path: root/build2/parser

// file      : build2/parser -*- C++ -*-
// copyright : Copyright (c) 2014-2016 Code Synthesis Ltd
// license   : MIT; see accompanying LICENSE file

#ifndef BUILD2_PARSER
#define BUILD2_PARSER

#include <stack>

#include <build2/types>
#include <build2/utility>

#include <build2/spec>
#include <build2/lexer>
#include <build2/token>
#include <build2/variable>
#include <build2/diagnostics>

namespace build2
{
  class scope;
  class target;

  class parser
  {
  public:
    using names_type = build2::names;
    using variable_type = build2::variable;

    // If boot is true, then we are parsing bootstrap.build and modules
    // should only be bootstrapped.
    //
    parser (bool boot = false): fail (&path_), boot_ (boot) {}

    // Issue diagnostics and throw failed in case of an error.
    //
    void
    parse_buildfile (istream&, const path& name, scope& root, scope& base);

    buildspec
    parse_buildspec (istream&, const path& name);

    token
    parse_variable (lexer&, scope&, const variable_type&, token_type kind);

    pair<value, token>
    parse_variable_value (lexer&, scope&, const variable_type&);

    names_type
    parse_export_stub (istream& is, const path& p, scope& r, scope& b)
    {
      parse_buildfile (is, p, r, b);
      return move (export_value_);
    }

    // Recursive descent parser.
    //
  protected:
    void
    clause (token&, token_type&);

    void
    print (token&, token_type&);

    void
    source (token&, token_type&);

    void
    include (token&, token_type&);

    void
    import (token&, token_type&);

    void
    export_ (token&, token_type&);

    void
    using_ (token&, token_type&);

    void
    define (token&, token_type&);

    void
    if_else (token&, token_type&);

    void
    variable (token&, token_type&, const variable_type&, token_type kind);

    string
    variable_name (names_type&&, const location&);

    // Note: calls attributes_push() that the caller must pop.
    //
    value
    variable_value (token&, token_type&);

    void
    variable_attributes (const variable_type&);

    void
    value_attributes (const variable_type*, // Optional.
                      value& lhs,
                      value&& rhs,
                      token_type kind);

    value
    eval (token&, token_type&);

    value
    eval_trailer (token&, token_type&);

    // Attributes stack. We can have nested attributes, for example:
    //
    // x = [bool] ([uint64] $x == [uint64] $y)
    //
    // In this example we only apply the value attributes after evaluating
    // the context, which has its own attributes.
    //
    struct attributes
    {
      bool has;                         // Has attributes flag.
      location loc;                     // Start of attributes location.
      vector<pair<string, string>> ats; // Attributes.

      explicit operator bool () const {return has;}
    };

    // Push a new entry into the attributes_ stack. If the next token is '['
    // parse the attribute sequence until ']' setting the 'has' flag and
    // storing the result on the stack. Then get the next token and, if
    // standalone is false, verify it is not newline/eos (i.e., there is
    // something after it). Return the top of the stack.
    //
    attributes&
    attributes_push (token&, token_type&, bool standalone = false);

    attributes
    attributes_pop ()
    {
      attributes r (move (attributes_.top ()));
      attributes_.pop ();
      return r;
    }

    attributes&
    attributes_top () {return attributes_.top ();}

    // If chunk is true, then parse the smallest but complete, name-wise,
    // chunk of input. Note that in this case you may still end up with
    // multiple names, for example, {foo bar}.
    //
    names_type
    names (token& t, token_type& tt, bool chunk = false)
    {
      names_type ns;
      names (t, tt, ns, chunk, 0, nullptr, nullptr, nullptr);
      return ns;
    }

    // As above but return the result as a value (always untyped) indicating
    // if it is NULL. The only way for the value to be NULL is if it is the
    // result of a sole, unquoted variable expansion, function call, or
    // context evaluation.
    //
    value
    names_value (token& t, token_type& tt)
    {
      names_type ns;
      return names (t, tt, ns, false, 0, nullptr, nullptr, nullptr)
        ? value (move (ns))
        : value (nullptr);
    }

    // Return true if the parsed value is NOT NULL.
    //
    bool
    names (token& t, token_type& tt, names_type& ns, bool chunk = false)
    {
      return names (t, tt, ns, chunk, 0, nullptr, nullptr, nullptr);
    }

    bool
    names (token&, token_type&,
           names_type&,
           bool chunk,
           size_t pair,
           const string* prj,
           const dir_path* dir,
           const string* type);

    size_t
    names_trailer (token&, token_type&,
                   names_type&,
                   size_t pair,
                   const string* prj,
                   const dir_path* dir,
                   const string* type);

    // Skip until newline or eos.
    //
    void
    skip_line (token&, token_type&);

    // Skip until block-closing } or eos, taking into account nested blocks.
    //
    void
    skip_block (token&, token_type&);

    // Return true if the name token can be considered a directive keyword.
    //
    bool
    keyword (token&);

    // Buildspec.
    //
    buildspec
    buildspec_clause (token&, token_type&, token_type end);

    // Utilities.
    //
  protected:
    class enter_scope;
    class enter_target;

    // Switch to a new current scope. Note that this function might also have
    // to switch to a new root scope if the new current scope is in another
    // project. So both must be saved and restored.
    //
    void
    switch_scope (const dir_path&);

    void
    process_default_target (token&);

    // Enter buildfile as a target.
    //
    void
    enter_buildfile (const path&);

    // Lexer.
    //
  protected:
    token_type
    next (token&, token_type&);

    // Be careful with peeking and switching the lexer mode. See keyword()
    // for more information.
    //
    token_type
    peek ();

    const token&
    peeked () const
    {
      assert (peeked_);
      return peek_;
    }

    void
    mode (lexer_mode m, char ps = '=')
    {
      if (replay_ != replay::play)
        lexer_->mode (m, ps);
    }

    lexer_mode
    mode () const
    {
      assert (replay_ != replay::play);
      return lexer_->mode ();
    }

    void
    expire_mode ()
    {
      if (replay_ != replay::play)
        lexer_->expire_mode ();
    }

    // Token saving and replaying. Note that it can only be used in certain
    // contexts. Specifically, the lexer mode should be the same and the code
    // that parses a replay must not interact with the lexer directly (e.g.,
    // the keyword() test). For now we don't enforce any of this.
    //
    // Note also that the peeked token is not part of the replay, until it
    // is "got".
    //
    //
    void
    replay_save ()
    {
      assert (replay_ == replay::stop);
      replay_ = replay::save;
    }

    void
    replay_play ()
    {
      assert ((replay_ == replay::save && !replay_data_.empty ()) ||
              (replay_ == replay::play && replay_i_ == replay_data_.size ()));

      replay_i_ = 0;
      replay_ = replay::play;
    }

    void
    replay_stop ()
    {
      replay_data_.clear ();
      replay_ = replay::stop;
    }

    const token&
    replay_next ()
    {
      assert (replay_i_ != replay_data_.size ());
      return replay_data_[replay_i_++];
    }

    struct replay_guard
    {
      replay_guard (parser& p, bool start = true)
          : p_ (start ? &p : nullptr)
      {
        if (p_ != nullptr)
          p_->replay_save ();
      }

      void
      play ()
      {
        if (p_ != nullptr)
          p_->replay_play ();
      }

      ~replay_guard ()
      {
        if (p_ != nullptr)
          p_->replay_stop ();
      }

    private:
      parser* p_;
    };

    // Diagnostics.
    //
  protected:
    const fail_mark<failed> fail;

  protected:
    bool boot_;

    const path* path_; // Current path.
    lexer* lexer_;
    target* target_; // Current target, if any.
    scope* scope_;   // Current base scope (out_base).
    scope* root_;    // Current root scope (out_root).

    std::stack<attributes> attributes_;

    target* default_target_;
    names_type export_value_;

    token peek_ = token (token_type::eos, false, 0, 0);
    bool peeked_ = false;

    enum class replay {stop, save, play} replay_ = replay::stop;
    vector<token> replay_data_;
    size_t replay_i_; // Position of the next token during replay.
  };
}

#endif // BUILD2_PARSER