Add support for package dependency and requirement alternatives representation new syntax

5 files changed, 2014 insertions, 111 deletions

diff --git a/libbpkg/buildfile-scanner.cxx b/libbpkg/buildfile-scanner.cxx
new file mode 100644
index 0000000..ec73d83
--- /dev/null
+++ b/libbpkg/buildfile-scanner.cxx
@@ -0,0 +1,38 @@
+// file      : libbpkg/buildfile-scanner.cxx -*- C++ -*-
+// license   : MIT; see accompanying LICENSE file
+
+#include <libbpkg/buildfile-scanner.hxx>
+
+#include <string>
+
+using namespace std;
+
+namespace bpkg
+{
+  // buildfile_scanning
+  //
+  static inline string
+  format (const string& n, uint64_t l, uint64_t c, const string& d)
+  {
+    string r;
+    if (!n.empty ())
+    {
+      r += n;
+      r += ':';
+    }
+
+    r += to_string (l);
+    r += ':';
+    r += to_string (c);
+    r += ": error: ";
+    r += d;
+    return r;
+  }
+
+  buildfile_scanning::
+  buildfile_scanning (const string& n, uint64_t l, uint64_t c, const string& d)
+      : runtime_error (format (n, l, c, d)),
+        name (n), line (l), column (c), description (d)
+  {
+  }
+}
diff --git a/libbpkg/buildfile-scanner.hxx b/libbpkg/buildfile-scanner.hxx
new file mode 100644
index 0000000..8b5f0d7
--- /dev/null
+++ b/libbpkg/buildfile-scanner.hxx
@@ -0,0 +1,96 @@
+// file      : libbpkg/buildfile-scanner.hxx -*- C++ -*-
+// license   : MIT; see accompanying LICENSE file
+
+#ifndef LIBBPKG_BUILDFILE_SCANNER_HXX
+#define LIBBPKG_BUILDFILE_SCANNER_HXX
+
+#include <string>
+#include <cstdint>   // uint64_t
+#include <cstddef>   // size_t
+#include <stdexcept> // runtime_error
+
+#include <libbutl/char-scanner.hxx>
+
+#include <libbpkg/export.hxx>
+
+namespace bpkg
+{
+  // Scan buildfile fragments, respecting the single- and double-quoted
+  // character sequences, backslash-escaping, comments, evaluation
+  // contexts, and nested blocks.
+  //
+  class LIBBPKG_EXPORT buildfile_scanning: public std::runtime_error
+  {
+  public:
+    buildfile_scanning (const std::string& name,
+                        std::uint64_t line,
+                        std::uint64_t column,
+                        const std::string& description);
+
+    std::string name;
+    std::uint64_t line;
+    std::uint64_t column;
+    std::string description;
+  };
+
+  template <typename V, std::size_t N>
+  class buildfile_scanner
+  {
+  public:
+    // Note that name is stored by shallow reference.
+    //
+    buildfile_scanner (butl::char_scanner<V, N>& s, const std::string& name)
+        : scan_ (s), name_ (name) {}
+
+    // Scan the buildfile line and return the scanned fragment. Optionally,
+    // specify an additional stop character. Leave the newline (or the stop
+    // character) in the stream. Throw buildfile_scanning on error
+    // (unterminated quoted sequence, etc).
+    //
+    std::string
+    scan_line (char stop = '\0');
+
+    // Scan the buildfile line until an unbalanced ')' character is encountered
+    // and return the scanned fragment, leaving ')' in the stream. Throw
+    // buildfile_scanning on error or if eos or newline is reached.
+    //
+    std::string
+    scan_eval ();
+
+    // Scan the buildfile block until an unbalanced block closing '}' character
+    // is encountered and return the scanned fragment. Throw buildfile_scanning
+    // on error or if eos is reached.
+    //
+    // Note that the block opening '{' and closing '}' characters are only
+    // considered as such, if they are the only characters on the line besides
+    // whitespaces and comments. Also note that the fragment terminating '}'
+    // line is consumed from the stream but is not included into the fragment.
+    //
+    std::string
+    scan_block ();
+
+  private:
+    using scanner = butl::char_scanner<V, N>;
+    using xchar = typename scanner::xchar;
+
+    xchar
+    peek ();
+
+    // Scan the buildfile line, saving the scanned characters into the
+    // specified string, leaving newline and the stop character, if specified,
+    // in the stream. Return '{' if this line is a block-opening curly brace,
+    // '}' if it is a block-closing curly brace, and '\0' otherwise.
+    //
+    char
+    scan_line (std::string& l, char stop = '\0');
+
+  private:
+    scanner&           scan_;
+    const std::string& name_;
+    std::string        ebuf_; // Error message buffer.
+  };
+}
+
+#include <libbpkg/buildfile-scanner.txx>
+
+#endif // LIBBPKG_BUILDFILE_SCANNER_HXX
diff --git a/libbpkg/buildfile-scanner.txx b/libbpkg/buildfile-scanner.txx
new file mode 100644
index 0000000..acd9037
--- /dev/null
+++ b/libbpkg/buildfile-scanner.txx
@@ -0,0 +1,272 @@
+// file      : libbpkg/buildfile-scanner.txx -*- C++ -*-
+// license   : MIT; see accompanying LICENSE file
+
+#include <cassert>
+
+#include <libbutl/optional.hxx>
+
+namespace bpkg
+{
+  template <typename V, std::size_t N>
+  typename buildfile_scanner<V, N>::xchar buildfile_scanner<V, N>::
+  peek ()
+  {
+    xchar c (scan_.peek (ebuf_));
+
+    if (scanner::invalid (c))
+      throw buildfile_scanning (name_, scan_.line, scan_.column, ebuf_);
+
+    return c;
+  }
+
+  template <typename V, std::size_t N>
+  char buildfile_scanner<V, N>::
+  scan_line (std::string& l, char stop)
+  {
+    using namespace std;
+
+    auto fail = [this] (const string& d)
+    {
+      throw buildfile_scanning (name_, scan_.line, scan_.column, d);
+    };
+
+    xchar c (peek ());
+
+    auto next = [&l, &c, this] ()
+    {
+      l += c;
+      scan_.get (c);
+    };
+
+    butl::optional<char> r;
+    bool double_quoted (false);
+
+    for (;
+         !scanner::eos (c) && (double_quoted || (c != '\n' && c != stop));
+         c = peek ())
+    {
+      switch (c)
+      {
+      case '\"':
+        {
+          // Start or finish scanning the double-quoted sequence.
+          //
+          double_quoted = !double_quoted;
+
+          r = '\0';
+          break;
+        }
+      case '\\':
+        {
+          next ();
+
+          c = peek ();
+
+          if (scanner::eos (c))
+            fail (double_quoted
+                  ? "unterminated double-quoted sequence"
+                  : "unterminated escape sequence");
+
+          r = '\0';
+          break;
+        }
+      case '(':
+        {
+          next ();
+
+          scan_line (l, ')');
+
+          c = peek ();
+
+          if (c != ')')
+            fail ("unterminated evaluation context");
+
+          next ();
+
+          r = '\0';
+          continue;
+        }
+      case '\'':
+        {
+          if (!double_quoted)
+          {
+            next ();
+
+            for (;;)
+            {
+              c = peek ();
+
+              if (scanner::eos (c))
+                fail ("unterminated single-quoted sequence");
+
+              next ();
+
+              if (c == '\'')
+                break;
+            }
+
+            r = '\0';
+            continue;
+          }
+
+          break;
+        }
+      case '#':
+        {
+          if (!double_quoted)
+          {
+            next ();
+
+            // See if this is a multi-line comment in the form:
+            //
+            /*
+              #\
+              ...
+              #\
+            */
+            auto ml = [&c, &next, this] () -> bool
+            {
+              if ((c = peek ()) == '\\')
+              {
+                next ();
+
+                if ((c = peek ()) == '\n' || scanner::eos (c))
+                  return true;
+              }
+
+              return false;
+            };
+
+            if (ml ())
+            {
+              // Scan until we see the closing one.
+              //
+              for (;;)
+              {
+                if (c == '#' && ml ())
+                  break;
+
+                if (scanner::eos (c = peek ()))
+                  fail ("unterminated multi-line comment");
+
+                next ();
+              }
+            }
+            else
+            {
+              // Read until newline or eos.
+              //
+              for (; !scanner::eos (c) && c != '\n'; c = peek ())
+                next ();
+            }
+
+            continue;
+          }
+
+          break;
+        }
+      case '{':
+      case '}':
+        {
+          if (!double_quoted)
+            r = !r ? static_cast<char> (c) : '\0';
+
+          break;
+        }
+      default:
+        {
+          if (!double_quoted && c != ' ' && c != '\t')
+            r = '\0';
+
+          break;
+        }
+      }
+
+      next ();
+    }
+
+    if (double_quoted)
+      fail ("unterminated double-quoted sequence");
+
+    return r ? *r : '\0';
+  }
+
+  template <typename V, std::size_t N>
+  std::string buildfile_scanner<V, N>::
+  scan_line (char stop)
+  {
+    std::string r;
+    scan_line (r, stop);
+    return r;
+  }
+
+  template <typename V, std::size_t N>
+  std::string buildfile_scanner<V, N>::
+  scan_eval ()
+  {
+    std::string r;
+    scan_line (r, ')');
+
+    if (peek () != ')')
+      throw buildfile_scanning (name_,
+                                scan_.line,
+                                scan_.column,
+                                "unterminated evaluation context");
+
+    return r;
+  }
+
+  template <typename V, std::size_t N>
+  std::string buildfile_scanner<V, N>::
+  scan_block ()
+  {
+    using namespace std;
+
+    auto fail = [this] (const string& d)
+    {
+      throw buildfile_scanning (name_, scan_.line, scan_.column, d);
+    };
+
+    string r;
+    for (size_t level (0);; )
+    {
+      if (scanner::eos (peek ()))
+        fail ("unterminated buildfile block");
+
+      size_t n (r.size ());
+      char bc (scan_line (r));
+
+      xchar c (peek ());
+
+      // Append the newline unless this is eos.
+      //
+      if (c == '\n')
+      {
+        r += c;
+        scan_.get (c);
+      }
+      else
+        assert (scanner::eos (c));
+
+      if (bc == '{')
+      {
+        ++level;
+      }
+      else if (bc == '}')
+      {
+        // If this is the fragment terminating line, then strip it from the
+        // fragment and bail out.
+        //
+        if (level == 0)
+        {
+          r.resize (n);
+          break;
+        }
+        else
+          --level;
+      }
+    }
+
+    return r;
+  }
+}
diff --git a/libbpkg/manifest.cxx b/libbpkg/manifest.cxx
index bbe8749..6a5ce25 100644
--- a/libbpkg/manifest.cxx
+++ b/libbpkg/manifest.cxx
@@ -18,15 +18,19 @@
 
 #include <libbutl/url.hxx>
 #include <libbutl/path.hxx>
+#include <libbutl/utf8.hxx>
 #include <libbutl/base64.hxx>
 #include <libbutl/utility.hxx>             // icasecmp(), lcase(), alnum(),
                                            // digit(), xdigit(), next_word()
 #include <libbutl/filesystem.hxx>          // dir_exist()
 #include <libbutl/small-vector.hxx>
+#include <libbutl/char-scanner.hxx>
 #include <libbutl/manifest-parser.hxx>
 #include <libbutl/manifest-serializer.hxx>
 #include <libbutl/standard-version.hxx>
 
+#include <libbpkg/buildfile-scanner.hxx>
+
 using namespace std;
 using namespace butl;
 
@@ -1123,141 +1127,1434 @@ namespace bpkg
 
   // dependency_alternative
   //
-  dependency_alternative::
-  dependency_alternative (const std::string& s)
+  string dependency_alternative::
+  string () const
   {
-    push_back (dependency (s)); // @@ DEP
+    std::string r (size () > 1 ? "{" : "");
+
+    bool first (true);
+    for (const dependency& d: *this)
+    {
+      if (!first)
+        r += ' ';
+      else
+        first = false;
+
+      r += d.string ();
+    }
+
+    if (size () > 1)
+      r += '}';
+
+    if (single_line ())
+    {
+      if (enable)
+      {
+        r += " ? (";
+        r += *enable;
+        r += ')';
+      }
+
+      if (reflect)
+      {
+        r += ' ';
+        r += *reflect;
+      }
+    }
+    else
+    {
+      // Add an extra newline between the clauses.
+      //
+      bool first (true);
+
+      r += "\n{";
+
+      if (enable)
+      {
+        first = false;
+
+        r += "\n  enable (";
+        r += *enable;
+        r += ')';
+      }
+
+      if (prefer)
+      {
+        if (!first)
+          r += '\n';
+        else
+          first = false;
+
+        r += "\n  prefer\n  {\n";
+        r += *prefer;
+        r += "  }";
+
+        assert (accept);
+
+        r += "\n\n  accept (";
+        r += *accept;
+        r += ')';
+      }
+      else if (require)
+      {
+        if (!first)
+          r += '\n';
+        else
+          first = false;
+
+        r += "\n  require\n  {\n";
+        r += *require;
+        r += "  }";
+      }
+
+      if (reflect)
+      {
+        if (!first)
+          r += '\n';
+        else
+          first = false;
+
+        r += "\n  reflect\n  {\n";
+        r += *reflect;
+        r += "  }";
+      }
+
+      r += "\n}";
+    }
+
+    return r;
   }
 
-  string dependency_alternative::
-  string () const
+  bool dependency_alternative::
+  single_line () const
   {
-    assert (size () == 1); // @@ DEP
-    return front ().string ();
+    return !prefer  &&
+           !require &&
+           (!reflect || reflect->find ('\n') == string::npos);
   }
 
   // dependency_alternatives
   //
-  dependency_alternatives::
-  dependency_alternatives (const std::string& s)
+  class dependency_alternatives_lexer: public char_scanner<utf8_validator>
   {
-    using std::string;
+  public:
+    enum class token_type
+    {
+      eos,
+      newline,
+      word,
+      buildfile,
+
+      question,        // ?
+
+      lcbrace,         // {
+      rcbrace,         // }
+
+      lparen,          // (
+      rparen,          // )
+
+      lsbrace,         // [
+      rsbrace,         // ]
 
-    // Allow specifying ?* in any order.
+      equal,           // ==
+      not_equal,       // !=
+      less,            // <
+      greater,         // >
+      less_equal,      // <=
+      greater_equal,   // >=
+
+      tilde,           // ~
+      caret,           // ^
+
+      bit_or           // |
+    };
+
+    struct token
+    {
+      token_type type;
+      std::string value;
+
+      uint64_t line;
+      uint64_t column;
+
+      std::string
+      string (bool diag = true) const;
+    };
+
+  public:
+    // Note that name is stored by shallow reference.
     //
-    size_t n (s.size ());
-    size_t cond ((n > 0 && s[0] == '?') || (n > 1 && s[1] == '?') ? 1 : 0);
-    size_t btim ((n > 0 && s[0] == '*') || (n > 1 && s[1] == '*') ? 1 : 0);
+    dependency_alternatives_lexer (istream& is,
+                                   const string& name,
+                                   uint64_t line,
+                                   uint64_t column)
+        : char_scanner (is,
+                        utf8_validator (codepoint_types::graphic, U"\n\r\t"),
+                        true /* crlf */,
+                        line,
+                        column),
+          name_ (name),
+          buildfile_scan_ (*this, name_) {}
+
+    // The following functions throw manifest_parsing on invalid UTF-8
+    // sequence.
+    //
+
+    // Peek the next non-whitespace character.
+    //
+    xchar
+    peek_char ();
+
+    // Extract next token (other than of the buildfile type) from the stream.
+    //
+    // Note that it is ok to call next() again after getting eos.
+    //
+    token
+    next ();
+
+    // The next_*() functions extract the buildfile token from the stream.
+    // Throw manifest_parsing on error (invalid buildfile fragment, etc).
+    //
+    // Note that they are just thin wrappers around the scan_*() functions
+    // (see buildfile-scanner.hxx for details).
+    //
+    token
+    next_eval ();
+
+    token
+    next_line (char stop);
+
+    token
+    next_block ();
+
+  private:
+    using base = char_scanner<utf8_validator>;
+
+    xchar
+    get ()
+    {
+      xchar c (base::get (ebuf_));
+
+      if (invalid (c))
+        throw parsing (name_, c.line, c.column, ebuf_);
+
+      return c;
+    }
+
+    void
+    get (const xchar& peeked)
+    {
+      base::get (peeked);
+    }
+
+    xchar
+    peek ()
+    {
+      xchar c (base::peek (ebuf_));
+
+      if (invalid (c))
+        throw parsing (name_, c.line, c.column, ebuf_);
+
+      return c;
+    }
+
+    void
+    skip_spaces ();
+
+  private:
+    const string& name_;
+
+    // Buffer for a get()/peek() potential error.
+    //
+    string ebuf_;
+
+    buildfile_scanner<utf8_validator, 1> buildfile_scan_;
+  };
+
+  dependency_alternatives_lexer::token dependency_alternatives_lexer::
+  next ()
+  {
+    using type = token_type;
+
+    skip_spaces ();
+
+    uint64_t ln (line);
+    uint64_t cl (column);
+
+    xchar c (get ());
+
+    auto make_token = [ln, cl] (type t, string v = string ())
+    {
+      return token {t, move (v), ln, cl};
+    };
+
+    if (eos (c))
+      return make_token (type::eos);
+
+    // NOTE: don't forget to also update the below separators list if changing
+    // anything here.
+    //
+    switch (c)
+    {
+    case '\n': return make_token (type::newline);
+    case '?':  return make_token (type::question);
+    case '(':  return make_token (type::lparen);
+    case ')':  return make_token (type::rparen);
+    case '{':  return make_token (type::lcbrace);
+    case '}':  return make_token (type::rcbrace);
+    case '[':  return make_token (type::lsbrace);
+    case ']':  return make_token (type::rsbrace);
+
+    case '=':
+    case '!':
+      {
+        if ((peek ()) == '=')
+        {
+          get ();
+          return make_token (c == '=' ? type::equal : type::not_equal);
+        }
+        break;
+      }
+
+    case '<':
+      {
+        if ((c = peek ()) == '=')
+        {
+          get (c);
+          return make_token (type::less_equal);
+        }
+        else
+          return make_token (type::less);
+      }
+
+    case '>':
+      {
+        if ((c = peek ()) == '=')
+        {
+          get (c);
+          return make_token (type::greater_equal);
+        }
+        else
+          return make_token (type::greater);
+      }
+
+    case '~':  return make_token (type::tilde);
+    case '^':  return make_token (type::caret);
+
+    case '|': return make_token (type::bit_or);
+    }
+
+    // Otherwise it is a word.
+    //
+    // Starts with a non-whitespace character which has not been recognized as
+    // a part of some other token.
+    //
+    string r (1, c);
+
+    // Add subsequent characters until eos or separator is encountered.
+    //
+    const char* s (" \n\t?(){}[]=!<>~^|");
+    for (c = peek (); !eos (c) && strchr (s, c) == nullptr; c = peek ())
+    {
+      r += c;
+      get (c);
+    }
+
+    return make_token (type::word, move (r));
+  }
+
+  dependency_alternatives_lexer::token dependency_alternatives_lexer::
+  next_eval ()
+  {
+    skip_spaces ();
+
+    uint64_t ln (line);
+    uint64_t cl (column);
+
+    try
+    {
+      // Strip the trailing whitespaces.
+      //
+      return token {token_type::buildfile,
+                    trim (buildfile_scan_.scan_eval ()),
+                    ln,
+                    cl};
+    }
+    catch (const buildfile_scanning& e)
+    {
+      throw parsing (e.name, e.line, e.column, e.description);
+    }
+  }
+
+  dependency_alternatives_lexer::token dependency_alternatives_lexer::
+  next_line (char stop)
+  {
+    skip_spaces ();
+
+    uint64_t ln (line);
+    uint64_t cl (column);
+
+    try
+    {
+      // Strip the trailing whitespaces.
+      //
+      return token {token_type::buildfile,
+                    trim (buildfile_scan_.scan_line (stop)),
+                    ln,
+                    cl};
+    }
+    catch (const buildfile_scanning& e)
+    {
+      throw parsing (e.name, e.line, e.column, e.description);
+    }
+  }
+
+  dependency_alternatives_lexer::token dependency_alternatives_lexer::
+  next_block ()
+  {
+    uint64_t ln (line);
+    uint64_t cl (column);
+
+    try
+    {
+      // Don't trim the token value not to strip the potential block indenting
+      // on the first line.
+      //
+      return token {token_type::buildfile,
+                    buildfile_scan_.scan_block (),
+                    ln,
+                    cl};
+    }
+    catch (const buildfile_scanning& e)
+    {
+      throw parsing (e.name, e.line, e.column, e.description);
+    }
+  }
+
+  dependency_alternatives_lexer::xchar dependency_alternatives_lexer::
+  peek_char ()
+  {
+    skip_spaces ();
+    return peek ();
+  }
+
+  void dependency_alternatives_lexer::
+  skip_spaces ()
+  {
+    xchar c (peek ());
+    bool start (c.column == 1);
+
+    for (; !eos (c); c = peek ())
+    {
+      switch (c)
+      {
+      case ' ':
+      case '\t': break;
+
+      case '\n':
+        {
+          // Skip empty lines.
+          //
+          if (start)
+            break;
+        }
+        // Fall through.
+      default: return;
+      }
+
+      get (c);
+    }
+  }
+
+  std::string dependency_alternatives_lexer::token::
+  string (bool diag) const
+  {
+    std::string q (diag ? "'" : "");
+
+    switch (type)
+    {
+    case token_type::eos:           return diag ? "<end of stream>" : "";
+    case token_type::newline:       return diag ? "<newline>" : "\n";
+    case token_type::word:          return q + value + q;
+    case token_type::buildfile:     return (diag
+                                            ? "<buildfile fragment>"
+                                            : value);
+    case token_type::question:      return q + "?" + q;
+    case token_type::lparen:        return q + "(" + q;
+    case token_type::rparen:        return q + ")" + q;
+    case token_type::lcbrace:       return q + "{" + q;
+    case token_type::rcbrace:       return q + "}" + q;
+    case token_type::lsbrace:       return q + "[" + q;
+    case token_type::rsbrace:       return q + "]" + q;
+    case token_type::equal:         return q + "==" + q;
+    case token_type::not_equal:     return q + "!=" + q;
+    case token_type::less:          return q + "<" + q;
+    case token_type::greater:       return q + ">" + q;
+    case token_type::less_equal:    return q + "<=" + q;
+    case token_type::greater_equal: return q + ">=" + q;
+    case token_type::tilde:         return q + "~" + q;
+    case token_type::caret:         return q + "^" + q;
+    case token_type::bit_or:        return q + "|" + q;
+    }
+
+    assert (false); // Can't be here.
+    return "";
+  }
+
+  class dependency_alternatives_parser
+  {
+  public:
+
+    // If the requirements flavor is specified, then only enable and reflect
+    // clauses are allowed in the multi-line representation.
+    //
+    explicit
+    dependency_alternatives_parser (bool requirements = false)
+        : requirements_ (requirements) {}
+
+    // Throw manifest_parsing if representation is invalid.
+    //
+    void
+    parse (const package_name& dependent,
+           istream&,
+           const string& name,
+           uint64_t line,
+           uint64_t column,
+           dependency_alternatives&);
+
+  private:
+    using lexer      = dependency_alternatives_lexer;
+    using token      = lexer::token;
+    using token_type = lexer::token_type;
+
+    token_type
+    next (token&, token_type&);
+
+    token_type
+    next_eval (token&, token_type&);
+
+    token_type
+    next_line (token&, token_type&);
+
+    token_type
+    next_block (token&, token_type&);
+
+    // Receive the token/type from which it should start consuming and in
+    // return the token/type contains the first token that has not been
+    // consumed (normally eos, newline, or '|').
+    //
+    dependency_alternative
+    parse_alternative (token&, token_type&, bool first);
+
+    // Helpers.
+    //
+    // Throw manifest_parsing with the `<what> expected instead of <token>`
+    // description.
+    //
+    [[noreturn]] void
+    unexpected_token (const token&, string&& what);
+
+    bool requirements_;
+
+    const package_name* dependent_;
+    const string* name_;
+    lexer* lexer_;
+    dependency_alternatives* result_;
+  };
+
+  [[noreturn]] void dependency_alternatives_parser::
+  unexpected_token (const token& t, string&& w)
+  {
+    w += " expected";
+
+    // Don't add the `instead of...` part, if the unexpected token is eos or
+    // an empty word/buildfile.
+    //
+    if (t.type != token_type::eos &&
+        ((t.type != token_type::word && t.type != token_type::buildfile) ||
+         !t.value.empty ()))
+    {
+      w += " instead of ";
+      w += t.string ();
+    }
+
+    throw parsing (*name_, t.line, t.column, w);
+  }
+
+  void dependency_alternatives_parser::
+  parse (const package_name& dependent,
+         istream& is,
+         const string& name,
+         uint64_t line,
+         uint64_t column,
+         dependency_alternatives& result)
+  {
+    lexer lexer (is, name, line, column);
+
+    dependent_ = &dependent;
+    name_ = &name;
+    lexer_ = &lexer;
+    result_ = &result;
+
+    string what (requirements_ ? "requirement" : "dependency");
+
+    token t;
+    token_type tt;
+    next (t, tt);
+
+    // Make sure the representation is not empty, unless we are in the
+    // requirements mode. In the latter case fallback to creating a simple
+    // unconditional requirement. Note that it's the caller's responsibility
+    // to verify that a non-empty comment is specified in this case.
+    //
+    if (tt == token_type::eos)
+    {
+      if (!requirements_)
+        unexpected_token (t, what + " alternatives");
+
+      dependency_alternative da;
+      da.push_back (dependency ());
+
+      result_->push_back (move (da));
+      return;
+    }
+
+    for (bool first (true); tt != token_type::eos; )
+    {
+      dependency_alternative da (parse_alternative (t, tt, first));
+
+      // Skip newline after the dependency alternative, if present.
+      //
+      if (tt == token_type::newline)
+        next (t, tt);
+
+      // Make sure that the simple requirement has the only alternative in the
+      // representation.
+      //
+      if (requirements_   &&
+          da.size () == 1 &&
+          (da[0].name.empty () || (da.enable && da.enable->empty ())))
+      {
+        assert (first);
+
+        if (tt != token_type::eos)
+          throw parsing (*name_,
+                         t.line,
+                         t.column,
+                         "end of simple requirement expected");
+      }
+      else
+      {
+        if (tt != token_type::eos && tt != token_type::bit_or)
+          unexpected_token (t, "end of " + what + " alternatives or '|'");
+      }
+
+      if (tt == token_type::bit_or)
+      {
+        next (t, tt);
+
+        // Skip newline after '|', if present.
+        //
+        if (tt == token_type::newline)
+          next (t, tt);
+
+        // Make sure '|' is not followed by eos.
+        //
+        if (tt == token_type::eos)
+          unexpected_token (t, move (what));
+      }
+
+      result_->push_back (move (da));
+
+      first = false;
+    }
+  }
+
+  dependency_alternative dependency_alternatives_parser::
+  parse_alternative (token& t, token_type& tt, bool first)
+  {
+    using type  = token_type;
+
+    dependency_alternative r;
+
+    string what (requirements_ ? "requirement" : "dependency");
+    string config ("config." + dependent_->variable () + ".");
+
+    auto bad_token = [&t, this] (string&& what)
+    {
+      unexpected_token (t, move (what));
+    };
+
+    // Check that the current token type matches the expected one. Throw
+    // manifest_parsing if that's not the case. Use the expected token type
+    // name for the error description or the custom name, if specified. For
+    // the word and buildfile token types the custom name must be specified.
+    //
+    // Only move from the custom name argument if throwing exception.
+    //
+    auto expect_token = [&tt, &bad_token] (type et,
+                                           string&& what = string ())
+    {
+      assert ((et != type::word && et != type::buildfile) || !what.empty ());
+
+      if (tt != et)
+      {
+        if (what.empty ())
+        {
+          token e {et, "", 0, 0};
+          bad_token (e.string ());
+        }
+        else
+          bad_token (move (what));
+      }
+    };
+
+    // Parse dependencies.
+    //
+    // If the current token starts the version constraint, then read its
+    // tokens, rejoin them, and return the constraint string representation.
+    // Otherwise return nullopt.
+    //
+    // Note that normally the caller reads the dependency package name, reads
+    // the version constraint and, if present, appends it to the dependency,
+    // and then creates the dependency object with a single constructor call.
+    //
+    // Note: doesn't read token that follows the constraint.
+    //
+    auto try_scan_version_constraint =
+      [&t, &tt, &bad_token, &expect_token, this] () -> optional<string>
+    {
+      switch (t.type)
+      {
+      case type::lparen:
+      case type::lsbrace:
+        {
+          string r (t.string (false /* diag */));
+
+          next (t, tt);
+
+          expect_token (type::word, "version");
+
+          r += t.string (false /* diag */);
+          r += ' ';
+
+          next (t, tt);
+
+          expect_token (type::word, "version");
+
+          r += t.string (false /* diag */);
+
+          next (t, tt);
+
+          if (tt != type::rparen && tt != type::rsbrace)
+            bad_token ("')' or ']'");
+
+          r += t.string (false /* diag */);
+
+          return optional<string> (move (r));
+        }
+
+      case type::equal:
+      case type::not_equal:
+      case type::less:
+      case type::greater:
+      case type::less_equal:
+      case type::greater_equal:
+      case type::tilde:
+      case type::caret:
+        {
+          string r (t.string (false /* diag */));
+
+          next (t, tt);
+
+          expect_token (type::word, "version");
+
+          r += t.string (false /* diag */);
+
+          return optional<string> (move (r));
+        }
+
+      default: return nullopt;
+      }
+    };
+
+    // Parse the evaluation context including the left and right parenthesis
+    // and return the enclosed buildfile fragment.
+    //
+    // Note: no token is read after terminating ')'.
+    //
+    auto parse_eval = [&t, &tt, &expect_token, &bad_token, this] ()
+    {
+      next (t, tt);
+      expect_token (type::lparen);
+
+      next_eval (t, tt);
+
+      if (t.value.empty ())
+        bad_token ("condition");
+
+      string r (move (t.value));
+
+      next (t, tt);
+      expect_token (type::rparen);
+
+      return r;
+    };
+
+    const char* vccs ("([<>=!~^");
+
+    bool group (tt == type::lcbrace); // Dependency group.
+
+    if (group)
+    {
+      next (t, tt);
+
+      if (tt == type::rcbrace)
+        bad_token (move (what));
+
+      while (tt != type::rcbrace)
+      {
+        expect_token (type::word, what + " or '}'");
+
+        string   d  (move (t.value));
+        uint64_t dl (t.line);
+        uint64_t dc (t.column);
+
+        next (t, tt);
+
+        optional<string> vc (try_scan_version_constraint ());
+
+        if (vc)
+        {
+          d += *vc;
+
+          next (t, tt);
+        }
+
+        try
+        {
+          r.emplace_back (d);
+        }
+        catch (const invalid_argument& e)
+        {
+          throw parsing (*name_, dl, dc, e.what ());
+        }
+      }
+
+      // See if a common version constraint follows the dependency group and
+      // parse it if that's the case.
+      //
+      // Note that we need to be accurate not to consume what may end up to be
+      // a part of the reflect config.
+      //
+      lexer::xchar c (lexer_->peek_char ());
+
+      if (!lexer::eos (c) && strchr (vccs, c) != nullptr)
+      {
+        next (t, tt);
+
+        uint64_t vcl (t.line);
+        uint64_t vcc (t.column);
+
+        optional<string> vc (try_scan_version_constraint ());
+
+        if (!vc)
+          bad_token ("version constraint");
+
+        try
+        {
+          version_constraint c (*vc);
+
+          for (dependency& d: r)
+          {
+            if (!d.constraint)
+              d.constraint = c;
+          }
+        }
+        catch (const invalid_argument& e)
+        {
+          throw parsing (*name_,
+                         vcl,
+                         vcc,
+                         string ("invalid version constraint: ") + e.what ());
+        }
+      }
+    }
+    else        // Single dependency.
+    {
+      // If we see the question mark instead of a word in the requirements
+      // mode, then this is a simple requirement. In this case parse the
+      // evaluation context, if present, and bail out.
+      //
+      if (requirements_ && first && tt == type::question)
+      {
+        r.emplace_back (dependency ());
+        r.enable = lexer_->peek_char () == '(' ? parse_eval () : string ();
+
+        next (t, tt);
+        return r;
+      }
+
+      expect_token (type::word, move (what));
+
+      string   d  (move (t.value));
+      uint64_t dl (t.line);
+      uint64_t dc (t.column);
+
+      // See if a version constraint follows the dependency package name and
+      // parse it if that's the case.
+      //
+      lexer::xchar c (lexer_->peek_char ());
+
+      if (!lexer::eos (c) && strchr (vccs, c) != nullptr)
+      {
+        next (t, tt);
+
+        optional<string> vc (try_scan_version_constraint ());
+
+        if (!vc)
+          bad_token ("version constraint");
+
+        d += *vc;
+      }
+
+      try
+      {
+        r.emplace_back (d);
+      }
+      catch (const invalid_argument& e)
+      {
+        throw parsing (*name_, dl, dc, e.what ());
+      }
+    }
+
+    // See if there is an enable condition and parse it if that's the case.
+    //
+    {
+      lexer::xchar c (lexer_->peek_char ());
+
+      if (c == '?')
+      {
+        next (t, tt);
+        expect_token (type::question);
+
+        // If we don't see the opening parenthesis in the requirements mode,
+        // then this is a simple requirement. In this case set the enable
+        // condition to an empty string and bail out.
+        //
+        c = lexer_->peek_char ();
+
+        if (requirements_ && first && !group && c != '(')
+        {
+          r.enable = "";
+
+          next (t, tt);
+          return r;
+        }
+
+        r.enable = parse_eval ();
+      }
+    }
+
+    // See if there is a reflect config and parse it if that's the case.
+    //
+    {
+      lexer::xchar c (lexer_->peek_char ());
+
+      if (!lexer::eos (c) && strchr ("|\n", c) == nullptr)
+      {
+        next_line (t, tt);
+
+        string& l (t.value);
+        if (l.compare (0, config.size (), config) != 0)
+          bad_token (config + "* variable assignment");
+
+        r.reflect = move (l);
+      }
+    }
+
+    // If the dependencies are terminated with the newline, then check if the
+    // next token is '{'. If that's the case, then this is a multi-line
+    // representation.
+    //
+    next (t, tt);
+
+    if (tt == type::newline)
+    {
+      next (t, tt);
+
+      if (tt == type::lcbrace)
+      {
+        if (r.enable)
+          throw parsing (
+            *name_,
+            t.line,
+            t.column,
+            "multi-line " + what + " form with inline enable clause");
+
+        if (r.reflect)
+          throw parsing (
+            *name_,
+            t.line,
+            t.column,
+            "multi-line " + what + " form with inline reflect clause");
+
+        next (t, tt);
+        expect_token (type::newline);
+
+        // Parse the clauses.
+        //
+        for (next (t, tt); tt == type::word; next (t, tt))
+        {
+          auto fail_dup = [&t, this] ()
+          {
+            throw parsing (*name_, t.line, t.column, "duplicate clause");
+          };
+
+          auto fail_precede = [&t, this] (const char* what)
+          {
+            throw parsing (
+              *name_,
+              t.line,
+              t.column,
+              t.value + " clause should precede " + what + " clause");
+          };
+
+          auto fail_conflict = [&t, this] (const char* what)
+          {
+            throw parsing (
+              *name_,
+              t.line,
+              t.column,
+              t.value + " and " + what + " clauses are mutually exclusive");
+          };
+
+          auto fail_requirements = [&t, this] ()
+          {
+            throw parsing (
+              *name_,
+              t.line,
+              t.column,
+              t.value + " clause is not permitted for requirements");
+          };
+
+          // Parse the buildfile fragment block including the left and right
+          // curly braces (expected to be on the separate lines) and return
+          // the enclosed fragment.
+          //
+          auto parse_block = [&t, &tt, &expect_token, &bad_token, this] ()
+          {
+            next (t, tt);
+            expect_token (type::newline);
+
+            next (t, tt);
+            expect_token (type::lcbrace);
+
+            next (t, tt);
+            expect_token (type::newline);
+
+            next_block (t, tt);
+
+            string r (move (t.value));
+
+            // Fail if the buildfile fragment is empty.
+            //
+            if (r.find_first_not_of (" \t\n") == string::npos)
+              bad_token ("buildfile fragment");
+
+            return r;
+          };
+
+          const string& v (t.value);
+
+          if (v == "enable")
+          {
+            if (r.enable)
+              fail_dup ();
+
+            if (r.prefer)
+              fail_precede ("prefer");
+
+            if (r.require)
+              fail_precede ("require");
+
+            if (r.reflect)
+              fail_precede ("reflect");
+
+            r.enable = parse_eval ();
+
+            next (t, tt);
+            expect_token (type::newline);
+          }
+          else if (v == "prefer")
+          {
+            if (requirements_)
+              fail_requirements ();
+
+            if (r.prefer)
+              fail_dup ();
+
+            if (r.require)
+              fail_conflict ("require");
+
+            if (r.reflect)
+              fail_precede ("reflect");
+
+            r.prefer = parse_block ();
+
+            // The accept clause must follow, so parse it.
+            //
+            next (t, tt);
+
+            if (tt != type::word || t.value != "accept")
+              bad_token ("accept clause");
+
+            r.accept = parse_eval ();
+
+            next (t, tt);
+            expect_token (type::newline);
+          }
+          else if (v == "require")
+          {
+            if (requirements_)
+              fail_requirements ();
+
+            if (r.require)
+              fail_dup ();
+
+            if (r.prefer)
+              fail_conflict ("prefer");
+
+            if (r.reflect)
+              fail_precede ("reflect");
+
+            r.require = parse_block ();
+          }
+          else if (v == "reflect")
+          {
+            if (r.reflect)
+              fail_dup ();
+
+            r.reflect = parse_block ();
+          }
+          else if (v == "accept")
+          {
+            if (requirements_)
+              fail_requirements ();
+
+            throw parsing (*name_,
+                           t.line,
+                           t.column,
+                           "accept clause should follow prefer clause");
+          }
+          else
+            bad_token (what + " alternative clause");
+        }
+
+        expect_token (type::rcbrace);
+        next (t, tt);
+      }
+    }
+
+    return r;
+  }
+
+  dependency_alternatives_parser::token_type dependency_alternatives_parser::
+  next (token& t, token_type& tt)
+  {
+    t = lexer_->next ();
+    tt = t.type;
+    return tt;
+  }
+
+  dependency_alternatives_parser::token_type dependency_alternatives_parser::
+  next_eval (token& t, token_type& tt)
+  {
+    t = lexer_->next_eval ();
+    tt = t.type;
+    return tt;
+  }
+
+  dependency_alternatives_parser::token_type dependency_alternatives_parser::
+  next_line (token& t, token_type& tt)
+  {
+    t = lexer_->next_line ('|');
+    tt = t.type;
+    return tt;
+  }
+
+  dependency_alternatives_parser::token_type dependency_alternatives_parser::
+  next_block (token& t, token_type& tt)
+  {
+    t = lexer_->next_block ();
+    tt = t.type;
+    return tt;
+  }
+
+  dependency_alternatives::
+  dependency_alternatives (const std::string& s,
+                           const package_name& dependent,
+                           const std::string& name,
+                           uint64_t line,
+                           uint64_t column)
+  {
+    using std::string;
 
     auto vc (parser::split_comment (s));
 
-    conditional = (cond != 0);
-    buildtime   = (btim != 0);
-    comment     = move (vc.second);
+    comment = move (vc.second);
 
-    const string& vl (vc.first);
+    const string& v (vc.first);
+    buildtime = (v[0] == '*');
 
-    string::const_iterator b (vl.begin ());
-    string::const_iterator e (vl.end ());
+    string::const_iterator b (v.begin ());
+    string::const_iterator e (v.end ());
 
-    if (conditional || buildtime)
+    if (buildtime)
     {
-      string::size_type p (vl.find_first_not_of (spaces, cond + btim));
+      string::size_type p (v.find_first_not_of (spaces, 1));
       b = p == string::npos ? e : b + p;
     }
 
-    list_parser lp (b, e, '|');
-    for (string lv (lp.next ()); !lv.empty (); lv = lp.next ())
-      push_back (dependency_alternative (lv));
+    dependency_alternatives_parser p;
+    istringstream is (b == v.begin () ? v : string (b, e));
+    p.parse (dependent, is, name, line, column, *this);
   }
 
   string dependency_alternatives::
   string () const
   {
-    std::string r (conditional
-                   ? (buildtime ? "?* " : "? ")
-                   : (buildtime ? "* " : ""));
+    std::string r (buildtime ? "* " : "");
 
-    bool f (true);
+    const dependency_alternative* prev (nullptr);
     for (const dependency_alternative& da: *this)
     {
-      r += (f ? (f = false, "") : " | ");
+      if (prev != nullptr)
+      {
+        r += prev->single_line () ? " |" : "\n|";
+        r += !da.single_line () || !prev->single_line () ? '\n' : ' ';
+      }
+
       r += da.string ();
+      prev = &da;
     }
 
     return serializer::merge_comment (r, comment);
   }
 
-  // requirement_alternative
-  //
-  requirement_alternative::
-  requirement_alternative (const std::string& s)
+  bool dependency_alternatives::
+  conditional () const
   {
-    push_back (s); // @@ DEP
+    for (const dependency_alternative& da: *this)
+    {
+      if (da.enable)
+        return true;
+    }
+
+    return false;
   }
 
+  // requirement_alternative
+  //
   string requirement_alternative::
   string () const
   {
-    assert (size () == 1); // @@ DEP
-    return front ();
+    using std::string;
+
+    string r (size () > 1 ? "{" : "");
+
+    bool first (true);
+    for (const string& rq: *this)
+    {
+      if (!first)
+        r += ' ';
+      else
+        first = false;
+
+      r += rq;
+    }
+
+    if (size () > 1)
+      r += '}';
+
+    if (single_line ())
+    {
+      if (enable)
+      {
+        if (!simple ())
+        {
+          r += " ? (";
+          r += *enable;
+          r += ')';
+        }
+        else
+        {
+          // Note that the (single) requirement id may or may not be empty.
+          //
+          if (!r.empty ())
+            r += ' ';
+
+          r += '?';
+
+          if (!enable->empty ())
+          {
+            r += " (";
+            r += *enable;
+            r += ')';
+          }
+        }
+      }
+
+      if (reflect)
+      {
+        r += ' ';
+        r += *reflect;
+      }
+    }
+    else
+    {
+      r += "\n{";
+
+      if (enable)
+      {
+        r += "\n  enable (";
+        r += *enable;
+        r += ')';
+      }
+
+      if (reflect)
+      {
+        if (enable)
+          r += '\n';
+
+        r += "\n  reflect\n  {\n";
+        r += *reflect;
+        r += "  }";
+      }
+
+      r += "\n}";
+    }
+
+    return r;
+  }
+
+  bool requirement_alternative::
+  single_line () const
+  {
+    return !reflect || reflect->find ('\n') == string::npos;
   }
 
   // requirement_alternatives
   //
   requirement_alternatives::
-  requirement_alternatives (const std::string& v)
+  requirement_alternatives (const std::string& s,
+                            const package_name& dependent,
+                            const std::string& name,
+                            uint64_t line,
+                            uint64_t column)
   {
     using std::string;
 
-    // Allow specifying ?* in any order.
-    //
-    size_t n (v.size ());
-    size_t cond ((n > 0 && v[0] == '?') || (n > 1 && v[1] == '?') ? 1 : 0);
-    size_t btim ((n > 0 && v[0] == '*') || (n > 1 && v[1] == '*') ? 1 : 0);
-
-    auto vc (parser::split_comment (v));
+    auto vc (parser::split_comment (s));
 
-    conditional = (cond != 0);
-    buildtime   = (btim != 0);
-    comment     = move (vc.second);
+    comment = move (vc.second);
 
-    const string& vl (vc.first);
+    const string& v (vc.first);
+    buildtime = (v[0] == '*');
 
-    string::const_iterator b (vl.begin ());
-    string::const_iterator e (vl.end ());
+    string::const_iterator b (v.begin ());
+    string::const_iterator e (v.end ());
 
-    if (conditional || buildtime)
+    if (buildtime)
     {
-      string::size_type p (vl.find_first_not_of (spaces, cond + btim));
+      string::size_type p (v.find_first_not_of (spaces, 1));
       b = p == string::npos ? e : b + p;
     }
 
-    list_parser lp (b, e, '|');
-    for (string lv (lp.next ()); !lv.empty (); lv = lp.next ())
-      push_back (requirement_alternative (lv));
+    // We will use the dependency alternatives parser to parse the
+    // representation into a temporary dependency alternatives in the
+    // requirements mode. Then we will move the dependency alternatives into
+    // the requirement alternatives using the string representation of the
+    // dependencies.
+    //
+    dependency_alternatives_parser p (true /* requirements */);
+    istringstream is (b == v.begin () ? v : string (b, e));
 
-    if (empty () && comment.empty ())
-      throw invalid_argument ("empty package requirement specification");
+    dependency_alternatives das;
+    p.parse (dependent, is, name, line, column, das);
+
+    for (dependency_alternative& da: das)
+    {
+      requirement_alternative ra (move (da.enable), move (da.reflect));
+
+      // Also handle the simple requirement.
+      //
+      for (dependency& d: da)
+        ra.push_back (!d.name.empty () ? d.string () : string ());
+
+      push_back (move (ra));
+    }
+
+    // Make sure that the simple requirement is accompanied with a non-empty
+    // comment.
+    //
+    if (simple () && comment.empty ())
+    {
+      // Let's describe the following error cases differently:
+      //
+      // requires: ?
+      // requires:
+      //
+      throw parsing (name,
+                     line,
+                     column,
+                     (back ().enable
+                      ? "no comment specified for simple requirement"
+                      : "requirement or comment expected"));
+    }
   }
 
   std::string requirement_alternatives::
   string () const
   {
-    std::string r (conditional
-                   ? (buildtime ? "?* " : "? ")
-                   : (buildtime ? "* " : ""));
+    using std::string;
+
+    string r (buildtime ? "* " : "");
 
-    bool f (true);
+    const requirement_alternative* prev (nullptr);
     for (const requirement_alternative& ra: *this)
     {
-      r += (f ? (f = false, "") : " | ");
+      if (prev != nullptr)
+      {
+        r += prev->single_line () ? " |" : "\n|";
+        r += !ra.single_line () || !prev->single_line () ? '\n' : ' ';
+      }
+
       r += ra.string ();
+      prev = &ra;
     }
 
+    // For better readability separate the comment from the question mark for
+    // the simple requirement with an empty condition.
+    //
+    if (simple () && conditional () && back ().enable->empty ())
+      r += ' ';
+
     return serializer::merge_comment (r, comment);
   }
 
+  bool requirement_alternatives::
+  conditional () const
+  {
+    for (const requirement_alternative& ra: *this)
+    {
+      if (ra.enable)
+        return true;
+    }
+
+    return false;
+  }
+
   // build_class_term
   //
   build_class_term::
@@ -1966,9 +3263,11 @@ namespace bpkg
 
     // We will cache the depends and the test dependency manifest values to
     // parse and, if requested, complete the version constraints later, after
-    // the version value is parsed.
+    // the version value is parsed. We will also cache the requires values to
+    // parse them later, after the package name is parsed.
     //
     vector<name_value> dependencies;
+    vector<name_value> requirements;
     small_vector<name_value, 1> tests;
 
     // We will cache the description and its type values to validate them
@@ -2254,16 +3553,13 @@ namespace bpkg
 
         m.license_alternatives.push_back (move (l));
       }
+      else if (n == "depends")
+      {
+        dependencies.push_back (move (nv));
+      }
       else if (n == "requires")
       {
-        try
-        {
-          m.requirements.push_back (requirement_alternatives (v));
-        }
-        catch (const invalid_argument& e)
-        {
-          bad_value (e.what ());
-        }
+        requirements.push_back (move (nv));
       }
       else if (n == "builds")
       {
@@ -2280,10 +3576,6 @@ namespace bpkg
         m.build_constraints.push_back (
           parse_build_constraint (nv, true /* exclusion */, name));
       }
-      else if (n == "depends")
-      {
-        dependencies.push_back (move (nv));
-      }
       // @@ TMP time to drop *-0.14.0?
       //
       else if (n == "tests"      || n == "tests-0.14.0"    ||
@@ -2467,8 +3759,8 @@ namespace bpkg
       }
       catch (const invalid_argument& e)
       {
-        throw invalid_argument (string ("invalid package constraint: ") +
-                                e.what ());
+        throw invalid_argument ("invalid package constraint '" +
+                                dep.constraint->string () + "': " + e.what ());
       }
 
       return move (dep);
@@ -2480,16 +3772,15 @@ namespace bpkg
     {
       nv = move (d); // Restore as bad_value() uses its line/column.
 
-      const string& v (nv.value);
-
       // Parse dependency alternatives.
       //
       try
       {
-        dependency_alternatives das (v);
-
-        if (das.empty ())
-          bad_value ("empty package dependency specification");
+        dependency_alternatives das (nv.value,
+                                     m.name,
+                                     name,
+                                     nv.value_line,
+                                     nv.value_column);
 
         for (dependency_alternative& da: das)
         {
@@ -2505,6 +3796,18 @@ namespace bpkg
       }
     }
 
+    // Parse the requirements.
+    //
+    for (const name_value& r: requirements)
+    {
+      m.requirements.push_back (
+        requirement_alternatives (r.value,
+                                  m.name,
+                                  name,
+                                  r.value_line,
+                                  r.value_column));
+    }
+
     // Parse the test dependencies.
     //
     for (name_value& t: tests)
diff --git a/libbpkg/manifest.hxx b/libbpkg/manifest.hxx
index d72a55e..499d64d 100644
--- a/libbpkg/manifest.hxx
+++ b/libbpkg/manifest.hxx
@@ -8,7 +8,7 @@
 #include <string>
 #include <vector>
 #include <cassert>
-#include <cstdint>    // uint16_t
+#include <cstdint>    // uint*_t
 #include <ostream>
 #include <utility>    // move()
 #include <stdexcept>  // logic_error
@@ -455,45 +455,175 @@ namespace bpkg
 
   // depends
   //
+  // The dependency alternative can be represented in one of the following
+  // forms.
+  //
+  // Single-line form:
+  //
+  //   <dependencies> ['?' <enable-condition>] [<reflect-config>]
+  //
+  //   <dependencies> = <dependency> |
+  //                    ({ <dependency> [ <dependency>]* } [<version-constraint>])
+  //
+  //   <enable-condition> - buildfile evaluation context
+  //   <reflect-config>   - dependent package configuration variable assignment
+  //
+  //   If the version constraint is specified after the dependency group, it
+  //   only applies to dependencies without a version constraint.
+  //
+  // Multi-line forms:
+  //
+  //   <dependencies>
+  //   {
+  //     enable <enable-condition>
+  //
+  //     prefer
+  //     {
+  //       <prefer-config>
+  //     }
+  //
+  //     accept <accept-condition>
+  //
+  //     reflect
+  //     {
+  //       <reflect-config>
+  //     }
+  //   }
+  //   |
+  //   <dependencies>
+  //   {
+  //     enable <enable-condition>
+  //
+  //     require
+  //     {
+  //       <require-config>
+  //     }
+  //
+  //     reflect
+  //     {
+  //       <reflect-config>
+  //     }
+  //   }
+  //
+  //   <prefer-config>    - buildfile fragment containing dependency packages
+  //                        configuration variables assignments
+  //
+  //   <accept-condition> - buildfile evaluation context
+  //
+  //   <require-config>   - buildfile fragment containing dependency packages
+  //                        configuration variables assignments
+  //
+  //   <reflect-config>   - buildfile fragment containing dependent package
+  //                        configuration variables assignments
+  //
+  // The dependency alternative is only considered by bpkg if the enable
+  // condition evaluates to true. If the enable clause is not specified, then
+  // it is always considered.
+  //
+  // The prefer clause specifies the preferred dependency package
+  // configuration that may potentially differ from the resulting
+  // configuration after the preferred/required configurations from all the
+  // selected dependency alternatives of all the dependent packages are
+  // "negotiated" by bpkg. The accept clause is used to verify that the
+  // resulting configuration is still acceptable for the dependent
+  // package. The accept clause must always be specified if the prefer clause
+  // is specified.
+  //
+  // The require clause specifies the only acceptable dependency packages
+  // configuration. It is a shortcut for specifying the prefer/accept clauses,
+  // where the accept condition verifies all the variable values assigned in
+  // the prefer clause. The require clause and the prefer/accept clause pair
+  // are optional and are mutually exclusive.
+  //
+  // The reflect clause specifies the dependent package configuration that
+  // should be used if the alternative is selected.
+  //
+  // All clauses are optional but at least one of them must be specified.
+  //
   class dependency_alternative: public butl::small_vector<dependency, 1>
   {
   public:
     butl::optional<std::string> enable;
+    butl::optional<std::string> reflect;
+    butl::optional<std::string> prefer;
+    butl::optional<std::string> accept;
+    butl::optional<std::string> require;
 
     dependency_alternative () = default;
-    dependency_alternative (butl::optional<std::string> e)
-        : enable (std::move (e)) {}
-
-    // Parse the dependency alternative string representation.
+    dependency_alternative (butl::optional<std::string> e,
+                            butl::optional<std::string> r,
+                            butl::optional<std::string> p,
+                            butl::optional<std::string> a,
+                            butl::optional<std::string> q)
+        : enable (std::move (e)),
+          reflect (std::move (r)),
+          prefer (std::move (p)),
+          accept (std::move (a)),
+          require (std::move (q)) {}
+
+    // Return the single-line representation if possible (the prefer and
+    // require clauses are absent and the reflect clause either absent or
+    // contains no newlines).
     //
-    explicit LIBBPKG_EXPORT
-    dependency_alternative (const std::string&);
-
     LIBBPKG_EXPORT std::string
     string () const;
+
+    // Return true if the string() function would return the single-line
+    // representation.
+    //
+    LIBBPKG_EXPORT bool
+    single_line () const;
   };
 
   class dependency_alternatives:
     public butl::small_vector<dependency_alternative, 1>
   {
   public:
-    bool conditional;
     bool buildtime;
     std::string comment;
 
     dependency_alternatives () = default;
-    dependency_alternatives (bool d, bool b, std::string c)
-        : conditional (d), buildtime (b), comment (std::move (c)) {}
+    dependency_alternatives (bool b, std::string c)
+        : buildtime (b), comment (std::move (c)) {}
 
-    // Parse the dependency alternatives string representation in the
-    // `[?][*] <dependency> [ '|' <dependency>]* [; <comment>]` form. Throw
-    // std::invalid_argument if the value is invalid. @@ DEP @@ TMP update.
+    // Parse the dependency alternatives string representation in the form:
+    //
+    // [*] <alternative> [ '|' <alternative>]* [; <comment>]
+    //
+    // Where <alternative> can be single or multi-line (see above). Note also
+    // that leading `*` and trailing comment can be on separate lines. Throw
+    // manifest_parsing if the value is invalid.
+    //
+    // Use the dependent package name to verify that the reflect clauses in
+    // the dependency alternative representations refer to the dependent
+    // package configuration variable.
+    //
+    // Optionally, specify the stream name to use when creating the
+    // manifest_parsing exception. The start line and column arguments can be
+    // used to align the exception information with a containing stream. This
+    // is useful when the alternatives representation is a part of some larger
+    // text (manifest, etc).
+    //
+    // Note that semicolons inside alternatives must be escaped with the
+    // backslash (not to be treated as the start of a comment). Backslashes at
+    // the end of buildfile fragment lines need to also be escaped, if
+    // dependency alternatives representation comes from the manifest file
+    // (since trailing backslashes in manifest lines has special semantics).
     //
     explicit LIBBPKG_EXPORT
-    dependency_alternatives (const std::string&);
+    dependency_alternatives (const std::string&,
+                             const package_name& dependent,
+                             const std::string& name = std::string (),
+                             std::uint64_t line = 1,
+                             std::uint64_t column = 1);
 
     LIBBPKG_EXPORT std::string
     string () const;
+
+    // Return true if there is a conditional alternative in the list.
+    //
+    LIBBPKG_EXPORT bool
+    conditional () const;
   };
 
   inline std::ostream&
@@ -504,45 +634,109 @@ namespace bpkg
 
   // requires
   //
+  // The requirement alternative string representation is similar to that of
+  // the dependency alternative with the following differences:
+  //
+  // - The requirement id (with or without version) can mean anything (but
+  //   must still be a valid package name).
+  //
+  // - Only the enable and reflect clauses are permitted (reflect is allowed
+  //   for potential future support of recognized requirement alternatives,
+  //   for example, C++ standard).
+  //
+  // - The simplified representation syntax, where the comment carries the
+  //   main information and thus is mandatory, is also supported (see
+  //   requirement_alternatives for details). For example:
+  //
+  //   requires: ; X11 libs.
+  //   requires: ? ($windows) ; Only 64-bit.
+  //   requires: ? ; Only 64-bit if on Windows.
+  //   requires: x86_64 ? ; Only if on Windows.
+  //
   class requirement_alternative: public butl::small_vector<std::string, 1>
   {
   public:
     butl::optional<std::string> enable;
+    butl::optional<std::string> reflect;
 
     requirement_alternative () = default;
-    requirement_alternative (butl::optional<std::string> e)
-        : enable (std::move (e)) {}
+    requirement_alternative (butl::optional<std::string> e,
+                             butl::optional<std::string> r)
+        : enable (std::move (e)), reflect (std::move (r)) {}
 
-    // Parse the requirement alternative string representation.
+    // Return the single-line representation if possible (the reflect clause
+    // either absent or contains no newlines).
     //
-    explicit LIBBPKG_EXPORT
-    requirement_alternative (const std::string&);
-
     LIBBPKG_EXPORT std::string
     string () const;
+
+    // Return true if the string() function would return the single-line
+    // representation.
+    //
+    LIBBPKG_EXPORT bool
+    single_line () const;
+
+    // Return true if this is a single requirement with an empty id or an
+    // empty enable condition.
+    //
+    bool
+    simple () const
+    {
+      return size () == 1 && (back ().empty () || (enable && enable->empty ()));
+    }
   };
 
   class requirement_alternatives:
     public butl::small_vector<requirement_alternative, 1>
   {
   public:
-    bool conditional;
     bool buildtime;
     std::string comment;
 
     requirement_alternatives () = default;
-    requirement_alternatives (bool d, bool b, std::string c)
-        : conditional (d), buildtime (b), comment (std::move (c)) {}
+    requirement_alternatives (bool b, std::string c)
+        : buildtime (b), comment (std::move (c)) {}
 
     // Parse the requirement alternatives string representation in the
-    // `[?] [<requirement> [ '|' <requirement>]*] [; <comment>]` form. Throw
-    // std::invalid_argument if the value is invalid. @@ DEP @@ TMP update.
+    // following forms:
+    //
+    // [*] <alternative> [ '|' <alternative>]* [; <comment>]
+    // [*] [<requirement-id>] [? [<enable-condition>]] ; <comment>
+    //
+    // Parsing the second form ends up with a single alternative with a single
+    // potentially empty requirement id, potentially with an enable condition
+    // with potentially empty value (see examples above).
+    //
+    // Throw manifest_parsing if the value is invalid.
+    //
+    // Optionally, specify the stream name to use when creating the
+    // manifest_parsing exception. The start line and column arguments can be
+    // used to align the exception information with a containing stream. This
+    // is useful when the alternatives representation is a part of some larger
+    // text (manifest, etc).
     //
     explicit LIBBPKG_EXPORT
-    requirement_alternatives (const std::string&);
+    requirement_alternatives (const std::string&,
+                              const package_name& dependent,
+                              const std::string& name = std::string (),
+                              std::uint64_t line = 1,
+                              std::uint64_t column = 1);
 
     LIBBPKG_EXPORT std::string
     string () const;
+
+    // Return true if there is a conditional alternative in the list.
+    //
+    LIBBPKG_EXPORT bool
+    conditional () const;
+
+    // Return true if this is a single simple requirement alternative.
+    //
+    bool
+    simple () const
+    {
+      return size () == 1 && back ().simple ();
+    }
   };
 
   inline std::ostream&