Implement low-level manifest parser

1 files changed, 453 insertions, 0 deletions

diff --git a/bpkg/manifest-parser.cxx b/bpkg/manifest-parser.cxx
new file mode 100644
index 0000000..374de3e
--- /dev/null
+++ b/bpkg/manifest-parser.cxx
@@ -0,0 +1,453 @@
+// file      : bpkg/manifest-parser.cxx -*- C++ -*-
+// copyright : Copyright (c) 2014-2015 Code Synthesis Ltd
+// license   : MIT; see accompanying LICENSE file
+
+#include <bpkg/manifest-parser>
+
+#include <cassert>
+#include <sstream>
+
+using namespace std;
+
+namespace bpkg
+{
+  using name_value = manifest_parser::name_value_type;
+
+  name_value manifest_parser::
+  next ()
+  {
+    if (s_ == eos)
+      return name_value {"", "", l_, c_, l_, c_};
+
+    xchar c (skip_spaces ());
+
+    // Here is the problem: if we are in the 'body' state (that is,
+    // we are parsing inside the manifest) and we see the special
+    // empty name, then before returning the "start" pair for the
+    // next manifest, we have to return the "end" pair. One way
+    // would be to cache the "start" pair and return it on the
+    // next call of next(). But that would require quite a bit
+    // of extra logic. The alternative is to detect the beginning
+    // of the empty name before parsing too far. This way, the
+    // next call to next() will start parsing where we left of
+    // and return the "start" pair naturally.
+    //
+    if (s_ == body && c == ':')
+    {
+      s_ = start;
+      uint64_t ln (c.line ()), cn (c.column ());
+      return name_value {"", "", ln, cn, ln, cn};
+    }
+
+    // Regardless of the state, what should come next is a name,
+    // potentially the special empty one.
+    //
+    name_value r;
+    parse_name (r);
+
+    skip_spaces ();
+    c = get ();
+
+    if (is_eos (c))
+    {
+      // This is ok as long as the name is empty.
+      //
+      if (!r.name.empty ())
+        throw manifest_parsing (name_, c.line (), c.column (),
+                                "':' expected after name");
+
+      s_ = eos;
+
+      // The "end" pair.
+      //
+      r.value_line = r.name_line;
+      r.value_column = r.name_column;
+      return r;
+    }
+
+    if (c != ':')
+      throw manifest_parsing (name_, c.line (), c.column (),
+                              "':' expected after name");
+
+    skip_spaces ();
+    parse_value (r);
+
+    c = peek ();
+
+    // The character after the value should be either a newline or eos.
+    //
+    assert (c == '\n' || is_eos (c));
+
+    if (c == '\n')
+      get ();
+
+    // Now figure out whether what we've got makes sense, depending
+    // on the state we are in.
+    //
+    if (s_ == start)
+    {
+      // Start of the (next) manifest. The first pair should be the
+      // special empty name/format version.
+      //
+      if (!r.name.empty ())
+        throw manifest_parsing (name_, r.name_line, r.name_column,
+                                "format version pair expected");
+
+      // The version value is only mandatory for the first manifest in
+      // a sequence.
+      //
+      if (r.value.empty ())
+      {
+        if (version_.empty ())
+          throw manifest_parsing (name_, r.value_line, r.value_column,
+                                  "format version value expected");
+        r.value = version_;
+      }
+      else
+      {
+        version_ = r.value; // Update with the latest.
+
+        if (version_ != "1")
+          throw manifest_parsing (name_, r.value_line, r.value_column,
+                                  "unsupported format version " + version_);
+      }
+
+      s_ = body;
+    }
+    else
+    {
+      // Parsing the body of the manifest.
+      //
+
+      // Should have been handled by the special case above.
+      //
+      assert (!r.name.empty ());
+    }
+
+    return r;
+  }
+
+  void manifest_parser::
+  parse_name (name_value& r)
+  {
+    xchar c (peek ());
+
+    r.name_line = c.line ();
+    r.name_column = c.column ();
+
+    for (; !is_eos (c); c = peek ())
+    {
+      if (c == ':' || c == ' ' || c == '\t' || c == '\n')
+        break;
+
+      r.name += c;
+      get ();
+    }
+  }
+
+  void manifest_parser::
+  parse_value (name_value& r)
+  {
+    xchar c (peek ());
+
+    r.value_line = c.line ();
+    r.value_column = c.column ();
+
+    string& v (r.value);
+    string::size_type n (0); // Size of last non-space character (simpel mode).
+
+    // Detect the multi-line mode introductor.
+    //
+    bool ml (false);
+    if (c == '\\')
+    {
+      get ();
+      xchar p (peek ());
+
+      if (p == '\n')
+      {
+        get (); // Newline is not part of the value so skip it.
+        c = peek ();
+        ml = true;
+      }
+      else if (is_eos (p))
+        ml = true;
+      else
+        unget (c);
+    }
+
+    // The nl flag signals that the preceding character was a "special
+    // newline", that is, a newline that was part of the milti-line mode
+    // introductor or an escape sequence.
+    //
+    for (bool nl (ml); !is_eos (c); c = peek ())
+    {
+      // Detect the special "\n\\\n" sequence. In the multi-line mode,
+      // this is a "terminator". In the simple mode, this is a way to
+      // specify a newline.
+      //
+      // The key idea here is this: if we "swallowed" any characters
+      // (i.e., called get() without a matching unget()), then we
+      // have to restart the loop in order to do all the tests for
+      // the next character. Also, for this to work, we can only
+      // add one character to v, which limits us to maximum three
+      // characters look-ahead: one in v, one "ungot", and one
+      // peeked.
+      //
+      // The first block handles the special sequence that starts with
+      // a special newline. In multi-line mode, this is an "immediate
+      // termination" where we "use" the newline from the introductor.
+      // Note also that in the simple mode the special sequence can
+      // only start with a special (i.e., escaped) newline.
+      //
+      if (nl)
+      {
+        nl = false;
+
+        if (c == '\\')
+        {
+          get ();
+          xchar c1 (peek ());
+
+          if (c1 == '\n' || is_eos (c1))
+          {
+            if (ml)
+              break;
+            else
+            {
+              if (c1 == '\n')
+                get ();
+
+              v += '\n'; // Literal newline.
+              n++;
+              continue; // Restart from the next character.
+            }
+          }
+          else
+            unget (c); // Fall through.
+        }
+      }
+
+      if (c == '\n')
+      {
+        if (ml)
+        {
+          get ();
+          xchar c1 (peek ());
+
+          if (c1 == '\\')
+          {
+            get ();
+            xchar c2 (peek ());
+
+            if (c2 == '\n' || is_eos (c2))
+              break;
+            else
+            {
+              v += '\n';
+              unget (c1);
+              continue; // Restart from c1 (slash).
+            }
+          }
+          else
+            unget (c); // Fall through.
+        }
+        else
+          break; // Simple value terminator.
+      }
+
+      // Detect the newline escape sequence. The same look-ahead
+      // approach as above.
+      //
+      if (c == '\\')
+      {
+        get ();
+        xchar c1 (peek ());
+
+        if (c1 == '\n' || is_eos (c1))
+        {
+          if (c1 == '\n')
+          {
+            get ();
+            nl = true; // This is a special newline.
+          }
+          continue; // Restart from the next character.
+        }
+        else if (c1 == '\\')
+        {
+          get ();
+          xchar c2 (peek ());
+
+          if (c2 == '\n' || is_eos (c1))
+          {
+            v += '\\';
+            n++;
+            // Restart from c2 (newline/eos).
+          }
+          else
+          {
+            v += '\\';
+            n++;
+            unget (c1); // Restart from c1 (second slash).
+          }
+
+          continue;
+        }
+        else
+          unget (c); // Fall through.
+      }
+
+      get ();
+      v += c;
+
+      if (!ml)
+      {
+        if (c != ' ' && c != '\t')
+          n++;
+      }
+    }
+
+    // Cut off trailing whitespaces.
+    //
+    if (!ml)
+      v.resize (n);
+  }
+
+  manifest_parser::xchar manifest_parser::
+  skip_spaces ()
+  {
+    xchar c (peek ());
+    bool start (c.column () == 1);
+
+    for (; !is_eos (c); c = peek ())
+    {
+      switch (c)
+      {
+      case ' ':
+      case '\t':
+        break;
+      case '\n':
+        {
+          // Skip empty lines.
+          //
+          if (!start)
+            return c;
+
+          break;
+        }
+      case '#':
+        {
+          // We only recognize '#' as a start of a comment at the beginning
+          // of the line (sans leading spaces).
+          //
+          if (!start)
+            return c;
+
+          get ();
+
+          // Read until newline or eos.
+          //
+          for (c = peek (); !is_eos (c) && c != '\n'; c = peek ())
+            get ();
+
+          continue;
+        }
+      default:
+        return c; // Not a space.
+      }
+
+      get ();
+    }
+
+    return c;
+  }
+
+  // Character interface.
+  //
+
+  manifest_parser::xchar manifest_parser::
+  peek ()
+  {
+    if (unget_)
+      return buf_;
+    else
+    {
+      if (eos_)
+        return xchar (xchar::traits_type::eof (), l_, c_);
+      else
+      {
+        xchar::int_type v (is_.peek ());
+
+        if (v == xchar::traits_type::eof ())
+          eos_ = true;
+
+        return xchar (v, l_, c_);
+      }
+    }
+  }
+
+  manifest_parser::xchar manifest_parser::
+  get ()
+  {
+    if (unget_)
+    {
+      unget_ = false;
+      return buf_;
+    }
+    else
+    {
+      // When is_.get () returns eof, the failbit is also set (stupid,
+      // isn't?) which may trigger an exception. To work around this
+      // we will call peek() first and only call get() if it is not
+      // eof. But we can only call peek() on eof once; any subsequent
+      // calls will spoil the failbit (even more stupid).
+      //
+      xchar c (peek ());
+
+      if (!is_eos (c))
+      {
+        is_.get ();
+
+        if (c == '\n')
+        {
+          l_++;
+          c_ = 1;
+        }
+        else
+          c_++;
+      }
+
+      return c;
+    }
+  }
+
+  void manifest_parser::
+  unget (const xchar& c)
+  {
+    // Because iostream::unget cannot work once eos is reached,
+    // we have to provide our own implementation.
+    //
+    buf_ = c;
+    unget_ = true;
+  }
+
+  // manifest_parsing
+  //
+
+  static string
+  format (const string& n, uint64_t l, uint64_t c, const string& d)
+  {
+    ostringstream os;
+    if (!n.empty ())
+      os << n << ':';
+    os << l << ':' << c << ": error: " << d;
+    return os.str ();
+  }
+
+  manifest_parsing::
+  manifest_parsing (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)
+  {
+  }
+}