<?xml version="1.0" encoding="utf-8" ?>
<!DOCTYPE chapter SYSTEM "chapter.dtd">

<chapter>
  <header>
    <copyright>
      <year>2001</year><year>2017</year>
      <holder>Ericsson AB. All Rights Reserved.</holder>
    </copyright>
    <legalnotice>
      Licensed under the Apache License, Version 2.0 (the "License");
      you may not use this file except in compliance with the License.
      You may obtain a copy of the License at

          http://www.apache.org/licenses/LICENSE-2.0

      Unless required by applicable law or agreed to in writing, software
      distributed under the License is distributed on an "AS IS" BASIS,
      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
      See the License for the specific language governing permissions and
      limitations under the License.

    </legalnotice>

    <title>The Abstract Format</title>
    <prepared>Arndt Jonasson</prepared>
    <responsible>Kenneth Lundin</responsible>
    <docno>1</docno>
    <approved></approved>
    <checked></checked>
    <date>2000-12-01</date>
    <rev>A</rev>
    <file>absform.xml</file>
  </header>
  <p>This section describes the standard representation of parse trees for Erlang
    programs as Erlang terms. This representation is known as the <em>abstract
    format</em>. Functions dealing with such parse trees are
    <seealso marker="compiler:compile#forms/1">
    <c>compile:forms/1,2</c></seealso> and functions in the following
    modules:</p>

  <list type="bulleted">
    <item><seealso marker="stdlib:epp">
      <c>epp(3)</c></seealso></item>
    <item><seealso marker="stdlib:erl_eval">
      <c>erl_eval(3)</c></seealso></item>
    <item><seealso marker="stdlib:erl_lint">
      <c>erl_lint(3)</c></seealso></item>
    <item><seealso marker="stdlib:erl_parse">
      <c>erl_parse(3)</c></seealso></item>
    <item><seealso marker="stdlib:erl_pp">
      <c>erl_pp(3)</c></seealso></item>
    <item><seealso marker="stdlib:io">
      <c>io(3)</c></seealso></item>
  </list>

  <p>The functions are also used as input and output for parse transforms, see
    the <seealso marker="compiler:compile"><c>compile(3)</c></seealso>
    module.</p>

  <p>We use the function <c>Rep</c> to denote the mapping from an Erlang source
    construct <c>C</c> to its abstract format representation <c>R</c>, and write
    <c>R = Rep(C)</c>.</p>

  <p>The word <c>LINE</c> in this section represents an integer, and denotes the
    number of the line in the source file where the construction occurred.
    Several instances of <c>LINE</c> in the same construction can denote
    different lines.</p>

  <p>As operators are not terms in their own right, when operators are
    mentioned below, the representation of an operator is to be taken to
    be the atom with a printname consisting of the same characters as the
    operator.</p>

  <section>
    <title>Module Declarations and Forms</title>
    <p>A module declaration consists of a sequence of forms, which are either
      function declarations or attributes.</p>

    <list type="bulleted">
      <item>
        <p>If D is a module declaration consisting of the forms
          <c>F_1</c>, ..., <c>F_k</c>, then
          Rep(D) = <c>[Rep(F_1), ..., Rep(F_k)]</c>.</p>
      </item>
      <item>
        <p>If F is an attribute <c>-export([Fun_1/A_1, ..., Fun_k/A_k])</c>,
          then Rep(F) =
          <c>{attribute,LINE,export,[{Fun_1,A_1}, ..., {Fun_k,A_k}]}</c>.</p>
      </item>
      <item>
        <p>If F is an attribute <c>-import(Mod,[Fun_1/A_1, ..., Fun_k/A_k])</c>,
          then Rep(F) =
          <c>{attribute,LINE,import,{Mod,[{Fun_1,A_1}, ...,
          {Fun_k,A_k}]}}</c>.</p>
      </item>
      <item>
        <p>If F is an attribute <c>-module(Mod)</c>, then
          Rep(F) = <c>{attribute,LINE,module,Mod}</c>.</p>
      </item>
      <item>
        <p>If F is an attribute <c>-file(File,Line)</c>, then
          Rep(F) = <c>{attribute,LINE,file,{File,Line}}</c>.</p>
      </item>
      <item>
        <p>If F is a function declaration <c>Name Fc_1 ; ... ; Name Fc_k</c>,
          where each <c>Fc_i</c> is a function clause with a pattern sequence of
          the same length <c>Arity</c>, then Rep(F) =
          <c>{function,LINE,Name,Arity,[Rep(Fc_1), ...,Rep(Fc_k)]}</c>.</p>
      </item>
      <item>
        <p>If F is a function specification <c>-Spec Name Ft_1; ...; Ft_k</c>,
          where <c>Spec</c> is either the atom <c>spec</c> or the atom
          <c>callback</c>, and each <c>Ft_i</c> is a possibly constrained
          function type with an argument sequence of the same length
          <c>Arity</c>, then Rep(F) =
          <c>{attribute,Line,Spec,{{Name,Arity},[Rep(Ft_1), ...,
          Rep(Ft_k)]}}</c>.</p>
      </item>
      <item>
        <p>If F is a function specification
          <c>-spec Mod:Name Ft_1; ...; Ft_k</c>, where each <c>Ft_i</c> is a
          possibly constrained function type with an argument sequence of the
          same length <c>Arity</c>, then Rep(F) =
          <c>{attribute,Line,spec,{{Mod,Name,Arity},[Rep(Ft_1), ...,
          Rep(Ft_k)]}}</c>.</p>
      </item>
      <item>
        <p>If F is a record declaration <c>-record(Name,{V_1, ..., V_k})</c>,
          where each <c>V_i</c> is a record field, then Rep(F) =
          <c>{attribute,LINE,record,{Name,[Rep(V_1), ..., Rep(V_k)]}}</c>.
          For Rep(V), see below.</p>
      </item>
      <item>
        <p>If F is a type declaration <c>-Type Name(V_1, ..., V_k) :: T</c>,
          where <c>Type</c> is either the atom <c>type</c> or the atom
          <c>opaque</c>, each <c>V_i</c> is a variable, and <c>T</c> is a type,
          then Rep(F) =
          <c>{attribute,LINE,Type,{Name,Rep(T),[Rep(V_1), ...,
          Rep(V_k)]}}</c>.</p>
      </item>
      <item>
        <p>If F is a wild attribute <c>-A(T)</c>, then
          Rep(F) = <c>{attribute,LINE,A,T}</c>.</p>
      </item>
    </list>

    <section>
      <title>Record Fields</title>
      <p>Each field in a record declaration can have an optional,
        explicit, default initializer expression, and an
        optional type.</p>

      <list type="bulleted">
        <item>
          <p>If V is <c>A</c>, then
            Rep(V) = <c>{record_field,LINE,Rep(A)}</c>.</p>
        </item>
        <item>
          <p>If V is <c>A = E</c>, where <c>E</c> is an expression, then
            Rep(V) = <c>{record_field,LINE,Rep(A),Rep(E)}</c>.</p>
        </item>
        <item>
          <p>If V is <c>A :: T</c>, where <c>T</c> is a type, then Rep(V) =
            <c>{typed_record_field,{record_field,LINE,Rep(A)},Rep(T)}</c>.</p>
        </item>
        <item>
          <p>If V is <c>A = E :: T</c>, where
            <c>E</c> is an expression and <c>T</c> is a type, then Rep(V) =
            <c>{typed_record_field,{record_field,LINE,Rep(A),Rep(E)},Rep(T)}</c>.
          </p>
        </item>
      </list>
    </section>

    <section>
      <title>Representation of Parse Errors and End-of-File</title>
      <p>In addition to the representations of forms, the list that represents
        a module declaration (as returned by functions in
        <seealso marker="stdlib:epp"><c>epp(3)</c></seealso> and
        <seealso marker="stdlib:erl_parse"><c>erl_parse(3)</c></seealso>)
        can contain the following:</p>

      <list type="bulleted">
        <item>
          <p>Tuples <c>{error,E}</c> and <c>{warning,W}</c>, denoting
            syntactically incorrect forms and warnings.
          </p>
        </item>
        <item>
          <p><c>{eof,LOCATION}</c>, denoting an end-of-stream
            encountered before a complete form had been parsed.
            The word <c>LOCATION</c> represents an integer, and denotes the
            number of the last line in the source file.
          </p>
        </item>
      </list>
    </section>
  </section>

  <section>
    <title>Atomic Literals</title>
    <p>There are five kinds of atomic literals, which are represented in the
      same way in patterns, expressions, and guards:</p>

    <list type="bulleted">
      <item>
        <p>If L is an atom literal, then Rep(L) = <c>{atom,LINE,L}</c>.</p>
      </item>
      <item>
        <p>If L is a character literal, then Rep(L) = <c>{char,LINE,L}</c>.</p>
      </item>
      <item>
        <p>If L is a float literal, then Rep(L) = <c>{float,LINE,L}</c>.</p>
      </item>
      <item>
        <p>If L is an integer literal, then
          Rep(L) = <c>{integer,LINE,L}</c>.</p>
      </item>
      <item>
        <p>If L is a string literal consisting of the characters
          <c>C_1</c>, ..., <c>C_k</c>, then
          Rep(L) = <c>{string,LINE,[C_1, ..., C_k]}</c>.</p>
      </item>
    </list>

    <p>Notice that negative integer and float literals do not occur as such;
      they are parsed as an application of the unary negation operator.</p>
  </section>

  <section>
    <title>Patterns</title>
    <p>If Ps is a sequence of patterns <c>P_1, ..., P_k</c>, then
      Rep(Ps) = <c>[Rep(P_1), ..., Rep(P_k)]</c>. Such sequences occur as the
      list of arguments to a function or fun.</p>

    <p>Individual patterns are represented as follows:</p>

    <list type="bulleted">
      <item>
        <p>If P is an atomic literal <c>L</c>, then Rep(P) = Rep(L).</p>
      </item>
      <item>
        <p>If P is a bitstring pattern
          <c>&lt;&lt;P_1:Size_1/TSL_1, ..., P_k:Size_k/TSL_k>></c>, where each
          <c>Size_i</c> is an expression that can be evaluated to an integer,
          and each <c>TSL_i</c> is a type specificer list, then Rep(P) =
          <c>{bin,LINE,[{bin_element,LINE,Rep(P_1),Rep(Size_1),Rep(TSL_1)},
          ..., {bin_element,LINE,Rep(P_k),Rep(Size_k),Rep(TSL_k)}]}</c>.
          For Rep(TSL), see below.
          An omitted <c>Size_i</c> is represented by <c>default</c>.
          An omitted <c>TSL_i</c> is represented by <c>default</c>.</p>
      </item>
      <item>
        <p>If P is a compound pattern <c>P_1 = P_2</c>, then Rep(P) =
          <c>{match,LINE,Rep(P_1),Rep(P_2)}</c>.</p>
      </item>
      <item>
        <p>If P is a cons pattern <c>[P_h | P_t]</c>, then Rep(P) =
          <c>{cons,LINE,Rep(P_h),Rep(P_t)}</c>.</p>
      </item>
      <item>
        <p>If P is a map pattern <c>#{A_1, ..., A_k}</c>, where each
          <c>A_i</c> is an association <c>P_i_1 := P_i_2</c>, then Rep(P) =
          <c>{map,LINE,[Rep(A_1), ..., Rep(A_k)]}</c>.
          For Rep(A), see below.</p>
      </item>
      <item>
        <p>If P is a nil pattern <c>[]</c>, then Rep(P) =
          <c>{nil,LINE}</c>.</p>
      </item>
      <item>
        <p>If P is an operator pattern <c>P_1 Op P_2</c>, where <c>Op</c> is a
          binary operator (this is either an occurrence of <c>++</c> applied to
          a literal string or character list, or an occurrence of an expression
          that can be evaluated to a number at compile time), then Rep(P) =
          <c>{op,LINE,Op,Rep(P_1),Rep(P_2)}</c>.</p>
      </item>
      <item>
        <p>If P is an operator pattern <c>Op P_0</c>, where <c>Op</c> is a
          unary operator (this is an occurrence of an expression that can be
          evaluated to a number at compile time), then Rep(P) =
          <c>{op,LINE,Op,Rep(P_0)}</c>.</p>
      </item>
      <item>
        <p>If P is a parenthesized pattern <c>( P_0 )</c>, then Rep(P) =
          <c>Rep(P_0)</c>, that is, parenthesized patterns cannot be
          distinguished from their bodies.</p>
      </item>
      <item>
        <p>If P is a record field index pattern <c>#Name.Field</c>,
          where <c>Field</c> is an atom, then Rep(P) =
          <c>{record_index,LINE,Name,Rep(Field)}</c>.</p>
      </item>
      <item>
        <p>If P is a record pattern <c>#Name{Field_1=P_1, ..., Field_k=P_k}</c>,
          where each <c>Field_i</c> is an atom or <c>_</c>, then Rep(P) =
          <c>{record,LINE,Name,[{record_field,LINE,Rep(Field_1),Rep(P_1)}, ...,
          {record_field,LINE,Rep(Field_k),Rep(P_k)}]}</c>.</p>
      </item>
      <item>
        <p>If P is a tuple pattern <c>{P_1, ..., P_k}</c>, then Rep(P) =
          <c>{tuple,LINE,[Rep(P_1), ..., Rep(P_k)]}</c>.</p>
      </item>
      <item>
        <p>If P is a universal pattern <c>_</c>, then Rep(P) =
          <c>{var,LINE,'_'}</c>.</p></item>
      <item>
        <p>If P is a variable pattern <c>V</c>, then Rep(P) =
          <c>{var,LINE,A}</c>, where A is an atom with a printname consisting
          of the same characters as <c>V</c>.</p>
      </item>
    </list>

    <p>Notice that every pattern has the same source form as some expression,
      and is represented in the same way as the corresponding expression.</p>
  </section>

  <section>
    <title>Expressions</title>
    <p>A body B is a non-empty sequence of expressions <c>E_1, ..., E_k</c>,
      and Rep(B) = <c>[Rep(E_1), ..., Rep(E_k)]</c>.</p>

    <p>An expression E is one of the following:</p>

    <list type="bulleted">
      <item>
        <p>If E is an atomic literal <c>L</c>, then Rep(E) = Rep(L).</p>
      </item>
      <item>
        <p>If E is a bitstring comprehension
          <c>&lt;&lt;E_0 || Q_1, ..., Q_k>></c>,
          where each <c>Q_i</c> is a qualifier, then Rep(E) =
          <c>{bc,LINE,Rep(E_0),[Rep(Q_1), ..., Rep(Q_k)]}</c>.
          For Rep(Q), see below.</p>
      </item>
      <item>
        <p>If E is a bitstring constructor
          <c>&lt;&lt;E_1:Size_1/TSL_1, ..., E_k:Size_k/TSL_k>></c>,
          where each <c>Size_i</c> is an expression and each
          <c>TSL_i</c> is a type specificer list, then Rep(E) =
          <c>{bin,LINE,[{bin_element,LINE,Rep(E_1),Rep(Size_1),Rep(TSL_1)},
          ..., {bin_element,LINE,Rep(E_k),Rep(Size_k),Rep(TSL_k)}]}</c>.
          For Rep(TSL), see below.
          An omitted <c>Size_i</c> is represented by <c>default</c>.
          An omitted <c>TSL_i</c> is represented by <c>default</c>.</p>
      </item>
      <item>
        <p>If E is a block expression <c>begin B end</c>,
          where <c>B</c> is a body, then Rep(E) =
          <c>{block,LINE,Rep(B)}</c>.</p>
      </item>
      <item>
        <p>If E is a case expression <c>case E_0 of Cc_1 ; ... ; Cc_k end</c>,
          where <c>E_0</c> is an expression and each <c>Cc_i</c> is a
          case clause, then Rep(E) =
          <c>{'case',LINE,Rep(E_0),[Rep(Cc_1), ..., Rep(Cc_k)]}</c>.</p>
      </item>
      <item>
        <p>If E is a catch expression <c>catch E_0</c>, then Rep(E) =
          <c>{'catch',LINE,Rep(E_0)}</c>.</p>
      </item>
      <item>
        <p>If E is a cons skeleton <c>[E_h | E_t]</c>, then Rep(E) =
          <c>{cons,LINE,Rep(E_h),Rep(E_t)}</c>.</p>
      </item>
      <item>
        <p>If E is a fun expression <c>fun Name/Arity</c>, then Rep(E) =
          <c>{'fun',LINE,{function,Name,Arity}}</c>.</p>
      </item>
      <item>
        <p>If E is a fun expression <c>fun Module:Name/Arity</c>, then Rep(E) =
          <c>{'fun',LINE,{function,Rep(Module),Rep(Name),Rep(Arity)}}</c>.
          (Before Erlang/OTP R15: Rep(E) =
          <c>{'fun',LINE,{function,Module,Name,Arity}}</c>.)</p>
      </item>
      <item>
        <p>If E is a fun expression <c>fun Fc_1 ; ... ; Fc_k end</c>,
          where each <c>Fc_i</c> is a function clause, then Rep(E) =
          <c>{'fun',LINE,{clauses,[Rep(Fc_1), ..., Rep(Fc_k)]}}</c>.</p>
      </item>
      <item>
        <p>If E is a fun expression <c>fun Name Fc_1 ; ... ; Name Fc_k end</c>,
          where <c>Name</c> is a variable and each
          <c>Fc_i</c> is a function clause, then Rep(E) =
          <c>{named_fun,LINE,Name,[Rep(Fc_1), ..., Rep(Fc_k)]}</c>.</p>
      </item>
      <item>
        <p>If E is a function call <c>E_0(E_1, ..., E_k)</c>, then Rep(E) =
          <c>{call,LINE,Rep(E_0),[Rep(E_1), ..., Rep(E_k)]}</c>.</p>
      </item>
      <item>
        <p>If E is a function call <c>E_m:E_0(E_1, ..., E_k)</c>, then Rep(E) =
          <c>{call,LINE,{remote,LINE,Rep(E_m),Rep(E_0)},[Rep(E_1), ...,
          Rep(E_k)]}</c>.</p>
      </item>
      <item>
        <p>If E is an if expression <c>if Ic_1 ; ... ; Ic_k  end</c>,
          where each <c>Ic_i</c> is an if clause, then Rep(E) =
          <c>{'if',LINE,[Rep(Ic_1), ..., Rep(Ic_k)]}</c>.</p>
      </item>
      <item>
        <p>If E is a list comprehension <c>[E_0 || Q_1, ..., Q_k]</c>,
          where each <c>Q_i</c> is a qualifier, then Rep(E) =
          <c>{lc,LINE,Rep(E_0),[Rep(Q_1), ..., Rep(Q_k)]}</c>.
          For Rep(Q), see below.</p>
      </item>
      <item>
        <p>If E is a map creation <c>#{A_1, ..., A_k}</c>,
          where each <c>A_i</c> is an association <c>E_i_1 => E_i_2</c>
          or <c>E_i_1 := E_i_2</c>, then Rep(E) =
          <c>{map,LINE,[Rep(A_1), ..., Rep(A_k)]}</c>.
          For Rep(A), see below.</p>
      </item>
      <item>
        <p>If E is a map update <c>E_0#{A_1, ..., A_k}</c>,
          where each <c>A_i</c> is an association <c>E_i_1 => E_i_2</c>
          or <c>E_i_1 := E_i_2</c>, then Rep(E) =
          <c>{map,LINE,Rep(E_0),[Rep(A_1), ..., Rep(A_k)]}</c>.
          For Rep(A), see below.</p>
      </item>
      <item>
        <p>If E is a match operator expression <c>P = E_0</c>,
          where <c>P</c> is a pattern, then Rep(E) =
          <c>{match,LINE,Rep(P),Rep(E_0)}</c>.</p>
      </item>
      <item>
        <p>If E is nil, <c>[]</c>, then Rep(E) = <c>{nil,LINE}</c>.</p>
      </item>
      <item>
        <p>If E is an operator expression <c>E_1 Op E_2</c>,
          where <c>Op</c> is a binary operator other than match operator
          <c>=</c>, then Rep(E) =
          <c>{op,LINE,Op,Rep(E_1),Rep(E_2)}</c>.</p>
      </item>
      <item>
        <p>If E is an operator expression <c>Op E_0</c>,
          where <c>Op</c> is a unary operator, then Rep(E) =
          <c>{op,LINE,Op,Rep(E_0)}</c>.</p>
      </item>
      <item>
        <p>If E is a parenthesized expression <c>( E_0 )</c>, then Rep(E) =
          <c>Rep(E_0)</c>, that is, parenthesized expressions cannot be
          distinguished from their bodies.</p>
      </item>
      <item>
        <p>If E is a receive expression <c>receive Cc_1 ; ... ; Cc_k end</c>,
          where each <c>Cc_i</c> is a case clause, then Rep(E) =
          <c>{'receive',LINE,[Rep(Cc_1), ..., Rep(Cc_k)]}</c>.</p>
      </item>
      <item>
        <p>If E is a receive expression
          <c>receive Cc_1 ; ... ; Cc_k after E_0 -> B_t end</c>,
          where each <c>Cc_i</c> is a case clause, <c>E_0</c> is an expression,
          and <c>B_t</c> is a body, then Rep(E) =
          <c>{'receive',LINE,[Rep(Cc_1), ...,
          Rep(Cc_k)],Rep(E_0),Rep(B_t)}</c>.</p>
      </item>
      <item>
        <p>If E is a record creation
          <c>#Name{Field_1=E_1, ..., Field_k=E_k}</c>,
          where each <c>Field_i</c> is an atom or <c>_</c>, then Rep(E) =
          <c>{record,LINE,Name,[{record_field,LINE,Rep(Field_1),Rep(E_1)},
          ..., {record_field,LINE,Rep(Field_k),Rep(E_k)}]}</c>.</p>
      </item>
      <item>
        <p>If E is a record field access <c>E_0#Name.Field</c>,
          where <c>Field</c> is an atom, then Rep(E) =
          <c>{record_field,LINE,Rep(E_0),Name,Rep(Field)}</c>.</p>
      </item>
      <item>
        <p>If E is a record field index <c>#Name.Field</c>,
          where <c>Field</c> is an atom, then Rep(E) =
          <c>{record_index,LINE,Name,Rep(Field)}</c>.</p></item>
      <item>
        <p>If E is a record update
          <c>E_0#Name{Field_1=E_1, ..., Field_k=E_k}</c>,
          where each <c>Field_i</c> is an atom, then Rep(E) =
          <c>{record,LINE,Rep(E_0),Name,[{record_field,LINE,Rep(Field_1),Rep(E_1)},
      ..., {record_field,LINE,Rep(Field_k),Rep(E_k)}]}</c>.</p>
      </item>
      <item>
        <p>If E is a tuple skeleton <c>{E_1, ..., E_k}</c>, then Rep(E) =
          <c>{tuple,LINE,[Rep(E_1), ..., Rep(E_k)]}</c>.</p>
      </item>
      <item>
        <p>If E is a try expression <c>try B catch Tc_1 ; ... ; Tc_k end</c>,
          where <c>B</c> is a body and each <c>Tc_i</c> is a catch clause,
          then Rep(E) =
          <c>{'try',LINE,Rep(B),[],[Rep(Tc_1), ..., Rep(Tc_k)],[]}</c>.</p>
      </item>
      <item>
        <p>If E is a try expression
          <c>try B of Cc_1 ; ... ; Cc_k catch Tc_1 ; ... ; Tc_n end</c>,
          where <c>B</c> is a body, each <c>Cc_i</c> is a case clause, and
          each <c>Tc_j</c> is a catch clause, then Rep(E) =
          <c>{'try',LINE,Rep(B),[Rep(Cc_1), ..., Rep(Cc_k)],[Rep(Tc_1), ...,
          Rep(Tc_n)],[]}</c>.</p>
      </item>
      <item>
        <p>If E is a try expression <c>try B after A end</c>,
          where <c>B</c> and <c>A</c> are bodies, then Rep(E) =
          <c>{'try',LINE,Rep(B),[],[],Rep(A)}</c>.</p>
      </item>
      <item>
        <p>If E is a try expression
          <c>try B of Cc_1 ; ... ; Cc_k after A end</c>,
          where <c>B</c> and <c>A</c> are a bodies,
          and each <c>Cc_i</c> is a case clause, then Rep(E) =
          <c>{'try',LINE,Rep(B),[Rep(Cc_1), ...,
          Rep(Cc_k)],[],Rep(A)}</c>.</p>
      </item>
      <item>
        <p>If E is a try expression
          <c>try B catch Tc_1 ; ... ; Tc_k after A end</c>,
          where <c>B</c> and <c>A</c> are bodies,
          and each <c>Tc_i</c> is a catch clause, then Rep(E) =
          <c>{'try',LINE,Rep(B),[],[Rep(Tc_1), ...,
          Rep(Tc_k)],Rep(A)}</c>.</p>
      </item>
      <item>
        <p>If E is a try expression
          <c>try B of Cc_1 ; ... ; Cc_k  catch Tc_1 ; ... ; Tc_n after A
          end</c>, where <c>B</c> and <c>A</c> are a bodies,
          each <c>Cc_i</c> is a case clause,
          and each <c>Tc_j</c> is a catch clause, then Rep(E) =
          <c>{'try',LINE,Rep(B),[Rep(Cc_1), ..., Rep(Cc_k)],[Rep(Tc_1), ...,
          Rep(Tc_n)],Rep(A)}</c>.</p>
      </item>
      <item>
        <p>If E is a variable <c>V</c>, then Rep(E) = <c>{var,LINE,A}</c>,
          where <c>A</c> is an atom with a printname consisting of the same
          characters as <c>V</c>.</p>
      </item>
    </list>

    <section>
      <title>Qualifiers</title>
      <p>A qualifier Q is one of the following:</p>

      <list type="bulleted">
        <item>
          <p>If Q is a filter <c>E</c>, where <c>E</c> is an expression, then
            Rep(Q) = <c>Rep(E)</c>.</p>
        </item>
        <item>
          <p>If Q is a generator <c>P &lt;- E</c>, where <c>P</c> is
            a pattern and <c>E</c> is an expression, then Rep(Q) =
            <c>{generate,LINE,Rep(P),Rep(E)}</c>.</p>
        </item>
        <item>
          <p>If Q is a bitstring generator <c>P &lt;= E</c>, where <c>P</c> is
            a pattern and <c>E</c> is an expression, then Rep(Q) =
            <c>{b_generate,LINE,Rep(P),Rep(E)}</c>.</p>
        </item>
      </list>
    </section>

    <section>
      <title>Bitstring Element Type Specifiers</title>
      <p>A type specifier list TSL for a bitstring element is a sequence
        of type specifiers <c>TS_1 - ... - TS_k</c>, and
        Rep(TSL) = <c>[Rep(TS_1), ..., Rep(TS_k)]</c>.</p>

      <list type="bulleted">
        <item>
          <p>If TS is a type specifier <c>A</c>, where <c>A</c> is an atom,
            then Rep(TS) = <c>A</c>.</p>
        </item>
        <item>
          <p>If TS is a type specifier <c>A:Value</c>,
            where <c>A</c> is an atom and <c>Value</c> is an integer,
            then Rep(TS) = <c>{A,Value}</c>.</p>
        </item>
      </list>
    </section>

    <section>
      <title>Associations</title>
      <p>An association A is one of the following:</p>

      <list type="bulleted">
         <item>
           <p>If A is an association <c>K => V</c>,
             then Rep(A) = <c>{map_field_assoc,LINE,Rep(K),Rep(V)}</c>.</p>
        </item>
        <item>
          <p>If A is an association <c>K := V</c>,
            then Rep(A) = <c>{map_field_exact,LINE,Rep(K),Rep(V)}</c>.</p>
        </item>
      </list>
    </section>
  </section>

  <section>
    <title>Clauses</title>
    <p>There are function clauses, if clauses, case clauses,
      and catch clauses.</p>

    <p>A clause C is one of the following:</p>

    <list type="bulleted">
      <item>
        <p>If C is a case clause <c>P -> B</c>,
          where <c>P</c> is a pattern and <c>B</c> is a body, then
          Rep(C) = <c>{clause,LINE,[Rep(P)],[],Rep(B)}</c>.</p>
      </item>
      <item>
        <p>If C is a case clause <c>P when Gs -> B</c>,
          where <c>P</c> is a pattern,
          <c>Gs</c> is a guard sequence, and <c>B</c> is a body, then
          Rep(C) = <c>{clause,LINE,[Rep(P)],Rep(Gs),Rep(B)}</c>.</p>
      </item>
      <item>
        <p>If C is a catch clause <c>P -> B</c>,
          where <c>P</c> is a pattern and <c>B</c> is a body, then
          Rep(C) = <c>{clause,LINE,[Rep({throw,P,_})],[],Rep(B)}</c>.</p>
      </item>
      <item>
        <p>If C is a catch clause <c>X : P -> B</c>,
          where <c>X</c> is an atomic literal or a variable pattern,
          <c>P</c> is a pattern, and <c>B</c> is a body, then
          Rep(C) = <c>{clause,LINE,[Rep({X,P,_})],[],Rep(B)}</c>.</p>
      </item>
      <item>
        <p>If C is a catch clause <c>P when Gs -> B</c>,
          where <c>P</c> is a pattern, <c>Gs</c> is a guard sequence,
          and <c>B</c> is a body, then
          Rep(C) = <c>{clause,LINE,[Rep({throw,P,_})],Rep(Gs),Rep(B)}</c>.</p>
      </item>
      <item>
        <p>If C is a catch clause <c>X : P when Gs -> B</c>,
          where <c>X</c> is an atomic literal or a variable pattern,
          <c>P</c> is a pattern, <c>Gs</c> is a guard sequence,
          and <c>B</c> is a body, then
          Rep(C) = <c>{clause,LINE,[Rep({X,P,_})],Rep(Gs),Rep(B)}</c>.</p>
      </item>
      <item>
        <p>If C is a function clause <c>( Ps ) -> B</c>,
          where <c>Ps</c> is a pattern sequence and <c>B</c> is a body, then
          Rep(C) = <c>{clause,LINE,Rep(Ps),[],Rep(B)}</c>.</p>
      </item>
      <item>
        <p>If C is a function clause <c>( Ps ) when Gs -> B</c>,
          where <c>Ps</c> is a pattern sequence,
          <c>Gs</c> is a guard sequence and <c>B</c> is a body, then
          Rep(C) = <c>{clause,LINE,Rep(Ps),Rep(Gs),Rep(B)}</c>.</p>
      </item>
      <item>
        <p>If C is an if clause <c>Gs -> B</c>,
          where <c>Gs</c> is a guard sequence and <c>B</c> is a body, then
          Rep(C) = <c>{clause,LINE,[],Rep(Gs),Rep(B)}</c>.</p>
      </item>
    </list>
  </section>

  <section>
    <title>Guards</title>
    <p>A guard sequence Gs is a sequence of guards <c>G_1; ...; G_k</c>, and
      Rep(Gs) = <c>[Rep(G_1), ..., Rep(G_k)]</c>. If the guard sequence is
      empty, then Rep(Gs) = <c>[]</c>.</p>

    <p>A guard G is a non-empty sequence of guard tests
      <c>Gt_1, ..., Gt_k</c>, and Rep(G) =
      <c>[Rep(Gt_1), ..., Rep(Gt_k)]</c>.</p>

    <p>A guard test Gt is one of the following:</p>

    <list type="bulleted">
      <item>
        <p>If Gt is an atomic literal <c>L</c>, then Rep(Gt) = Rep(L).</p>
      </item>
      <item>
        <p>If Gt is a bitstring constructor
          <c>&lt;&lt;Gt_1:Size_1/TSL_1, ..., Gt_k:Size_k/TSL_k>></c>,
          where each <c>Size_i</c> is a guard test and each
          <c>TSL_i</c> is a type specificer list, then Rep(Gt) =
          <c>{bin,LINE,[{bin_element,LINE,Rep(Gt_1),Rep(Size_1),Rep(TSL_1)},
          ..., {bin_element,LINE,Rep(Gt_k),Rep(Size_k),Rep(TSL_k)}]}</c>.
          For Rep(TSL), see above.
          An omitted <c>Size_i</c> is represented by <c>default</c>.
          An omitted <c>TSL_i</c> is represented by <c>default</c>.</p>
      </item>
      <item>
        <p>If Gt is a cons skeleton <c>[Gt_h | Gt_t]</c>, then Rep(Gt) =
          <c>{cons,LINE,Rep(Gt_h),Rep(Gt_t)}</c>.</p>
      </item>
      <item>
        <p>If Gt is a function call <c>A(Gt_1, ..., Gt_k)</c>,
          where <c>A</c> is an atom, then Rep(Gt) =
          <c>{call,LINE,Rep(A),[Rep(Gt_1), ..., Rep(Gt_k)]}</c>.</p>
      </item>
      <item>
        <p>If Gt is a function call <c>A_m:A(Gt_1, ..., Gt_k)</c>,
          where <c>A_m</c> is the atom <c>erlang</c> and <c>A</c> is
          an atom or an operator, then Rep(Gt) =
          <c>{call,LINE,{remote,LINE,Rep(A_m),Rep(A)},[Rep(Gt_1), ...,
          Rep(Gt_k)]}</c>.</p>
      </item>
      <item>
        <p>If Gt is a map creation <c>#{A_1, ..., A_k}</c>,
          where each <c>A_i</c> is an association <c>Gt_i_1 => Gt_i_2</c>
          or <c>Gt_i_1 := Gt_i_2</c>, then Rep(Gt) =
          <c>{map,LINE,[Rep(A_1), ..., Rep(A_k)]}</c>.
          For Rep(A), see above.</p>
      </item>
      <item>
        <p>If Gt is a map update <c>Gt_0#{A_1, ..., A_k}</c>,
          where each <c>A_i</c> is an association <c>Gt_i_1 => Gt_i_2</c>
          or <c>Gt_i_1 := Gt_i_2</c>, then Rep(Gt) =
          <c>{map,LINE,Rep(Gt_0),[Rep(A_1), ..., Rep(A_k)]}</c>.
          For Rep(A), see above.</p>
      </item>
      <item>
        <p>If Gt is nil, <c>[]</c>, then Rep(Gt) = <c>{nil,LINE}</c>.</p>
      </item>
      <item>
        <p>If Gt is an operator guard test <c>Gt_1 Op Gt_2</c>,
          where <c>Op</c> is a binary operator other than match
          operator <c>=</c>, then Rep(Gt) =
          <c>{op,LINE,Op,Rep(Gt_1),Rep(Gt_2)}</c>.</p>
      </item>
      <item>
        <p>If Gt is an operator guard test <c>Op Gt_0</c>,
          where <c>Op</c> is a unary operator, then Rep(Gt) =
          <c>{op,LINE,Op,Rep(Gt_0)}</c>.</p>
      </item>
      <item>
        <p>If Gt is a parenthesized guard test <c>( Gt_0 )</c>, then Rep(Gt) =
          <c>Rep(Gt_0)</c>, that is, parenthesized
          guard tests cannot be distinguished from their bodies.</p>
      </item>
      <item>
        <p>If Gt is a record creation
          <c>#Name{Field_1=Gt_1, ..., Field_k=Gt_k}</c>,
           where each <c>Field_i</c> is an atom or <c>_</c>, then Rep(Gt) =
           <c>{record,LINE,Name,[{record_field,LINE,Rep(Field_1),Rep(Gt_1)},
           ..., {record_field,LINE,Rep(Field_k),Rep(Gt_k)}]}</c>.</p>
      </item>
      <item>
        <p>If Gt is  a record field access <c>Gt_0#Name.Field</c>,
          where <c>Field</c> is an atom, then Rep(Gt) =
          <c>{record_field,LINE,Rep(Gt_0),Name,Rep(Field)}</c>.</p>
      </item>
      <item>
        <p>If Gt is a record field index <c>#Name.Field</c>,
          where <c>Field</c> is an atom, then Rep(Gt) =
          <c>{record_index,LINE,Name,Rep(Field)}</c>.</p>
      </item>
      <item>
        <p>If Gt is a tuple skeleton <c>{Gt_1, ..., Gt_k}</c>, then Rep(Gt) =
          <c>{tuple,LINE,[Rep(Gt_1), ..., Rep(Gt_k)]}</c>.</p>
      </item>
      <item>
        <p>If Gt is a variable pattern <c>V</c>, then Rep(Gt) =
          <c>{var,LINE,A}</c>, where A is an atom with
          a printname consisting of the same characters as <c>V</c>.</p>
      </item>
    </list>

    <p>Notice that every guard test has the same source form as some expression,
      and is represented in the same way as the corresponding expression.</p>
  </section>

  <section>
    <title>Types</title>
    <list type="bulleted">
      <item>
        <p>If T is an annotated type <c>A :: T_0</c>,
          where <c>A</c> is a variable, then Rep(T) =
          <c>{ann_type,LINE,[Rep(A),Rep(T_0)]}</c>.</p>
      </item>
      <item>
        <p>If T is an atom or integer literal L, then Rep(T) = Rep(L).</p>
      </item>
      <item>
        <p>If T is a bitstring type <c>&lt;&lt;_:M,_:_*N>></c>,
          where <c>M</c> and <c>N</c> are singleton integer types, then Rep(T) =
          <c>{type,LINE,binary,[Rep(M),Rep(N)]}</c>.</p>
      </item>
      <item>
        <p>If T is the empty list type <c>[]</c>, then Rep(T) =
          <c>{type,Line,nil,[]}</c>.</p>
      </item>
      <item>
        <p>If T is a fun type <c>fun()</c>, then Rep(T) =
          <c>{type,LINE,'fun',[]}</c>.</p>
      </item>
      <item>
        <p>If T is a fun type <c>fun((...) -> T_0)</c>, then Rep(T) =
          <c>{type,LINE,'fun',[{type,LINE,any},Rep(T_0)]}</c>.</p>
      </item>
      <item>
        <p>If T is a fun type <c>fun(Ft)</c>, where
          <c>Ft</c> is a function type, then Rep(T) = <c>Rep(Ft)</c>.
          For Rep(Ft), see below.</p>
      </item>
      <item>
        <p>If T is an integer range type <c>L .. H</c>,
          where <c>L</c> and <c>H</c> are singleton integer types, then Rep(T) =
          <c>{type,LINE,range,[Rep(L),Rep(H)]}</c>.</p>
      </item>
      <item>
        <p>If T is a map type <c>map()</c>, then Rep(T) =
          <c>{type,LINE,map,any}</c>.</p>
      </item>
      <item>
        <p>If T is a map type <c>#{A_1, ..., A_k}</c>, where each
          <c>A_i</c> is an association type, then Rep(T) =
          <c>{type,LINE,map,[Rep(A_1), ..., Rep(A_k)]}</c>.
          For Rep(A), see below.</p>
      </item>
      <item>
        <p>If T is an operator type <c>T_1 Op T_2</c>,
          where <c>Op</c> is a binary operator (this is an occurrence of
          an expression that can be evaluated to an integer at compile
          time), then Rep(T) =
          <c>{op,LINE,Op,Rep(T_1),Rep(T_2)}</c>.</p>
      </item>
      <item>
        <p>If T is an operator type <c>Op T_0</c>, where <c>Op</c> is a
          unary operator (this is an occurrence of an expression that can
          be evaluated to an integer at compile time), then Rep(T) =
          <c>{op,LINE,Op,Rep(T_0)}</c>.</p>
      </item>
      <item>
        <p>If T is <c>( T_0 )</c>, then Rep(T) = <c>Rep(T_0)</c>, that is,
          parenthesized types cannot be distinguished from their bodies.</p>
      </item>
      <item>
        <p>If T is a predefined (or built-in) type <c>N(T_1, ..., T_k)</c>,
          then Rep(T) = <c>{type,LINE,N,[Rep(T_1), ..., Rep(T_k)]}</c>.</p>
      </item>
      <item>
        <p>If T is a record type <c>#Name{F_1, ..., F_k}</c>,
          where each <c>F_i</c> is a record field type, then Rep(T) =
          <c>{type,LINE,record,[Rep(Name),Rep(F_1), ..., Rep(F_k)]}</c>.
          For Rep(F), see below.</p>
      </item>
      <item>
        <p>If T is a remote type <c>M:N(T_1, ..., T_k)</c>, then Rep(T) =
          <c>{remote_type,LINE,[Rep(M),Rep(N),[Rep(T_1), ...,
          Rep(T_k)]]}</c>.</p>
      </item>
      <item>
        <p>If T is a tuple type <c>tuple()</c>, then Rep(T) =
          <c>{type,LINE,tuple,any}</c>.</p>
      </item>
      <item>
        <p>If T is a tuple type <c>{T_1, ..., T_k}</c>, then Rep(T) =
          <c>{type,LINE,tuple,[Rep(T_1), ..., Rep(T_k)]}</c>.</p>
      </item>
      <item>
        <p>If T is a type union <c>T_1 | ... | T_k</c>, then Rep(T) =
          <c>{type,LINE,union,[Rep(T_1), ..., Rep(T_k)]}</c>.</p>
      </item>
      <item>
        <p>If T is a type variable <c>V</c>, then Rep(T) =
          <c>{var,LINE,A}</c>, where <c>A</c> is an atom with a printname
          consisting of the same characters as <c>V</c>. A type variable
          is any variable except underscore (<c>_</c>).</p>
      </item>
      <item>
        <p>If T is a user-defined type <c>N(T_1, ..., T_k)</c>, then Rep(T) =
          <c>{user_type,LINE,N,[Rep(T_1), ..., Rep(T_k)]}</c>.</p>
      </item>
    </list>

    <section>
      <title>Function Types</title>
      <p>A function type Ft is one of the following:</p>

      <list type="bulleted">
        <item>
          <p>If Ft is a constrained function type <c>Ft_1 when Fc</c>,
            where <c>Ft_1</c> is a function type and
            <c>Fc</c> is a function constraint, then Rep(T) =
            <c>{type,LINE,bounded_fun,[Rep(Ft_1),Rep(Fc)]}</c>.
            For Rep(Fc), see below.</p>
        </item>
        <item>
          <p>If Ft is a function type <c>(T_1, ..., T_n) -> T_0</c>,
            where each <c>T_i</c> is a type, then Rep(Ft) =
            <c>{type,LINE,'fun',[{type,LINE,product,[Rep(T_1), ...,
            Rep(T_n)]},Rep(T_0)]}</c>.</p>
        </item>
      </list>
    </section>

    <section>
      <title>Function Constraints</title>
      <p>A function constraint Fc is a non-empty sequence of constraints
        <c>C_1, ..., C_k</c>, and
        Rep(Fc) = <c>[Rep(C_1), ..., Rep(C_k)]</c>.</p>

      <list type="bulleted">
        <item>If C is a constraint <c>V :: T</c>,
          where <c>V</c> is a type variable
          and <c>T</c> is a type, then Rep(C) =
          <c>{type,LINE,constraint,[{atom,LINE,is_subtype},[Rep(V),Rep(T)]]}</c>.
        </item>
      </list>
    </section>

    <section>
      <title>Association Types</title>
      <list type="bulleted">
        <item>
          <p>If A is an association type <c>K => V</c>,
            where <c>K</c> and <c>V</c> are types, then Rep(A) =
            <c>{type,LINE,map_field_assoc,[Rep(K),Rep(V)]}</c>.</p>
        </item>
        <item>
          <p>If A is an association type <c>K := V</c>,
            where <c>K</c> and <c>V</c> are types, then Rep(A) =
            <c>{type,LINE,map_field_exact,[Rep(K),Rep(V)]}</c>.</p>
        </item>
      </list>
    </section>

    <section>
      <title>Record Field Types</title>
      <list type="bulleted">
        <item>If F is a record field type <c>Name :: Type</c>,
          where <c>Type</c> is a type, then Rep(F) =
          <c>{type,LINE,field_type,[Rep(Name),Rep(Type)]}</c>.
        </item>
      </list>
    </section>
  </section>

  <section>
    <title>The Abstract Format after Preprocessing</title>
    <p>The compilation option <c>debug_info</c> can be specified to the
      compiler to have the abstract code stored in
      the <c>abstract_code</c> chunk in the Beam file
      (for debugging purposes).</p>

    <p>As from Erlang/OTP R9C, the <c>abstract_code</c> chunk contains
      <c>{raw_abstract_v1,AbstractCode}</c>, where <c>AbstractCode</c> is the
      abstract code as described in this section.</p>

    <p>In OTP releases before R9C, the abstract code after some more
      processing was stored in the Beam file. The first element of the
      tuple would be either <c>abstract_v1</c> (in OTP R7B) or
      <c>abstract_v2</c> (in OTP R8B).</p>
  </section>
</chapter>