationships above are not completely precise. However, I'm interested in (often occurring) cases when a task can be accomplished equally well by several extensions.
"supersede" is very ambiguous here.
For language features that have semantics that overlap with other features, we can recognize a few cases:
some features generalize existing ones
some offer alternative approaches
In the case of more general features, there are risks and benefits. The generality can impact many other parts of the language, or be restricted. In the latter, choose the general case.
If it is a choice between alternative approaches, with no strict ordering, such as FDs or ATs, then it is harder to give general advice.
So, in summary, "no" , there's no obvious reason to use "new" things over "old" things, when it comes to language features. They don't sort that way.
In short, yes, and for the reasons you state: cleaner and clearer.
Occasionally I've come across places where there's a class better expressed as a multi-parameter type class with functional dependency because you can get more parametricity more simply that way, but usually an associated type family is a clearer and more flexible way of epressing it. Type families are much more intuitive to a functional programmer than functional dependencies.
GADTs have much cleaner, clearer syntax than existential quantification, with fewer conceptual hurdles, and it's easy to get the behaviour you want.
I'd recommend using the new stuff if you can, and the old ways if necessary. These lovely newer extensions could do with a higher profile.
I'm trying to learn how various aspects of GHC, like type inference, pattern matching, and other code transformations, are implemented.
However the codebase is fairly large and the file names use a lot of acronyms (simpl, stg, stranal...). What do these mean, and how is the code organized?
As hammar says, the GHC commentary is probably the best place to start for learning about GHC itself. This does assume some prior knowledge of compilers in general, but if your primary interest is in modifying GHC you can probably get by with just a basic CS background.
If you're interested more generally in the principles behind GHC, e.g. if you want to learn how to write your own compiler, you'd probably be better served by perusing Simon Peyton-Jones' myriad publications on relevant topics, including an entire book on implementing functional languages and a "tutorial" book that goes through the implementation of a non-strict functional language.
The GHC commentary is a good place to start.
I currently program in Perl, Python, C#, C, C++, Java, and a few other languages, and I'm looking for a new language to use as a primary when doing personal projects.
My current criteria are:
can be run as an interpreted language (i.e., run without having to wait to compile it);
can be compiled to native code;
are strongly typed (even if optionally);
support macros/templating/code morphing/wtf you want to call it;
has a decent number of libraries for it, or easily accessible to it;
Ideas? Suggestions?
I would suggest that Haskell suits your criteria.
Can be run as an interpreted language? Yes, via GHCI.
Can be compiled to native code? Yes.
Is strongly typed? Very much so. Perhaps even the most strongly typed language today, with the exception of some theorem provers like Agda.
Support macros/templating/morphing? If you use template haskell. This is an optional extension of the language however, so most libraries don't use macros. I haven't used template haskell myself so i can't comment on if it's any good.
Has decent library support? The standard library is not bad. There is also Hackage, an open repository of Haskell libraries a bit in the style of CPAN.
Additionally, it sounds like you already know a lot of imperative/object oriented languages. IMHO if you learn another one of those langs. it will probably be a slightly different permutation of features you've already seen somewhere else. Adding another programming paradigm like functional programming to your toolbox will probably be a better learning experience. Though I guess whether that's an advantage or not depends on if you want to learn new things or be productive quickly.
Common Lisp fits: there is an optional typing, efficient native compilation is available, powerful REPL makes it a perfect choice for scripting, and there is a powerful macro metaprogramming.
OCaml fits as well, with CamlP4 for metaprogramming.
Scala? It does run scripts, although they are compiled (transparently) first. I'm not sure what you mean by code morphing etc, but it's pretty good for DSLs. It meets all your other requirements - compiled as much as Java is, strongly typed, and has a reasonable number of its own libraries as well as all of Java's. I'm still a beginner with it, but I like it so far.
Let's assume for the moment that C++ is not a functional programming language. If you want to write a compiler using LLVM for the back-end, and you want to use a functional programming language and its bindings to LLVM to do your work, you have two choices as far as I know: Objective Caml and Haskell. If there are others, then I'd like to know about those too.
I'm not asking for subjective opinions, so please don't give this the subjective tag. I want to make up my own mind about this, but I'm not sure I know what are all the trade-offs. So, StackOverflow to the rescue. What are the trade-offs?
Either OCaml or Haskell would be a good choice. Why not check out the LLVM tutorials for each language? The LLVM tutorial for OCaml is here: http://llvm.org/docs/tutorial/OCamlLangImpl1.html
Haskell has more momentum these days, but there are plenty of good parsing libraries for OCaml as well including the PEG parser generator Aurochs, Menhir, and the GLR parser generator Dypgen. Also check out this presentation on pcl a monadic parser combinator library for OCaml (like Parsec for Haskell) there's some good info in there comparing Haskell's and OCaml's approach: http://osp.janestreet.com/files/pcl.pdf
Some will say that laziness gives Haskell the edge in parsing, but you can get laziness in OCaml as well.
Haskell has higher level bindings to LLVM than OCaml (the Haskell ones provide some interesting type safety guarantees) and Haskell has by far more libraries to use (1700 packages on http://hackage.haskell.org) making it easier to glue together components.
Availability of native bindings need not constrain your choice of language. There is a third option, apart from using bindings or generating IR text directly:
You can use a language-neutral serialization format, such as Google's Protocol Buffers, to serve as the bridge from your front-end to your back-end. Protocol buffers are, after all, just ASTs in disguise.
Your front end, implemented in a functional language, then does what it is best at -- parsing, type checking, desugaring, core-to-core transformations, etc -- and the C++ backend takes the IR from your frontend and uses LLVM's feature-complete-by-definition native C++ API to do lowering from your-language-IR to LLVM IR. This makes it much easier to handle "advanced" features of LLVM such as debug metadata.
I'm using this strategy with hprotoc and associated Haskell bindings for protocol buffers, and am very happy with the results. There is much to be said for using the right tool for the job!
OCaml is the only functional language with bindings in the LLVM distro itself and documentation on llvm.org such as the Kaleidoscope tutorial. If you have OCaml installed when you build and install LLVM then it will automatically build and install the LLVM bindings for OCaml as well. Moreover, these OCaml bindings have been in use for years so they are mature and reliable.
I have been developing HLVM in OCaml using the standard LLVM bindings and found OCaml+LLVM to be an extremely powerful combination. HLVM provides tuples, arrays, unions, TCO of all tail calls, generic printing, FFI to C, JIT compilation and parallel garbage collection with a VM weighing in at under 2kLOC of OCaml code that took only a few man-weeks to develop from scratch. HLVM's numerical performance already far exceeds that of today's fastest open source FPLs including OCaml itself. I have published articles in the OCaml Journal describing how LLVM can be used from OCaml for everything from basic expression evaluation to advanced topics such as parallelism and garbage collection. You may also like this mini example.
I need to get around to writing that programming language I've been meaning to write. How do you kids do it these days? I've been out of the loop for over a decade; are you doing it any differently now than we did back in the pre-internet, pre-windows days? You know, back when "real" coders coded in C, used the command line, and quibbled over which shell was superior?
Just to clarify, I mean, not how do you DESIGN a language (that I can figure out fairly easily) but how do you build the compiler and standard libraries and so forth? What tools do you kids use these days?
One consideration that's new since the punched card era is the existence of virtual machines already bountifully provided with "standard libraries." Targeting the JVM or the .NET CLR instead of ye olde "language walled garden" saves you a lot of bootstrapping. If you're creating a compiled language, you may also find Java byte code or MSIL an easier compile target than machine code (of course, if you're in this for the fun of creating a tight optimising compiler then you'll see this as a bug rather than a feature).
On the negative side, the idioms of the JVM or CLR may not be what you want for your language. So you may still end up building "standard libraries" just to provide idiomatic interfaces over the platform facility. (An example is that every languages and its dog seems to provide its own method for writing to the console, rather than leaving users to manually call System.out.println or Console.WriteLine.) Nevertheless, it enables an incremental development of the idiomatic libraries, and means that the more obscure libraries for which you never get round to building idiomatic interfaces are still accessible even if in an ugly way.
If you're considering an interpreted language, .NET also has support for efficient interpretation via the Dynamic Language Runtime (DLR). (I don't know if there's an equivalent for the JVM.) This should help free you up to focus on the language design without having to worry so much about the optimisation of the interpreter.
I've written two compilers now in Haskell for small domain-specific languages, and have found it to be an incredibly productive experience. The parsec library makes playing with syntax easy, and interpreters are very simple to write over a Haskell data structure. There is a description of writing a Lisp interpreter in Haskell that I found helpful.
If you are interested in a high-performance backend, I recommend LLVM. It has a concise and elegant byte-code and the best x86/amd64 generating backend you can find. There is an optional garbage collector, and some experimental backends that target the JVM and CLR.
You can write a compiler in any language that produces LLVM bytecode. If you are adventurous enough to learn Haskell but want LLVM, there are a set of Haskell-LLVM bindings.
What has changed considerably but hasn't been mentioned yet is IDE support and interoperability:
Nowadays we pretty much expect Intellisense, step-by-step execution and state inspection "right in the editor window", new types that tell the debugger how to treat them and rather helpful diagnostic messages. The old "compile .x -> .y" executable is not enough to create a language anymore. The environment is nothing to focus on first, but affects willingness to adopt.
Also, libraries have become much more powerful, noone wants to implement all that in yet another language. Try to borrow, make it easy to call existing code, and make it easy to be called by other code.
Targeting a VM - as itowlson suggested - is probably a good way to get started. If that turns out a problem, it can still be replaced by native compilers.
I'm pretty sure you do what's always been done.
Write some code, and show your results to the world.
As compared to the olden times, there are some tools to make your job easier though. Might I suggest ANTLR for parsing your language grammar?
Speaking as someone who just built a very simple assembly like language and interpreter, I'd start out with the .NET framework or similar. Nothing can beat the powerful syntax of C# + the backing of the entire .NET community when attempting to write most things. From here i designed a simple bytecode format and assembly syntax and proceeeded to write my interpreter + assembler.
Like i said, it was a very simple language.
You should not accept wimpy solutions like using the latest tools. You should bootstrap the language by writing a minimal compiler in Visual Basic for Applications or a similar language, then write all the compilation tools in your new language and then self-compile it using only the language itself.
Also, what is the proposed name of the language?
I think recently there have not been languages with ALL CAPITAL LETTER names like COBOL and FORTRAN, so I hope you will call it something like MIKELANG with all capital letters.
Not so much an implementation but a design decision which effects implementation - if you make every statement of your language have a unique parse tree without context, you'll get something that it's easy to hand-code a parser, and that doesn't require large amounts of work to provide syntax highlighting for. Similarly simple things like using a different symbol for module namespaces and object namespaces ( unlike Java which uses . for both package and class namespaces ) means you can parse the code without loading every module that it refers to.
Standard libraries - include the equivalent of everything in C99 standard libraries other than setjmp. Add whatever else you need for your domain. Work out an easy way to do this, either something like SWIG or an in-line FFI such as Ruby's [can't remember module name] and Python's ctypes.
Building as much of the language in the language is an option, but projects which start out doing either give up (rubinius moved to using C++ for parts of its standard library), or is only for research purposes (Mozilla Narcissus)
I am actually a kid, haha. I've never written an actual compiler before or designed a language, but I have finished The Red Dragon Book, so I suppose I have somewhat of an idea (I hope).
It would depend firstly on the grammar. If it's LR or LALR I suppose tools like Bison/Flex would work well. If it's more LL, I'd use Spirit, which is a component of Boost. It allows you to write the language's grammar in C++ in an EBNF-like syntax, so no muddling around with code generators; the C++ compiler compiles the grammar for you. If any of these fail, I'd write an EBNF grammar on paper, and then proceed to do some heavy recursive descent parsing, which seems to work; if C++ can be parsed pretty well using RDP (as GCC does it), then I suppose with enough unit tests and patience you could write entire compilers using RDP.
Once I have a parser running and some sort of intermediate representation, it then depends on how it runs. If it's some bytecode or native code compiler, I'll use LLVM or libJIT to process it. LLVM is more suited for general compilation, but I like the libJIT API and documentation better. Alternatively, if I'm really lazy, I'll generate C code and let GCC do the actual compilation. Another alternative, is to target an existing VM, like Parrot or the JVM or the CLR. Parrot is the VM being designed for Perl. If it's just an interpreter, I'll walk the syntax tree.
A radical alternative is to use Prolog, which has syntax features which remarkably simulate EBNF. I have no experience with it though, and if I am not wrong (which I am almost certainly going to be), Prolog would be quite slow if used to parse heavy duty programming languages with a lot of syntactical constructs and quirks (read: C++ and Perl).
All this I'll do in C++, if only because I am more used to writing in it than C. I'd stay away from Java/Python or anything of that sort for the actual production code (writing compilers in C/C++ help to make it portable), but I could see myself using them as a prototyping language, especially Python, which I am partial towards. Of course, I've never actually done any of this before, so I'm not one to say.
On lambda-the-ultimate there's a link to Create Your Own Programming Language by Marc-André Cournoyer, which appears to describe how to leverage some modern tools for creating little languages.
Just to clarify, I mean, not how do you DESIGN a language (that I can figure out fairly easily)
Just a hint: Look at some quite different languages first, before designing a new languge (i.e. languages with a very different evaluation strategy). Haskell and Oz come to mind. Though you should also know Prolog and Scheme. A year ago I also was like "hey, let's design a language that behaves exactly as I want", but fortunatly I looked at those other languages first (or you could also say unfortunatly, because now I don't know how I want a language to behave anymore...).
Before you start creating a language you should read this:
Hanspeter Moessenboeck, The Art of Niklaus Wirth
ftp://ftp.ssw.uni-linz.ac.at/pub/Papers/Moe00b.pdf
There's a big shortcut to implementing a language that I don't see in the other answers here. If you use one of Lukasiewicz's "unparenthesized" forms (ie. Forward Polish or Reverse Polish) you don't need a parser at all! With reverse polish, the dependencies go right-to-left so you simply execute each token as it's scanned. With forward polish, it's the reverse of that, so you actually execute the program "backwards", simplifying subexpressions until reaching the starting token.
To understand why this works, you should investigate the 3 primary tree-traversal algorithms: pre-order, in-order, post-order. These three traversals are the inverse of the parsing task that a language reader (i. parser) has to perform. Only the in-order notation "requires" a recursive decent to re-construct the expression tree. With the other two, you can get away with just a stack.
This may require more "thinking' and less "implementing".
BTW, if you've already found an answer (this question is a year old), you can post that and accept it.
Real coders still code in C. Just that it's a litte sharper.
Hmmm... language design? or writing a compiler?
If you want to write a compiler, you'd use Flex + Bison. (google)
Not an easy answer, but..
You essentially want to define a set of rules written in text (tokens) and then some parser that checks these rules and assembles them into fragments.
http://www.mactech.com/articles/mactech/Vol.16/16.07/UsingFlexandBison/
People can spend years on this, The above article talks about using two tools (Flex and Bison) That can be used to turn text into code you can feed to a compiler.
First I spent a year or so to actually think how the language should look like. At the same time I helped in developing Ioke (www.ioke.org) to learn language internals.
I have chosen Objective-C as implementation platform as it's fast (enough), simple and rich language. It also provides test framework so agile approach is a go. It also has a rich standard library I can build upon.
Since my language is simple on syntactic level (no keywords, only literals, operators and messages) I could go with Ragel (http://www.complang.org/ragel/) for building scanner. It's fast as hell and simple to use.
Now I have a working object model, scanner and simple operator shuffling plus standard library bootstrap code. I can even run a simple programs - as long as they fit in one file that is :)
Of course older techniques are still common (e.g. using Flex and Bison) many newer language implementations combine the lexing and parsing phase, by using a parser based on a parsing expression grammar (PEG). This works for recursive descent parsers created using combinators, or memoizing Packrat parsers. Many compilers are built using the Antlr framework also.
Use bison/flex which is the gnu version of yacc/lex. This book is extremely helpful.
The reason to use bison is it catches any conflicts in the language. I used it and it made my life many years easier (ok so i'm on my 2nd year but the first 6months was a few years ago writing it in C++ and the parsing/conflicts/results were terrible! :(.)
If you want to write a compiler obviously you need to read the Dragon Book ;)
Here is another good book that I have just read. It is practical and easier to understand than the Dragon Book:
http://www.amazon.co.uk/s/ref=nb_sb_noss?url=search-alias%3Daps&field-keywords=language+implementation+patterns&x=0&y=0
Mike --
If you're interested in an efficient native-code-generating compiler for Windows so you can get your bearings -- without wading through all the unnecessary widgets, gadgets, and other nonsense that clutter today's machines -- I recommend the Osmosian Order's Plain English development system. It includes a unique interface, a simplified file manager, a friendly text editor, a handy hexadecimal dumper, the compiler/linker (of course), and a wysiwyg page-layout application for documentation. Written entirely in Plain English, it is a quick download (less than a megabyte), small enough to understand in short order (about 25,000 lines of Plain English code, with just 4,000 in the compiler/linker), yet powerful enough to reproduce itself on a bottom-of-the-line Dell in less than three seconds. Really: three seconds. And it's free to all who write and ask for a copy, including the source code and and a rather humorous tongue-in-cheek 100-page manual. See www.osmosian.com for details on how to get a copy, or write to me directly with questions or comments: Gerry.Rzeppa#pobox.com