felise

felise is a programming language occupying a space between concatenative and procedural programming (with, like many modern languages, some influence from functional programming). It is a hobby language designed as a successor to nimf (an earlier programming language by the same author). The language is implemented as a basic AST walking interpreter (though a compiler, JIT, or bytecode/vm would also be possible) in golang. It is not dynamically typed and all variables need to be declared as a given type (even control structures like if and while are technically types as well). Type checks are generally handled at runtime, with a few such checks handled in the lex/parse phases of runtime (rather than as encountered in running code).

As a new language there is no idiomatic felise style at present and anything goes. Deliberate effort has gone into bridging some gaps between the stack based concatenative style and a procedural/imperative style. The later is represented by a heavy focus on variable use, lexical scoping/nested procedures forming closures.

This README has gotten a little messy with lots of random information. TODO: clean it up.

Example

Since things are still in a major early development phase I figure an example of the code and a metric will suffice for the readme for now.

A file, fibtest.fe:

proc fib
  | calculates the nth fibonacci number
    where n is an INT on TOS at call time |

  INT var! fib# set! fib#
  INT var! a
  INT var! b 1 set! b
  INT var! s

  proc continue-loop? 
    fib# 0 >
  end

  continue-loop?
  while
    a b + set! s
    b     set! a
    s     set! b
    fib# 1 - set! fib#
    continue-loop?
  end

  a
end

# run it!
sys-args length 0 >

if
  sys-args 1 <- INT cast fib println
else
  "No input" stderr file-write
end

end
^
This last `end` closes the global scope, so anything
below it is unread by the interpreter. As such, this
is basically a comment. `end` always closes the current
scope for any construct that creates a scope (while,
proc, if, dowhile, proc!, try). The global scope can
be closed too, even though it doesn't have an explicit
start tag. You can use `.` instead of `end` if you
prefer.

Running time felise ./fibtest.fe 40 on my Linux system outputs:

102334155

real    0m0.010s
user    0m0.001s
sys     0m0.012s

Pretty snappy for this initial test as I am building things. Granted, it is iterative and not recursive (which would certainly slow down).

Available types in the language are STRING, INT, BOOL, FLOAT, PROC, TYPE, and LIST. There a few other types available to the interpreter, but that a user would not use themselves. The LIST type is the core data structure of the language, beyond the stack, and a LIST can hold a mix of any of the above types, including LIST items. As such, this creates a strange blend of typed variables and untyped data structures. Still working out the ramifications of this decision...

Building

Assuming you have the mainline go compiler and git installed (and are using a posix-ish shell on a non-windows system):

git clone https://tildegit.org/sloum/felise && cd felise && go install

You should then be able to run felise to load the repl (assuming the path that go install installs to is on your path -- you can always use go build to build in the local directory instead). As things develop a Makefile will likely be added, but right now it just isnt necessary.

Libraries

Libraries, written in felise, are baked into the executable at present. This makes it easier to work things out than depending on a file-system based approach. The available libraries are as follows:

1. std (loaded by the repl automatically)
2. math
3. paths
4. stack
5. strings

To load a library follow this pattern:

"strings" import

You could also do something like the following to load arbitrary code from another file:

"~/my-code/felise-code.fe" import

Features

• Types
  • BOOL
  • FLOAT
  • INT
  • LIST
  • STRING
  • TYPE
• Blocks
  • Blocks create their own scope as a child to the scope they were created in
  • All block types end with end, which always closes out the current scope (including the main scope)
  • Block types:
    • Loops
      • while
      • dowhile
    • Procedures
      • proc
      • proc!
        • User defined forward reading procedures (proc!) enable things like writing procedures that act on variables directly, like how var! and set! do
    • Conditionals
      • if
      • else
    • Error handling
      • try
      • catch (often used with throw)
• Lexical scoping with explicit scope jumping for variable creation/setting and closures
• A solid base library of built-in procedures developed in golang directly, providing the language primitives and base features to build on without sacrificing speed
• A list type that can contain any other type, including lists themselves
  • felise lists are 1 indexed
  • Most list operations allow negative index references. -1 would be the last item in the list, -2 the 2nd to last, and so on...
  • Setting indexes, getting indexes, slicing, joining two lists, and appending are all available
• Variables are typed at creation: INT var! myvar creates an int variable named myvar and is set to the type's zero value (0 in this case) at creation
• Helpful error messages with stack straces
• Repl with tab completions for built-ins and imports
• Loading of a single file works, and any arguments passed at the shell will be passed in as a list named sys-args
• Variables can be created in parent scopes with scoped-var! or updated in parent scopes with scoped-set!
• There is a lot of "operator overloading" going on with +, -, *, / to have them work on strings and lists in interesting ways.
• Procedures have the ability to add a docstring proc myproc | I am a docstring | "hello" println end defines a procedure, and docstring! myproc would add the docstring for that proc to TOS (whence it could be printed). Built-ins also have docstrings available (learning the stack order for everything can be a pain without them). To see the available words/procedures the system knows use words, which will add a string containing them to TOS. You can then try to docstring! them and find out what they do
• The functional programming operations each! and filter! are available. each!, by nature of operating on the stack, can also be used as reduce as well (0 [1 2 3] each! + sums all of the numbers in the list starting at 0, resulting in 6 on TOS)
• Type comparisons are as easy as casting to TYPE and comparing. Assuming two items are on the stack: TYPE cast swap TYPE cast =
• Procedures/built-ins that end their identifier with a ! indicate that they will "forward read" a word. So the code INT var! my-int will place INT (a TYPE) on the stack, then var! will pop the type and create a variable in the current scope with that type, var! will then "forward read" the token stream to get my-int and assign that SYMBOL to the newly created INT variable. Users can create their own forward reading procedures by using proc! instead of proc. This works sort of like a macro in some other languages: in the procedure body you can use the symbol procarg and when the procedure is run it will forward read to get a symbol or value and replace all instances of procarg with that value before executing the procedure. This allows for some form of argument passing beyond using the stack and is also the only way to reference a variable symbol without referencing its value

Gotchas / Weird Things

• Unlike a traditional forth style system, felise encourages use of local variables. Its lexical scoping and garbage collection mean that the global scope does not get littered with random variables, and so it is very comfortable and easy to use local variables when defining procedures and not have to do the stack manipulation dance except where it makes sense to do so
• LIST type items have a starting index of 1
• When using cast to go from FLOAT to INT, the number will be floored rather than truncated (as is common). Positive numbers will function like a trunc, but negative numbers will not (the FLOAT -1.57 would cast to the INT -2)
• Using divide on two strings will do a split and produce a list of s1 split by s2 (TOS); this can be reversed by multiplying the list by a string. Similarly, subtracting anything from a LIST or STRING will attempt to remove the thing from the LIST or remove a stringified version of the thing from the STRING. This is just a sample of the "fun" overloading that is happening. Enjoy!
• File I/O is done without file handles (from a user perspective) and consists only of append (file-write) and read all (file-read). When combined with file-remove and file-exists?, most common file actions can be accomplished with this set of features (when was the last time you needed to seek in a file). It makes both writing and reading a bit more expensive, but simplicity if the tradeoff
• You can close out the global scope with a terminal end (at your option) and anything after it in the file will be ignored by the interpreter, allowing for long comments/notes at the end of the file
• While I write above that a non-windows system is required... it may build and work on windows. I'm not sure, as I do not use that OS and do not have a system to test it on. YMMV

License

Notice: This is not free/libre software as defined by the FSF or OSC. I do not ascribe to their vision of free software. I believe that capitalism is an inherent threat to freedom and is an injustice to everyone. As such I use the Floodgap Free Software License (Version 1). The license is included in the repo (see LICENSE file), and is mentioned and linked in each source code file. The basic idea is that you as an individual can do anything you like with this software that does not involve, directly or indirectly, money changing hands. There are a few exceptions to this noted in the license, but in general, that is the idea. Free to use in any way you like, so long as you don't charge for it. You are, of course, welcome to license code produced for the felise interpreter/language in any way you choose. The license I have chosen only applies to my interpreter code.