jkxr/Projects/Android/jni/OpenJK/code/game/genericparser2.cpp
Simon 4597b03873 Initial Commit
Opens in Android Studio but haven't even tried to build it yet (it won't.. I know that much!)
2022-09-18 16:37:21 +01:00

310 lines
6.7 KiB
C++

/*
===========================================================================
Copyright (C) 2000 - 2013, Raven Software, Inc.
Copyright (C) 2001 - 2013, Activision, Inc.
Copyright (C) 2013 - 2015, OpenJK contributors
This file is part of the OpenJK source code.
OpenJK is free software; you can redistribute it and/or modify it
under the terms of the GNU General Public License version 2 as
published by the Free Software Foundation.
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.
You should have received a copy of the GNU General Public License
along with this program; if not, see <http://www.gnu.org/licenses/>.
===========================================================================
*/
// Filename:- genericparser2.cpp
#include "common_headers.h"
#include "genericparser2.h"
#ifdef _JK2EXE
#include "../qcommon/qcommon.h"
#endif
#include <algorithm>
#include <cctype>
static void skipWhitespace( gsl::cstring_view& text, const bool allowLineBreaks )
{
gsl::cstring_view::iterator whitespaceEnd = text.begin();
while( whitespaceEnd != text.end() // No EOF
&& std::isspace( *whitespaceEnd ) // No End of Whitespace
&& ( allowLineBreaks || *whitespaceEnd != '\n' ) ) // No unwanted newline
{
++whitespaceEnd;
}
text = { whitespaceEnd, text.end() };
}
static void skipWhitespaceAndComments( gsl::cstring_view& text, const bool allowLineBreaks )
{
skipWhitespace( text, allowLineBreaks );
// skip single line comment
if( text.size() >= 2 && text[ 0 ] == '/' && text[ 1 ] == '/' )
{
auto commentEnd = std::find( text.begin() + 2, text.end(), '\n' );
if( commentEnd == text.end() )
{
text = { text.end(), text.end() };
return;
}
else
{
text = { commentEnd, text.end() };
skipWhitespaceAndComments( text, allowLineBreaks );
return;
}
}
// skip multi line comments
if( text.size() >= 2 && text[ 0 ] == '/' && text[ 1 ] == '*' )
{
static const std::array< char, 2 > endStr{ '*', '/' };
auto commentEnd = std::search( text.begin(), text.end(), endStr.begin(), endStr.end() );
if( commentEnd == text.end() )
{
text = { text.end(), text.end() };
return;
}
else
{
text = { commentEnd + endStr.size(), text.end() };
skipWhitespace( text, allowLineBreaks );
return;
}
}
// found start of token
return;
}
static gsl::cstring_view removeTrailingWhitespace( const gsl::cstring_view& text )
{
return{
text.begin(),
std::find_if_not(
std::reverse_iterator< const char *>( text.end() ), std::reverse_iterator< const char* >( text.begin() ),
static_cast< int( *)( int ) >( std::isspace )
).base()
};
}
/**
Skips whitespace (including lineBreaks, if desired) & comments, then reads one token.
A token can be:
- a string ("" delimited; ignores readToEOL)
- whitespace-delimited (if readToEOL == false)
- EOL- or comment-delimited (if readToEOL == true); i.e. reads to end of line or the first // or /*
@param text adjusted to start beyond the read token
*/
static gsl::cstring_view GetToken( gsl::cstring_view& text, bool allowLineBreaks, bool readToEOL = false )
{
skipWhitespaceAndComments( text, allowLineBreaks );
// EOF
if( text.empty() )
{
return{};
}
// string. ignores readToEOL.
if( text[ 0 ] == '"' )
{
// there are no escapes, string just ends at the next "
auto tokenEnd = std::find( text.begin() + 1, text.end(), '"' );
if( tokenEnd == text.end() )
{
gsl::cstring_view token = { text.begin() + 1, text.end() };
text = { text.end(), text.end() };
return token;
}
else
{
gsl::cstring_view token = { text.begin() + 1, tokenEnd };
text = { tokenEnd + 1, text.end() };
return token;
}
}
else if( readToEOL )
{
// find the first of '\n', "//" or "/*"; that's end of token
auto tokenEnd = std::find( text.begin(), text.end(), '\n' );
static const std::array< char, 2 > commentPatterns[]{
{ { '/', '*' } },
{ { '/', '/' } }
};
for( auto& pattern : commentPatterns )
{
tokenEnd = std::min(
tokenEnd,
std::search(
text.begin(), tokenEnd,
pattern.begin(), pattern.end()
)
);
}
gsl::cstring_view token{ text.begin(), tokenEnd };
text = { tokenEnd, text.end() };
return removeTrailingWhitespace( token );
}
else
{
// consume until first whitespace (if allowLineBreaks == false, that may be text.begin(); in that case token is empty.)
auto tokenEnd = std::find_if( text.begin(), text.end(), static_cast< int( *)( int ) >( std::isspace ) );
gsl::cstring_view token{ text.begin(), tokenEnd };
text = { tokenEnd, text.end() };
return token;
}
}
CGPProperty::CGPProperty( gsl::cstring_view initKey, gsl::cstring_view initValue )
: mKey( initKey )
{
if( !initValue.empty() )
{
mValues.push_back( initValue );
}
}
void CGPProperty::AddValue( gsl::cstring_view newValue )
{
mValues.push_back( newValue );
}
CGPGroup::CGPGroup( const gsl::cstring_view& initName )
: mName( initName )
{
}
bool CGPGroup::Parse( gsl::cstring_view& data, const bool topLevel )
{
while( true )
{
gsl::cstring_view token = GetToken( data, true );
if( token.empty() )
{
if ( topLevel )
{
// top level parse; there was no opening "{", so there should be no closing one either.
return true;
}
else
{
// end of data - error!
return false;
}
}
else if( token == CSTRING_VIEW( "}" ) )
{
if( topLevel )
{
// top-level group; there was no opening "{" so there should be no closing one, either.
return false;
}
else
{
// ending brace for this group
return true;
}
}
gsl::cstring_view lastToken = token;
// read ahead to see what we are doing
token = GetToken( data, true, true );
if( token == CSTRING_VIEW( "{" ) )
{
// new sub group
mSubGroups.emplace_back( lastToken );
if( !mSubGroups.back().Parse( data, false ) )
{
return false;
}
}
else if( token == CSTRING_VIEW( "[" ) )
{
// new list
mProperties.emplace_back( lastToken );
CGPProperty& list = mProperties.back();
while( true )
{
token = GetToken( data, true, true );
if( token.empty() )
{
return false;
}
if( token == CSTRING_VIEW( "]" ) )
{
break;
}
list.AddValue( token );
}
}
else
{
// new value
mProperties.emplace_back( lastToken, token );
}
}
}
bool CGenericParser2::Parse( gsl::czstring filename )
{
Clear();
mFileContent = FS::ReadFile( filename );
if( !mFileContent.valid() )
{
return false;
}
auto view = mFileContent.view();
return mTopLevel.Parse( view );
}
void CGenericParser2::Clear() NOEXCEPT
{
mTopLevel.Clear();
}
//////////////////// eof /////////////////////