mirror of
https://git.do.srb2.org/STJr/UltimateZoneBuilder.git
synced 2024-11-23 20:32:34 +00:00
497 lines
13 KiB
C#
497 lines
13 KiB
C#
|
|
#region ================== Copyright (c) 2007 Pascal vd Heiden
|
|
|
|
/*
|
|
* Copyright (c) 2007 Pascal vd Heiden, www.codeimp.com
|
|
* This program is released under GNU General Public License
|
|
*
|
|
* This program is distributed in the hope that it will be useful,
|
|
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
* GNU General Public License for more details.
|
|
*
|
|
*/
|
|
|
|
#endregion
|
|
|
|
#region ================== Namespaces
|
|
|
|
using System;
|
|
using System.Globalization;
|
|
using System.Text;
|
|
using System.IO;
|
|
|
|
#endregion
|
|
|
|
namespace CodeImp.DoomBuilder.ZDoom
|
|
{
|
|
public abstract class ZDTextParser
|
|
{
|
|
#region ================== Constants
|
|
|
|
#endregion
|
|
|
|
#region ================== Variables
|
|
|
|
// Parsing
|
|
protected string whitespace = "\n \t\r\u00A0\0"; //mxd. non-breaking space is also space :)
|
|
protected string specialtokens = ":{}+-\n;";
|
|
|
|
// Input data stream
|
|
protected Stream datastream;
|
|
protected BinaryReader datareader;
|
|
protected string sourcename;
|
|
|
|
// Error report
|
|
private int errorline;
|
|
private string errordesc;
|
|
private string errorsource;
|
|
private long prevstreamposition; //mxd. Text stream position storted before poerforming ReadToken.
|
|
|
|
#endregion
|
|
|
|
#region ================== Properties
|
|
|
|
internal Stream DataStream { get { return datastream; } }
|
|
internal BinaryReader DataReader { get { return datareader; } }
|
|
public int ErrorLine { get { return errorline; } }
|
|
public string ErrorDescription { get { return errordesc; } }
|
|
public string ErrorSource { get { return errorsource; } }
|
|
public bool HasError { get { return (errordesc != null); } }
|
|
|
|
#endregion
|
|
|
|
#region ================== Constructor / Disposer
|
|
|
|
// Constructor
|
|
protected ZDTextParser()
|
|
{
|
|
// Initialize
|
|
errordesc = null;
|
|
}
|
|
|
|
#endregion
|
|
|
|
#region ================== Parsing
|
|
|
|
// This parses the given decorate stream
|
|
// Returns false on errors
|
|
public virtual bool Parse(Stream stream, string sourcefilename)
|
|
{
|
|
return Parse(stream, sourcefilename, false);
|
|
}
|
|
|
|
// This parses the given decorate stream (mxd)
|
|
// Returns false on errors
|
|
public virtual bool Parse(Stream stream, string sourcefilename, bool clearerrors)
|
|
{
|
|
// Clear error status (mxd)
|
|
if(clearerrors)
|
|
{
|
|
errordesc = null;
|
|
errorsource = null;
|
|
errorline = -1;
|
|
}
|
|
|
|
datastream = stream;
|
|
datareader = new BinaryReader(stream, Encoding.ASCII);
|
|
sourcename = sourcefilename;
|
|
datastream.Seek(0, SeekOrigin.Begin);
|
|
return true;
|
|
}
|
|
|
|
// This returns true if the given character is whitespace
|
|
private bool IsWhitespace(char c)
|
|
{
|
|
return (whitespace.IndexOf(c) > -1);
|
|
}
|
|
|
|
// This returns true if the given character is a special token
|
|
private bool IsSpecialToken(char c)
|
|
{
|
|
return (specialtokens.IndexOf(c) > -1);
|
|
}
|
|
|
|
// This returns true if the given character is a special token
|
|
protected internal bool IsSpecialToken(string s)
|
|
{
|
|
if(s.Length > 0) return (specialtokens.IndexOf(s[0]) > -1);
|
|
return false;
|
|
}
|
|
|
|
//mxd. This removes beginning and ending quotes from a token
|
|
protected internal string StripTokenQuotes(string token)
|
|
{
|
|
return StripQuotes(token);
|
|
}
|
|
|
|
// This removes beginning and ending quotes from a token
|
|
internal static string StripQuotes(string token)
|
|
{
|
|
// Remove first character, if it is a quote
|
|
if(!string.IsNullOrEmpty(token) && (token[0] == '"'))
|
|
token = token.Substring(1);
|
|
|
|
// Remove last character, if it is a quote
|
|
if(!string.IsNullOrEmpty(token) && (token[token.Length - 1] == '"'))
|
|
token = token.Substring(0, token.Length - 1);
|
|
|
|
return token;
|
|
}
|
|
|
|
// This skips whitespace on the stream, placing the read
|
|
// position right before the first non-whitespace character
|
|
// Returns false when the end of the stream is reached
|
|
protected internal bool SkipWhitespace(bool skipnewline)
|
|
{
|
|
int offset = skipnewline ? 0 : 1;
|
|
char c;
|
|
prevstreamposition = datastream.Position; //mxd
|
|
|
|
do
|
|
{
|
|
if(datastream.Position == datastream.Length) return false;
|
|
c = (char)datareader.ReadByte();
|
|
|
|
// Check if this is comment
|
|
if(c == '/')
|
|
{
|
|
if(datastream.Position == datastream.Length) return false;
|
|
char c2 = (char)datareader.ReadByte();
|
|
if(c2 == '/')
|
|
{
|
|
// Check if not a special comment with a token
|
|
if(datastream.Position == datastream.Length) return false;
|
|
char c3 = (char)datareader.ReadByte();
|
|
if(c3 != '$')
|
|
{
|
|
// Skip entire line
|
|
char c4 = ' ';
|
|
while((c4 != '\n') && (datastream.Position < datastream.Length)) { c4 = (char)datareader.ReadByte(); }
|
|
c = c4;
|
|
}
|
|
else
|
|
{
|
|
// Not a comment
|
|
c = c3;
|
|
}
|
|
}
|
|
else if(c2 == '*')
|
|
{
|
|
// Skip until */
|
|
char c4, c3 = '\0';
|
|
prevstreamposition = datastream.Position; //mxd
|
|
do
|
|
{
|
|
if(datastream.Position == datastream.Length) //mxd
|
|
{
|
|
// ZDoom doesn't give even a warning message about this, so we shouldn't report error or fail parsing.
|
|
General.ErrorLogger.Add(ErrorType.Warning, "DECORATE warning in '" + sourcename + "', line " + GetCurrentLineNumber() + ". Block comment is not closed.");
|
|
return false;
|
|
}
|
|
|
|
c4 = c3;
|
|
c3 = (char)datareader.ReadByte();
|
|
}
|
|
while((c4 != '*') || (c3 != '/'));
|
|
c = ' ';
|
|
}
|
|
else
|
|
{
|
|
// Not a comment, rewind from reading c2
|
|
datastream.Seek(-1, SeekOrigin.Current);
|
|
}
|
|
}
|
|
}
|
|
while(whitespace.IndexOf(c, offset) > -1);
|
|
|
|
// Go one character back so we can read this non-whitespace character again
|
|
datastream.Seek(-1, SeekOrigin.Current);
|
|
return true;
|
|
}
|
|
|
|
// This reads a token (all sequential non-whitespace characters or a single character)
|
|
// Returns null when the end of the stream has been reached
|
|
protected internal string ReadToken()
|
|
{
|
|
//mxd. Return empty string when the end of the stream has been reached
|
|
if(datastream.Position == datastream.Length) return string.Empty;
|
|
|
|
//mxd. Store starting position
|
|
prevstreamposition = datastream.Position;
|
|
|
|
string token = "";
|
|
bool quotedstring = false;
|
|
|
|
// Start reading
|
|
char c = (char)datareader.ReadByte();
|
|
while(!IsWhitespace(c) || quotedstring || IsSpecialToken(c))
|
|
{
|
|
// Special token?
|
|
if(!quotedstring && IsSpecialToken(c))
|
|
{
|
|
// Not reading a token yet?
|
|
if(token.Length == 0)
|
|
{
|
|
// This is our whole token
|
|
token += c;
|
|
break;
|
|
}
|
|
else
|
|
{
|
|
// This is a new token and shouldn't be read now
|
|
// Go one character back so we can read this token again
|
|
datastream.Seek(-1, SeekOrigin.Current);
|
|
break;
|
|
}
|
|
}
|
|
else
|
|
{
|
|
// Quote?
|
|
if(c == '"')
|
|
{
|
|
// Quote to end the string?
|
|
if(quotedstring) quotedstring = false;
|
|
|
|
// First character is a quote?
|
|
if(token.Length == 0) quotedstring = true;
|
|
|
|
token += c;
|
|
}
|
|
// Potential comment?
|
|
else if((c == '/') && !quotedstring)
|
|
{
|
|
// Check the next byte
|
|
if(datastream.Position == datastream.Length) return token;
|
|
char c2 = (char)datareader.ReadByte();
|
|
if((c2 == '/') || (c2 == '*'))
|
|
{
|
|
// This is a comment start, so the token ends here
|
|
// Go two characters back so we can read this comment again
|
|
datastream.Seek(-2, SeekOrigin.Current);
|
|
break;
|
|
}
|
|
else
|
|
{
|
|
// Not a comment
|
|
// Go one character back so we can read this char again
|
|
datastream.Seek(-1, SeekOrigin.Current);
|
|
token += c;
|
|
}
|
|
}
|
|
else
|
|
{
|
|
token += c;
|
|
}
|
|
}
|
|
|
|
// Next character
|
|
if(datastream.Position < datastream.Length)
|
|
c = (char)datareader.Read();
|
|
else
|
|
break;
|
|
}
|
|
|
|
return token;
|
|
}
|
|
|
|
// This reads a token (all sequential non-whitespace characters or a single character) using custom set of special tokens
|
|
// Returns null when the end of the stream has been reached (mxd)
|
|
protected internal string ReadToken(string specialTokens)
|
|
{
|
|
// Return null when the end of the stream has been reached
|
|
if(datastream.Position == datastream.Length) return null;
|
|
|
|
//mxd. Store starting position
|
|
prevstreamposition = datastream.Position;
|
|
|
|
string token = "";
|
|
bool quotedstring = false;
|
|
|
|
// Start reading
|
|
char c = (char)datareader.ReadByte();
|
|
while(!IsWhitespace(c) || quotedstring || specialTokens.IndexOf(c) != -1)
|
|
{
|
|
// Special token?
|
|
if(!quotedstring && specialTokens.IndexOf(c) != -1)
|
|
{
|
|
// Not reading a token yet?
|
|
if(token.Length == 0)
|
|
{
|
|
// This is our whole token
|
|
token += c;
|
|
break;
|
|
}
|
|
|
|
// This is a new token and shouldn't be read now
|
|
// Go one character back so we can read this token again
|
|
datastream.Seek(-1, SeekOrigin.Current);
|
|
break;
|
|
}
|
|
else
|
|
{
|
|
// Quote?
|
|
if(c == '"')
|
|
{
|
|
// Quote to end the string?
|
|
if(quotedstring) quotedstring = false;
|
|
|
|
// First character is a quote?
|
|
if(token.Length == 0) quotedstring = true;
|
|
|
|
token += c;
|
|
}
|
|
// Potential comment?
|
|
else if((c == '/') && !quotedstring)
|
|
{
|
|
// Check the next byte
|
|
if(datastream.Position == datastream.Length) return token;
|
|
char c2 = (char)datareader.ReadByte();
|
|
if((c2 == '/') || (c2 == '*'))
|
|
{
|
|
// This is a comment start, so the token ends here
|
|
// Go two characters back so we can read this comment again
|
|
datastream.Seek(-2, SeekOrigin.Current);
|
|
break;
|
|
}
|
|
else
|
|
{
|
|
// Not a comment
|
|
// Go one character back so we can read this char again
|
|
datastream.Seek(-1, SeekOrigin.Current);
|
|
token += c;
|
|
}
|
|
}
|
|
else
|
|
{
|
|
token += c;
|
|
}
|
|
}
|
|
|
|
// Next character
|
|
if(datastream.Position < datastream.Length)
|
|
c = (char)datareader.Read();
|
|
else
|
|
break;
|
|
}
|
|
|
|
return token;
|
|
}
|
|
|
|
// This reads the rest of the line
|
|
// Returns null when the end of the stream has been reached
|
|
protected internal string ReadLine()
|
|
{
|
|
string token = "";
|
|
|
|
// Return null when the end of the stream has been reached
|
|
if(datastream.Position == datastream.Length) return null;
|
|
|
|
// Start reading
|
|
char c = (char)datareader.ReadByte();
|
|
while(c != '\n')
|
|
{
|
|
token += c;
|
|
|
|
// Next character
|
|
if(datastream.Position < datastream.Length)
|
|
c = (char)datareader.Read();
|
|
else
|
|
break;
|
|
}
|
|
|
|
return token.Trim();
|
|
}
|
|
|
|
//mxd
|
|
protected bool NextTokenIs(string expectedtoken)
|
|
{
|
|
return NextTokenIs(expectedtoken, true);
|
|
}
|
|
|
|
//mxd
|
|
protected bool NextTokenIs(string expectedtoken, bool reporterror)
|
|
{
|
|
if(!SkipWhitespace(true)) return false;
|
|
string token = ReadToken();
|
|
|
|
if(token != expectedtoken)
|
|
{
|
|
if(reporterror) General.ErrorLogger.Add(ErrorType.Error, "Error in '" + sourcename + "' at line " + GetCurrentLineNumber() + ": expected '" + expectedtoken + "', but got '" + token + "'");
|
|
|
|
// Rewind so this structure can be read again
|
|
DataStream.Seek(-token.Length - 1, SeekOrigin.Current);
|
|
return false;
|
|
}
|
|
|
|
return true;
|
|
}
|
|
|
|
//mxd
|
|
protected internal bool ReadSignedFloat(string token, ref float value)
|
|
{
|
|
int sign = 1;
|
|
if (token == "-")
|
|
{
|
|
sign = -1;
|
|
token = StripTokenQuotes(ReadToken());
|
|
}
|
|
|
|
float val;
|
|
bool success = float.TryParse(token, NumberStyles.Float, CultureInfo.InvariantCulture, out val);
|
|
if (success) value = val * sign;
|
|
return success;
|
|
}
|
|
|
|
//mxd
|
|
protected internal bool ReadSignedInt(string token, ref int value)
|
|
{
|
|
int sign = 1;
|
|
if (token == "-")
|
|
{
|
|
sign = -1;
|
|
token = StripTokenQuotes(ReadToken());
|
|
}
|
|
|
|
int val;
|
|
bool success = int.TryParse(token, NumberStyles.Integer, CultureInfo.InvariantCulture, out val);
|
|
if (success) value = val * sign;
|
|
return success;
|
|
}
|
|
|
|
// This reports an error
|
|
protected internal void ReportError(string message)
|
|
{
|
|
// Set error information
|
|
errordesc = message;
|
|
errorline = GetCurrentLineNumber();
|
|
errorsource = sourcename;
|
|
}
|
|
|
|
//mxd
|
|
protected internal int GetCurrentLineNumber()
|
|
{
|
|
long position = Math.Min(prevstreamposition, datastream.Position);
|
|
long readpos = 0;
|
|
int linenumber = 0;
|
|
|
|
// Find the line on which we found this error
|
|
datastream.Seek(0, SeekOrigin.Begin);
|
|
StreamReader textreader = new StreamReader(datastream, Encoding.ASCII);
|
|
while (readpos < position)
|
|
{
|
|
string line = textreader.ReadLine();
|
|
if (line == null) break;
|
|
readpos += line.Length + 2;
|
|
linenumber++;
|
|
}
|
|
|
|
// Return to original position
|
|
datastream.Seek(position, SeekOrigin.Begin);
|
|
|
|
return linenumber;
|
|
}
|
|
|
|
#endregion
|
|
}
|
|
}
|