Chapter 2
ZL Tutorial

This chapter will give you an overview of ZL with a focus on ZL extensibility features.

2.1 Hello World

ZL basic syntax is the same as it is in C or C++ therefor a simple Hello World program is nothing but:

  int main() {
    printf("Hello World!")

To compile, save this file to hello.zl and then use zlc:

  zlc hello.zl

which should create an excitable a.out which you can run.

Note that there is no #include line in the source file. The ZL prelude includes some of the more common functions from the C standard library and ZL source files are not run through the C preprocessor by default.

The driver script zlc is meant to act as a drop drop in replacement for GCC. For example,

  zlc main.zl file1.cpp file2.zlp -o main

compiles the pure ZL source file main.zl, the C++ source file file1.cpp, and the ZL source file file2.zlp into an executable main. Each file is compiled slightly differently based on the extension as ZL has different modes for C, C++, and ZL source files. In addition C, C++, and ZL source files with the .zlp extension are run through the C preprocessor to handle includes and other token-level macros that the higher-level ZL macro processor can not handle. As already mentioned, pure ZL files with .zl extension are not preprocessed.

2.2 Macro Introduction: A Perl Like Or

C’s or operator returns a boolean value. Sometimes it would be more convenient if it returns the first true value like Perl does. For example if we wanted to use the first non-NULL pointer we could use:


where p1 and p2 are pointers. Defining or as a function (or even a function template) will not work because the or macro should not evaluate the second argument if the first one evaluates to a true value.

Fortunately, ZL makes it very easy to define simple macros such as this using the macro form:

  macro or(x, y) { ({typeof(x) t = x; t ? t : y;}); }

(In ZL, as in GCC, the ({...}) is a statement expression whose value is the result of the last expression, and typeof(x) gets the type of a variable.) Like Scheme macros [6], ZL macros are hygienic, which means that they respect lexical scope. For example, the t used in or(0.0, t) and the t introduced by the or macro remain separate, even though they have the same symbol name.

Syntax Macros. It is even possible to locally redefine the behavior of the built-in || operator so that we could instead just use p1 || p2. All that is required to archive this transformation is to change the macro to smacro:

  smacro or(x, y) { ({typeof(x) t = x; t ? t : y;}); }

The smacro form defines a syntax macro which operates on syntax, as oppose to the macro form which defines a function-call macro in which calls to it take the shape of a function call.

The syntax version of the or macro works because ZL performs most of its operations using interment S-expression like language. For example p1 || p2 is parsed as (or p1 p2), before being converted into an AST (details of which are given in Chapter 3)

Keyword Parameters. The or macro above has too positional parameters. Macros can also have keyword parameters and default values. For example:

  macro sort(list, :compar = strcmp) {...}

defines the macro sort, which takes the keyword argument compar, with a default value of strcmp. A call to sort will look something like sort(list, :compar = mycmp).

2.3 Classes and Templates

ZL also supports a limited subset of C++ which includes classes and very basic template support. Classes are defined if the usual way. ZL recognizes the template syntax, but templates are defined using a macro and must be explicitly instantiated by calling the macro. For example, to use a vector of type int:

  int main() {
    vector<int> con;

We will revisit the mk_vector macro in Section 2.6.

2.4 Extending The Grammar: Foreach

One very common operation is to iterate through the elements of a container. It is such a common operation that support for it was incorporated latest version of the C++ standard via special version of the for loop. ZL doesn’t support this version, but we can very easily define a similar construct. Instead of extending the syntax of for we will define a new syntax form: foreach. The usage of the form is straightforward; for example, the following code prints all the elements of a vector:

  int main() {
    vector<int> con;
    foreach (elm in con) {

The first thing we need to do to support the new foreach form is to define some new syntax. ZL provides basic support for new syntax by extending the PEG using the new_syntax form. We add support the new loop form by rewriting the CUSTOM_STMT production as follows:

  new_syntax {
    CUSTOM_STMT := _cur / <foreach> "foreach" "(" {ID} "in" {EXP} ")" {STMT};

The special _cur production refers to the previous version of the CUSTOM_STMT and anything between {} becomes a subpart of the syntax object that is named between the <>. With the new syntax defined the definition of the macro is fairly straightforward:

  smacro foreach (VAR, WHAT, BODY) {
    typeof(WHAT) & what = WHAT;
    typeof(what.begin()) i=what.begin(), e=what.end();
    for (; i != e; ++i) {
      typeof(*i) & VAR = *i;

Chapter 3 gives more details on the PEG and how to extend it.

2.5 Procedural Macros

All the macros shows so far are simple pattern-based macros. Sometimes we need to do more than simply rearrange syntax. For example, in the foreach macro of the previous section, if the container lacks a begin or end method ZL will return a cryptic error message which will refer to the expanded output, rather than the user input; it would be far better to give a straightforward error message which refers to the user input. For these more complex tasks ZL provides support for procedural macros which can take action base on their input.

A procedural macro is a function which maps one syntax object to another. ZL provides special syntax support for defending these functions. They look a lot like pattern-based macros expect that the keyword proc proceeds the macro and the body of the macro is ZL code rather than the expanded output. Syntax is created using special quasiquote syntax. For example, the following macro defines a version of the foreach which returns an error message if the container lacks the proper methods:

  proc smacro foreach (VAR, WHAT, BODY) {
    Syntax * what = ‘‘WHAT;
    if (!symbol_exists(‘‘begin, what) || !symbol_exists(‘‘end, what))
      return error(what, "Container lacks begin or end method.");
    return ‘{
      typeof(WHAT) & what = WHAT;
      typeof(what.begin()) i=what.begin(), e=what.end();
      for (;i != e; ++i) {typeof(*i) & VAR = *i; BODY;}

The symbol_exists and error are API functions which are callbacks into the compiler. The ‘‘ and ‘{ are quasiquotes; the first forms quotes an identifier; the second form quotes arbitrary syntax. Pattern variables can only be used inside quasiquotes; however, the symbol_exists and error function requires a syntax object. Hence, ‘‘WHAT is used to extract the syntax object associated with the pattern variable and store it in a local variable.

Antiquotes. Rather than using pattern variables, it is possible to use antiquotes to store the result of the match in local variables by prefixing the variable with a $. For example, here is a version of foreach that uses antiquotes instead:

  proc smacro foreach ($var, $what, $body) {
    if (!symbol_exists(‘‘begin, what) || !symbol_exists(‘‘end, what))
      return error(what, "Container lacks begin or end method.");
    return ‘{
      typeof($what) & what = $what;
      typeof(what.begin()) i=what.begin(), e=what.end();
      for (;i != e; ++i) {typeof(*i) & $var = *i; $body;}

When an antiquote is used when matching syntax the results of the match are stored in a local variable hence $body gets stored in the local variable body, which is then used for by the symbol_exist and error function. When an antiquote is used inside quasiquotes than it is replaced with the value of the expression; in most cases this is a local variable.

Additional Info. Shown here is only a small taste of what procedural macros can do. ZL procedural macros are very powerful and in many ways, act more as an extension to the compiler than a simple macro that transforms text from one form to another. Chapter 4 gives additional information on procedural macros, which includes a lower level and more powerful API.

2.6 Templates and Bending Hygiene

As already mentioned in Section 2.3, ZL does not have builtin template support. Instead a macro can be used to create instances of a template. Figure 2.1 shows the macro and supporting code to create for the vector template class.

  template vector;
  macro mk_vector (T,@_) :(*)
    __once vector<T>
    class vector<T> {
      T * data_;
      T * end_;
      T * storage_end_;
      typedef T * iterator;
      typedef const T * const_iterator;
      typedef size_t size_type;
      iterator begin() const {return data_;}
      iterator end()   const {return end_;}
      bool empty() const {return data_ == end_;}
      size_t size() const {return end_ - data_;}
      size_t max_size() const {return (size_t)-1;}
      size_t capacity() const {return storage_end_ - data_;}
      T & front() const {return *data_;}
      T & back() const {return *(end_-1);}
      T & operator[](size_t n) const {return data_[n];}
      vector<T>() : data_(NULL), end_(NULL), storage_end_(NULL) {}
      vector<T>(size_t n) : ... {resize(n);}
      vector<T>(const vector<T> & other) : ... {...}
      void push_back(const T & el) {...}
      void pop_back() {...}
      void assign(const T * d, unsigned sz) {...}
      void resize(size_t sz) {...}
      void erase(T * pos) {...}
      ~vector<T>() {...}

Figure 2.1: ZL implementation of vector template class

There are two things to note about this macro, the first is some special syntax to make the template instance to behave as expected, and the second is to bend normal hygiene rules so that vector and the members of the class are visible outside the macro.

Special Template Syntax. Template classes have special syntax that ZL needs to know to recognize; which is done using the template form (which is used differently than it is in C++). In addition templates have special linkage rules so that there are not multiple instances of the same template in the executable, hence __once is used to declare that duplicate symbols should be merged when linking. Other than that, the body of the mk_vector is the same as it would be for a template class in ZL.

Bending Hygiene. Normal hygiene rules will normally prevent the mk_vector macro from working as expected as all symbols created will only be visible to the macro itself. To fix this ZL macros supports special syntax, the :(*), which will export top-level symbols from the macro (which includes the members of the vector class). Details of exactly what is exported and how are given in Section 6.2.1.

It is also possible to export specific symbols by specifying the symbol name instead of *, again see Section 6.2.1 for the details.

2.7 Bending Hygiene Part 2: Fancy Loops

In the previous section we needed to make the macro introduced symbols visible outside the context of the macro. Sometimes it is necessary to make symbols visible to the parameters of the macros. For this case, ZL provides a different mechanism, the fluid binding. For example, if we wanted to create a version of the for loop which also supports a Perl style redo we might try

  macro redo() {goto redo;}
  smacro for (INIT, TEST, INC, BODY) {
    for (INIT; TEST; INC) {

(where the for used in the macro will refer to the builtin for and not the macro itself and redo: is a label for the goto in the redo macro). A simple usage of the macro could be:

  int main() {
    for (int i = 0; i < 10; ++i) {
      printf("%d\n", i);
      if (i % 3 == 0)

Unfortunately this will not work. The problem is the redo label introduced in the for macro is private to that macro. Exporting the label will not help, as the label used in the redo macro is still lexically scoped. The export syntax only has meaning to new binding forms not existing ones, hence that form can’t be used. ZL does provide a way to make both redo labels totally unhygienic (see Section 6.2.3), but that solution will not compose well [4].

What is really needed is something akin to dynamic scoping in the hygiene system. That is, for the redo label to be scoped based on where it is used when expanded, rather than where it is written in the macro definition. This can be done by marking the redo symbol as fluid using fluid_label at the top level and then using fluid when defining the symbol in local scope. For example, the following will allow the above to compile:

  fluid_label redo;
  macro redo() {goto redo;}
  smacro for (INIT, TEST, INC, BODY) {
    for (INIT; TEST; INC) {
      fluid redo:

The general form to declare fluid binding is fluid_binding, the details of which are described in Section 6.2.2

2.8 User Types

ZL does not have a built-in notion of classes, rather the type system has what is known as a user type of which classes are built from using macros. A user type has two parts: a type, generally a struct, to hold the data for the class instance, and a collection of symbols for manipulating the data.

The collection of symbols is a module. For example,

  module M { int x;
             int foo(); }

defines a module with two symbols. Module symbols are used by either importing them into the current namespace, or by using the special syntax M::x, which accesses the x variable in the above module.

A user type is created by using the user_type primitive, which serves as the module associated with the user type. A type for the instance data is specified using associate_type.

As an example, the class1

  class C { int i;
            int f(int j) {return i + j;} };

roughly expands to:

  user_type C {
    struct Data {int i;};
    associate_type struct Data;
    macro i (:this ths = this) {...}
    macro f(j, :this ths = this) {f‘internal(this, j);}
    int f‘internal(...) {...}

which creates a user type C to represent a class C; the structural type Data is used for the underlying storage. The macro i implements the i field, while the f macro implements the f method by calling the function f‘internal with ths as the first parameter. Additional details of how classes are implemented are given in Chapter 9.

Converted From LaTeX using TeX4ht. PDF Version