XL - An extensible language

WARNING: XL is a work in progress. Even if there are some bits and pieces that happen to already work, XL is presently not suitable for any serious programming. Examples given below may sometimes simply not work. Take it as a painful reminder that the work is far from finished, and, who knows, as an idea for a contribution. See HISTORY for how we came to the present mess, and Compiler status for information about what is expected to work, and Compiler overview for a quick overview of the compiler internals.

XL is an extensible programming language designed to accomodate a variety of programming needs with ease. Being extensible means that the language is designed to make it very easy for programmers to adapt the language to suit their needs, for example by adding new programming constructs. In XL, extending the language is a routine operation, much like adding a function or creating a class in more traditional programming languages.

As a validation of this bold claim, XL has a single fundamental operator, the definition operator, which you write [Pattern] is [Implementation, where] [Pattern] is a program pattern, like X+Y, and [Implementation] explains how to translate that pattern, for example Add X, Y.

Everything that is built-in in most other programming languages, from basic data types to arithmetic to conditionals to loops is provided by the standard library in XL. You can replace these constructs if you want, or add your own. Adding a new kind of loop is not more difficult in XL than adding a function, and it uses the same syntax.

• Simple examples
• Dialects and use cases
• If you come from another language
• One operator to rule them all
• Syntax: Look, Ma, no keywords!
• XL as a functional language
• Subtelty #1: Expression vs statement
• Subtelty #2: Infix vs. Prefix
• Subtelty #3: Delayed evaluation
• Subtelty #4: Closures

For more information, please consult the XL handbook, also available in asciidoc format and PDF format

WARNING This documentation, like the compiler, is work in progress and presently extremely messy, incomplete and inaccurate.

A few simple examples

A program computing the factorial of numbers between 1 and 5 would be written as follows:

0! is 1
N! is N * (N-1)!

for I in 1..5 loop
    print "The factorial of ", I, " is ", I!

As a testament to its extensible nature, fundamental operations in XL are defined in the standard library, including operations that would be implemented using keywords in more traditional languages. For example, the if statement in XL is defined by the following code:

if [[true]]  then TrueClause else FalseClause   is TrueClause
if [[false]] then TrueClause else FalseClause   is FalseClause
if [[true]]  then TrueClause                    is TrueClause
if [[false]] then TrueClause                    is false

Similarly, the while loop is defined as follows:

while Condition loop Body is
    if Condition then
        Body
        while Condition loop Body

The standard library also provides implementations for usual operations. For example, if you evaluate 1+3, this is done through a definition for + on integer values that looks like the following (where ... denotes some implementation-dependent code):

X:integer + Y:integer is ...

Dialects and use cases

Two dialects of XL further demonstrate the extensibility of the language

• Tao3D focuses on real-time 3D animations and can be used as a scriptable presentation software, or as someone once described it, a sort of real-time 3D LaTeX Lisp. In Tao3D, you describe a slide with a program that looks like the following code:

  import Slides
  
  slide "A simple slide example",
      * "This looks like some kind of markdown language"
      * "But code makes it powerful: your mouse is on the " & position
      position is
          if mouse_x < 0 then "left" else "right"
  

  The examples above use the new syntax in XL, with is as its definition operator. Older variants of the language used -> instead. If you downloaded a pre-built binary of Tao3D, chances are that you need to replace is with -> for the code above to work as intended.

• ELFE, formerly ELIOT (Extensible Language for the Internet of things) was an experiment on how to write distributed software that looks like a single program, for example to control swarms of devices in the context of the Internet of Things. An example of a simple ELFE program would be:

  WORKER is "worker.mycorp.com"
  MIN_TEMP is 25
  MAX_TEMP is 55
  
  invoke WORKER,
      every 2s,
          reply
              display temperature
  
  display Temp:real is
      print "The temperature of ", WORKER, " is ", Temp
  

The present branch, bigmerge, is an ongoing effort to reconverge the various dialects of XL. At the moment, it should pass most of the ELFE-level tests, although this is not actively tested. Getting it to support Tao3D is somewhat more difficult and may take some time.

If you come from another language

If you are familiar with other programming languages, here are a few things that may surprise you about XL.

• There are no keywords. In C, if is a keyword. In XL, it's just a name.
• The language is designed primarily to be readable and writable by humans. For example, there are special parsing rules to match how we read the code.
• The language is homoiconic, i.e. programs are data, like in Lisp. This forms the basis of XL extensibility.
• Evaluation is defined entirely in terms of rewrites of a very simple abstract. syntax tree that represents the program being evaluated.
• The precedence of all operators is dynamic, in the sense that it's loaded from a configuration file
• The language is primarily defined by its own standard library, rather than by special rules in the compiler.

Semantics: One operator to rule them all

XL has one fundamental operator, is, the definition operator. This operator can be read as transforms into, i.e. it transforms the code that is on the left into the code that is on the right.

It can define simple variables

pi              is      3.1415926

It can define lists

words           is      "xylophage", "zygomatic", "barfitude"

It can define functions

abs X           is      if X < 0 then -X else X

It can define operators

X ≠ Y           is      not X = Y

It can define specializations for particular inputs

0!              is      1
N!  when N > 0  is      N * (N-1)!

It can define notations using arbitrary combinations of operators

A in B..C       is      A >= B and A <= C

It can define optimizations using specializations

X * 1           is      X
X + 0           is      X

It can define program structures

loop Body       is      Body; loop Body

It can define types

type complex    is      polar or cartesian
type cartesian  is      cartesian(re:number, im:number)
type polar      is      polar(mod:number, arg:number)

Note that types in XL indicate the shape of parse trees. In other words, the cartesian type above will match any parse tree that takes the shape of the word cartesian followed by two numbers, like for example cartesian(1,5).

It can define higher-order functions, i.e. functions that return functions

adder N         is      (X is N + X)
add3            is      adder 3

 // This will compute 8
 add3 5

This makes XL a truly functional language.

It can define maps associating a key to a value

my_map is
    0 is 4
    1 is 0
    8 is "World"
    27 is 32
    N when N < 45 is N + 1

// The following is "World"
my_map 8

// The following is 32
my_map[27]

// The following is 45
my_map (44)

This provides a functionality roughly equivalent to std::map in C++. However, it's really nothing more than a regular function with a number of special cases. The compiler can optimize some special kinds of mapping to provide an efficient implementation.

It can define templates (C++ terminology) or generics (Ada terminology)

_

// An (inefficient) implementation of a generic 1-based array type
type array [1] of T is
    Value : T
    1 is Value
type array [N] of T when N > 1 is
    Head  : array[N-1] of T
    Tail  : T
    I when I<N is Head[I]
    I when I=N is Tail

A : array[5] of integer
for I in 1..5 loop
    A[I] := I * I

It can define variadic functions

min X       is X
min X, Y    is { Z is min Y; if X < Z then X else Z }

// Computes 4
min 7, 42, 20, 8, 4, 5, 30

In short, the single is operator covers all kinds of declarations that are found in other languages, using a single, easy to read syntax.

Syntax: Look, Ma, no keywords!

XL has no keywords. Instead, the syntax relies on a rather simple recursive descent parser.

THe parser generates a parse tree made of 8 node types. The first four node types are leaf nodes:

• Integer is for integer numbers such as 2 or 16#FFFF_FFFF.
• Real is for real numbers such as 2.5 or 2#1.001_001_001#e-3
• Text is for text values such as "Hello" or 'World'. Text can be encoded using UTF-8
• Name is for names and symbols such as ABC or **

The last four node types are inner nodes:

• Infix are nodes where a named operator separates the operands, e.g. A+B or A and B.
• Prefix are nodes where the operator precedes the operand, e.g. +X or sin X. By default, functions are prefix.
• Postfix are nodes where the operator follows the operand, e.g. 3% or 5km.
• Block are nodes with only one child surrounded by delimiters, such as (A), [A] or {A}.

Of note, the line separator is an infix that separates statements, much like the semi-colon ;. The comma , infix is traditionally used to build lists or to separate the argument of functions. Indentation forms a special kind of block.

For example, the following code:

    tell "foo",
        if A < B+C then
            hello
        world

parses as a prefix tell, with an infix , as its right argument. On the left of the , there is the text "foo". On the right, there is an indentation block with a child that is an infix line separator. On the left of the line separator is the if statement. On the right is the name world.

This parser is dynamically configurable, with the default priorities being defined by the xl.syntax file.

Parse trees are the fundamendal data structure in XL. Any data or program can be represented as a parse tree. Program evaluation is defined as transformation of parse trees.

XL as a functional language

XL can be seen as a functional language, where functions are first-class entities, i.e. you can manipulate them, pass them around, etc:

adder X:integer is (Y is Y + X)

add3 is adder 3
add5 is adder 5

print "3+2=", add3 2
print "5+17=", add5 17
print "8+2=", (adder 8) 2

However, it is a bit different in the sense that the core data structure is the parse tree. Some specific parse trees, for example A+B, are not naturally reduced to a function call, although they are subject to the same evaluation rules based on tree rewrites.

Subtlety #1: expression vs. statement

The XL parse tree is designed to represent programs in a way that is relatively natural for human beings. In that sense, it departs from languages such as Lisp or SmallTalk.

However, being readable for humans requires a few special rules to match the way we read expressions. Consider for example the following:

write sin X, cos Y

Most human beings parse this as meaning write (sin(X),cos(Y)), i.e. we call write with two values resulting from evaluating sin X and cos Y. This is not entirely logical. If write takes comma-separated arguments, why wouldn't sin also take comma-separated arguments? In other words, why doesn't this parse as write(sin(X, cos(Y))?

This shows that humans have a notion of expressions vs. statements. Expressions such as sin X have higher priority than commas and require parentheses if you want multiple arguments. By contrast, statements such as write have lower priority, and will take comma-separated argument lists. An indent or { } block begins a statement, whereas parentheses () or square brackets [] begin an expression.

There are rare cases where the default rule will not achieve the desired objective, and you will need additional parentheses.

Subtlety #2: infix vs. prefix

Another special rule is that XL will use the presence of a space on only one side of an operator to disambiguate between an infix or a prefix. For example:

write -A   // write (-A)
B - A      // (B - A)

Subtlety #3: Delayed evaluation

When you pass an argument to a function, evaluation happens only when necessary. Deferred evaluation may happen multiple times, which is necessary in many cases, but awful for performance if you do it by mistake.

Consider the following definition of every:

every Duration, Body is
    loop
        Body
        sleep Duration

In that case, we want the Body to be evaluated every iteration, since this is typically an operation that we want to execute at each loop. Is the same true for Duration? Most likely, no.

One way to force evaluation is to give a type to the argument. If you want to force early evaluation of the argument, and to check that it is a real value, you can do it as follows:

every Duration:real, Body is
    loop
        Body
        sleep Duration

Subtlelty #4: Closures

Like many functional languages, XL ensures that the value of variables is preserved for the evaluation of a given body. Consider for example:

adder X:integer is (Y is Y + X)
add3 := adder 3

In that case, adder 3 will bind X to value 3, but then the returned value outlives the scope where X was declared. However, X is referred to in the code. So the returned value is a closure which integrates the binding X is 3.

Compiler status

Work items for the XL compiler (will be turned into GitHub issues)

Language definition

• Language definition documentation published
• Quickly-scanned To-Do list (this section)
• Finish chapter on compilation
• Finish chapter on programming paradigms
• Description of the module system
• Description of the standard library
• Decide whether it's import, use or both (import doing the importing and use bringing the referenced expression in scope).

Recent language changes

• Switch from -> to is as the definition operator
• Switch type definition from type Pattern to matching Pattern (issue #5)
• Implement syntactic sugar (can it be lib only?), e.g. type X is Y and module X with Y or to Copy(...) is blah (issue #6)
• Support for nested functions and proper nested scopes. See issue #8.
• Scope injection and scoping, i.e. meaning of scope.Foo and scope Foo in the language. Deal with (scope) foo, etc. See scoping and issue #9.
• Safe implementation of for loop using scope injection (see name parameters). This is issue #7.
• Implement metabox ([[true]]). This is issue #10.
• Write the interface and implementation of the type type (issue #11).
• Implement union types (T1 or T2), as well as and and not. See issue #12
• Revisit dynamic dispatch based on types. See issue #13
• Implement type inheritance checks (Derived like Base). See issue #14
• Generate lifetime values. See issue #15
• Implement own and ref types. See issue #16
• Implement in, out, inout types. See issue #17
• Implement creation and destruction. See issue #18

Symbol table

• Symbol table as an XL parse tree
• Symbol table storing actual definitions, using is operator. Currently, the symbol table uses a "nonsensical" prefix. See issue #19

Scanner

• Scanner
• Scan version numbers correctly, e.g. 1.20.5 not parsed as (1.2).5
• Better Unicode classification of letters versus punctuation. Currently, only ASCII punctuation is recognized as such.
• Option to transparently convert -> to is (for Tao3D importing)
• Scanner-time processing of syntax statements from imported files

Parser

• Parser
• "Binary" terminal node holding arbitrary binary data
• "Literal" terminal node holding arbitrary text parsed from regexp
• Add more error checks for failure cases, better error reporting

Renderer

• Renderer (-show option)
• Style renderer with -style option and .stylesheet files
• Preserving comments in renderer output
• Rendering binary or regexp nodes

Runtime

• Foreign function interface (FFI) with easy macro (see native.h
• Tag all useful runtime functions with FFI
• Find a strategy for migrating Tao3D 1500 runtime calls
• Cleanup the runtime from obsolete / useless functions

Interpreter (-O0)

• Simple interpreter (-O0 or -i)
• Explicit FFI for interpreter (extern C syntax)
• Add "sandboxed" mode (on by default for interpreter) that disables the above
• Connect native.h FFI to interpreter
• Implement modules in interpreter
• Implement error and compile_error evaluation rules in interpreter
• Update type system to recognize matching prefix instead of type

Bytecode interpreter (-O1) BROKEN, LIKELY NOT TO BE REPAIRED

• Bytecode interpreter (broken, do not use)
• Fix bytecode interpreter or get rid of its remnants (-O2)
• Redesign bytecode interpreter as emitting LLVM byte code?

LLVM "fast compiler" (-O2)

Simplistic compiler that does only run-time type analysis

• Fast compiler functionally similar to what was used in Tao3D
• Find strategy to re-connect it to Tao3D

LLVM optimized compiler (-O3)

• LLVM-based compiler
• LLVM-CRAP (adapting to multiple versions of LLVM)
• native.h for building FFI
• Option to emit LLVM bitcode (-B or -emit_ir)
• Option to pass bitcode to LLVM bitcode compiler (Automatically do something like xl -B ... | llc -filetype=asm)
• Option to directly emit disassembly
• Pass LLVM options to LLVM (-llvm-foo)
• Replace algo-W type analysis with forward-only type analysis.
• Implement auto-boxing for parameter scopes
• Self-compiling compiler
• Rebuild Tao3D on top of "new" XL

Build system

• Portable auto-configuring makefile (Using make-it-quick)
• Targets for opt, debug, release and check
• Target for install
• Fix issue with too much stuff being rebuilt for install

CI / Testing

• Basic QE framework / testing script (alltest script, make check)
• Fix breakage in make check with macOS Catalina DYLD_LIBRARY_PATH
• Basic CI in GitLab
• Repair breakage in GitLab CI
• Add tests for variants of LLVM in GitLab CI
• Basic CI in GitHub

Modules

• Reimplement module system

Standard library

• Standard library
• Overall structure (WIP)
• Types
• Arrays
• STL-style containers
• Algorithms
• I/Os
• Threading / synchronization
• Math and numerics
• Standard math functions
• Complex module
• Vectors
• Matrices
• Text processing
• Time
• Graphics (Tao3D based?)

Compiler overview

The interpreter / compiler recently went through a rather heavy merge of several incompatible branches. As a result, it inherited the best of the various branches, but is overall quite broken. There are actually several, somewhat incompatible versions of the language, some of which reside in different binaries, some in the primary binary.

The primary binary resides in the src directory. It is written in C++, and there is currently no real plan to self-compile it, although there are plans to use it as a basis for a self-compiling compiler bootstrap someday.

That primary binary contains a single scanner and parser for the XL language, but three different ways to evaluate it, which are instances of the C++ Evaluator class. These three ways to evaluate XL are selected using the -O option.

• -O0 or -i selects an interpreter. This interpreter is essentially similar to what used to be the ELFE implementation of the language, i.e. a very small implementation that performs evaluation using parse tree rewrites. It sort of works, passes most tests with make check, and is overall sane, if a bit slow (similar to bash in my testing). It can be used for example as an extension language for your application, and does not draw much in terms of dependencies. You would add your own vocabulary using simple-to-write "modules". See the Makefile for examples. That part is the only one I can advertise as possibly useful. In particular, it correctly runs the examples in the demo directory, which are the older ELFE demos, i.e. distributed programming from a single source code.

• -O1 selects the FastCompiler, which is basically an adaptation of the compiler used in the Tao3D program, with the intent to allow the master branch of XL to be able to support Tao3D again without major incompatibilities. It generates machine code using LLVM, but the generated code is relatively inefficient because it manipulates the parse tree. For example, an integer value is always represented by an Integer pointer, so there is always an indirection to access it. Also, while forward-porting that compiler to a version of the compiler that had dropped it, I broke a number of things. So under repair, and not currently able to support Tao3D yet.

• -O2 and above select the Compiler class, which is an ongoing attempt at implementing XL the way I always wanted to, using type inference to generate efficient code. Presently, the type inference is so badly broken that it's likely to reject a number of very valid programs, including the basic factorial example. I have hope, though. At some point, that implementation was able to compete with C on relatively simple programs, but only with a lot of type annotations. I'm trying to achieve the same result without the type annotations. We're getting there. Like -O1, -O2 output uses LLVM to generate machine code, but that time, it's good machine code.

If you think 3 implementations is bad, wait. There is more. There is a Bytecode class that is yet another evaluator that attempted to generate a bytecode so as to accelerate interpreted evaluation, without having to bring in LLVM and all the risks associated with dynamic code generation (e.g. if you use XL as an extension language). Unfortunately, that bytecode experiment went nowhere. It's slow, ridden with bugs, and has severely damaged other parts of the compiler. I can't wait to expurge it from the compiler.

So now that's it, right? Well... No.

You see, the current XL started life as a "runtime" language for the "real" XL. The original XL looked more like Ada, and had very different type semantics. See HISTORY for all the gory details. Suffice it to say here that this compiler resides in the xl2 directory (because, yes, it was already version 2 of the language). One reason for me to keep it is that it's the only version of the compiler that ever came close to self-compiling it. So I keep it around to remind myself of various neat tricks that XL made possible, like the translate instruction.

Now, you are really done, right? Well... There's one more.

See, I really want the compiler to self-compile. So in order to prepare for that, there is a native directory where I store tidbits of what the future compiler and library would look like. Except that this is really an exploratory scratchpad, so the various modules are not even consistent with one another... But ultimately, if everything goes according to plan, the C++ compiler should be able to compile native in order to generate a compiler that would compile itself.