kicad/common/string.cpp

911 lines
22 KiB
C++
Raw Normal View History

/*
* This program source code file is part of KiCad, a free EDA CAD application.
*
* Copyright (C) 2004-2019 KiCad Developers, see AUTHORS.txt for contributors.
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public License
* as published by the Free Software Foundation; either version 2
* of the License, or (at your option) any later version.
*
* 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, you may find one here:
* http://www.gnu.org/licenses/old-licenses/gpl-2.0.html
* or you may search the http://www.gnu.org website for the version 2 license,
* or you may write to the Free Software Foundation, Inc.,
* 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA
*/
/**
* @file string.cpp
* @brief Some useful functions to handle strings.
*/
2007-05-06 16:03:28 +00:00
#include <clocale>
#include <cmath>
#include <macros.h>
#include <richio.h> // StrPrintf
#include <kicad_string.h>
/**
* Illegal file name characters used to insure file names will be valid on all supported
* platforms. This is the list of illegal file name characters for Windows which includes
* the illegal file name characters for Linux and OSX.
*/
static const char illegalFileNameChars[] = "\\/:\"<>|";
bool ConvertSmartQuotesAndDashes( wxString* aString )
{
bool retVal = false;
for( wxString::iterator ii = aString->begin(); ii != aString->end(); ++ii )
{
if( *ii == L'\u00B4' || *ii == L'\u2018' || *ii == L'\u2019' )
{
*ii = '\'';
retVal = true;
}
if( *ii == L'\u201C' || *ii == L'\u201D' )
{
*ii = '"';
retVal = true;
}
if( *ii == L'\u2013' || *ii == L'\u2014' )
{
*ii = '-';
retVal = true;
}
}
return retVal;
}
/**
* These Escape/Unescape routines use HTML-entity-reference-style encoding to handle
* characters which are:
* (a) not legal in filenames
* (b) used as control characters in LIB_IDs
* (c) used to delineate hierarchical paths
*/
wxString EscapeString( const wxString& aSource, ESCAPE_CONTEXT aContext )
{
wxString converted;
for( wxUniChar c: aSource )
{
if( aContext == CTX_NETNAME )
{
if( c == '/' )
converted += "{slash}";
else if( c == '\n' || c == '\r' )
converted += ""; // drop
else
converted += c;
}
else if( aContext == CTX_LIBID )
{
if( c == '{' )
converted += "{brace}";
else if( c == ':' )
converted += "{colon}";
else if( c == '\n' || c == '\r' )
converted += ""; // drop
else
converted += c;
}
else if( aContext == CTX_QUOTED_STR )
{
if( c == '\"' )
converted += "{dblquote}";
else
converted += c;
}
else if( aContext == CTX_LINE )
{
if( c == '\n' || c == '\r' )
converted += "{return}";
else
converted += c;
}
else if( aContext == CTX_FILENAME )
{
if( c == '{' )
converted += "{brace}";
else if( c == '/' )
converted += "{slash}";
else if( c == '\\' )
converted += "{backslash}";
else if( c == '\"' )
converted += "{dblquote}";
else if( c == '<' )
converted += "{lt}";
else if( c == '>' )
converted += "{gt}";
else if( c == '|' )
converted += "{bar}";
else if( c == ':' )
converted += "{colon}";
else if( c == '\t' )
converted += "{tab}";
else if( c == '\n' || c == '\r' )
converted += "{return}";
else
converted += c;
}
else
converted += c;
}
return converted;
}
wxString UnescapeString( const wxString& aSource )
{
wxString newbuf;
size_t sourceLen = aSource.length();
for( size_t i = 0; i < sourceLen; ++i )
{
if( ( aSource[i] == '$' || aSource[i] == '^' || aSource[i] == '_' )
&& i + 1 < sourceLen && aSource[i+1] == '{' )
{
for( ; i < sourceLen; ++i )
{
newbuf += aSource[i];
if( aSource[i] == '}' )
break;
}
}
else if( aSource[i] == '{' )
{
wxString token;
int depth = 1;
for( i = i + 1; i < sourceLen; ++i )
{
if( aSource[i] == '{' )
depth++;
else if( aSource[i] == '}' )
depth--;
if( depth <= 0 )
break;
else
token.append( aSource[i] );
}
if( token == wxS( "dblquote" ) ) newbuf.append( wxS( "\"" ) );
else if( token == wxS( "quote" ) ) newbuf.append( wxS( "'" ) );
else if( token == wxS( "lt" ) ) newbuf.append( wxS( "<" ) );
else if( token == wxS( "gt" ) ) newbuf.append( wxS( ">" ) );
else if( token == wxS( "backslash" ) ) newbuf.append( wxS( "\\" ) );
else if( token == wxS( "slash" ) ) newbuf.append( wxS( "/" ) );
else if( token == wxS( "bar" ) ) newbuf.append( wxS( "|" ) );
else if( token == wxS( "colon" ) ) newbuf.append( wxS( ":" ) );
else if( token == wxS( "space" ) ) newbuf.append( wxS( " " ) );
else if( token == wxS( "dollar" ) ) newbuf.append( wxS( "$" ) );
else if( token == wxS( "tab" ) ) newbuf.append( wxS( "\t" ) );
else if( token == wxS( "return" ) ) newbuf.append( wxS( "\n" ) );
else if( token == wxS( "brace" ) ) newbuf.append( wxS( "{" ) );
else if( token.IsEmpty() ) newbuf.append( wxS( "{" ) );
else
{
newbuf.append( "{" + UnescapeString( token ) + "}" );
}
}
else
{
newbuf.append( aSource[i] );
}
}
return newbuf;
}
wxString TitleCaps( const wxString& aString )
{
wxArrayString words;
wxString result;
wxStringSplit( aString, words, ' ' );
for( const wxString& word : words )
{
if( !result.IsEmpty() )
result += wxT( " " );
result += word.Capitalize();
}
return result;
}
int ReadDelimitedText( wxString* aDest, const char* aSource )
{
std::string utf8; // utf8 but without escapes and quotes.
bool inside = false;
const char* start = aSource;
char cc;
while( (cc = *aSource++) != 0 )
{
if( cc == '"' )
{
if( inside )
break; // 2nd double quote is end of delimited text
inside = true; // first delimiter found, make note, do not copy
}
else if( inside )
{
if( cc == '\\' )
{
cc = *aSource++;
if( !cc )
break;
// do no copy the escape byte if it is followed by \ or "
if( cc != '"' && cc != '\\' )
utf8 += '\\';
utf8 += cc;
}
else
{
utf8 += cc;
}
}
}
*aDest = FROM_UTF8( utf8.c_str() );
return aSource - start;
}
int ReadDelimitedText( char* aDest, const char* aSource, int aDestSize )
{
if( aDestSize <= 0 )
return 0;
bool inside = false;
const char* start = aSource;
char* limit = aDest + aDestSize - 1;
char cc;
while( (cc = *aSource++) != 0 && aDest < limit )
{
2011-02-27 08:16:05 +00:00
if( cc == '"' )
{
if( inside )
break; // 2nd double quote is end of delimited text
inside = true; // first delimiter found, make note, do not copy
}
else if( inside )
{
2011-02-27 08:16:05 +00:00
if( cc == '\\' )
{
cc = *aSource++;
if( !cc )
break;
2011-02-27 08:16:05 +00:00
// do no copy the escape byte if it is followed by \ or "
if( cc != '"' && cc != '\\' )
*aDest++ = '\\';
if( aDest < limit )
*aDest++ = cc;
}
else
{
2011-02-27 08:16:05 +00:00
*aDest++ = cc;
}
}
}
*aDest = 0;
return aSource - start;
}
std::string EscapedUTF8( wxString aString )
2007-05-06 16:03:28 +00:00
{
// No new-lines allowed in quoted strings
aString.Replace( "\r\n", "\r" );
aString.Replace( "\n", "\r" );
std::string utf8 = TO_UTF8( aString );
2007-12-01 05:53:52 +00:00
std::string ret;
2011-03-22 18:29:26 +00:00
ret += '"';
for( std::string::const_iterator it = utf8.begin(); it!=utf8.end(); ++it )
2007-12-01 05:53:52 +00:00
{
// this escaping strategy is designed to be compatible with ReadDelimitedText():
if( *it == '"' )
2007-12-01 05:53:52 +00:00
{
ret += '\\';
ret += '"';
2007-12-01 05:53:52 +00:00
}
else if( *it == '\\' )
2007-12-01 05:53:52 +00:00
{
ret += '\\'; // double it up
ret += '\\';
2007-12-01 05:53:52 +00:00
}
else
{
ret += *it;
}
2007-12-01 05:53:52 +00:00
}
2011-03-22 18:29:26 +00:00
ret += '"';
return ret;
2007-05-06 16:03:28 +00:00
}
wxString EscapeHTML( const wxString& aString )
{
wxString converted;
for( wxUniChar c: aString )
{
if( c == '\"' )
converted += "&quot;";
else if( c == '\'' )
converted += "&apos;";
else if( c == '&' )
converted += "&amp;";
else if( c == '<' )
converted += "&lt;";
else if( c == '>' )
converted += "&gt;";
else
converted += c;
}
return converted;
}
bool NoPrintableChars( wxString aString )
{
return aString.Trim( true ).Trim( false ).IsEmpty();
}
char* StrPurge( char* text )
2007-05-06 16:03:28 +00:00
{
2011-01-16 13:14:08 +00:00
static const char whitespace[] = " \t\n\r\f\v";
2007-12-01 05:53:52 +00:00
2011-01-16 05:11:17 +00:00
if( text )
{
while( *text && strchr( whitespace, *text ) )
2011-01-16 05:11:17 +00:00
++text;
2007-12-01 05:53:52 +00:00
2011-01-16 05:11:17 +00:00
char* cp = text + strlen( text ) - 1;
2007-12-01 05:53:52 +00:00
2011-01-16 05:11:17 +00:00
while( cp >= text && strchr( whitespace, *cp ) )
*cp-- = '\0';
2007-12-01 05:53:52 +00:00
}
return text;
2007-05-06 16:03:28 +00:00
}
char* GetLine( FILE* File, char* Line, int* LineNum, int SizeLine )
2007-05-06 16:03:28 +00:00
{
do {
2007-12-01 05:53:52 +00:00
if( fgets( Line, SizeLine, File ) == NULL )
return NULL;
2007-12-01 05:53:52 +00:00
if( LineNum )
*LineNum += 1;
} while( Line[0] == '#' || Line[0] == '\n' || Line[0] == '\r' || Line[0] == 0 );
2007-12-01 05:53:52 +00:00
strtok( Line, "\n\r" );
return Line;
2007-05-06 16:03:28 +00:00
}
wxString DateAndTime()
2007-05-06 16:03:28 +00:00
{
wxDateTime datetime = wxDateTime::Now();
datetime.SetCountry( wxDateTime::Country_Default );
return datetime.Format( wxDefaultDateTimeFormat, wxDateTime::Local );
2007-05-06 16:03:28 +00:00
}
int StrNumCmp( const wxString& aString1, const wxString& aString2, bool aIgnoreCase )
2007-05-06 16:03:28 +00:00
{
2007-12-01 05:53:52 +00:00
int nb1 = 0, nb2 = 0;
auto str1 = aString1.begin();
auto str2 = aString2.begin();
while( str1 != aString1.end() && str2 != aString2.end() )
2007-12-01 05:53:52 +00:00
{
wxUniChar c1 = *str1;
wxUniChar c2 = *str2;
if( wxIsdigit( c1 ) && wxIsdigit( c2 ) ) // Both characters are digits, do numeric compare.
2007-12-01 05:53:52 +00:00
{
nb1 = 0;
nb2 = 0;
do
2007-12-01 05:53:52 +00:00
{
c1 = *str1;
nb1 = nb1 * 10 + (int) c1 - '0';
++str1;
} while( str1 != aString1.end() && wxIsdigit( *str1 ) );
2007-12-01 05:53:52 +00:00
do
2007-12-01 05:53:52 +00:00
{
c2 = *str2;
nb2 = nb2 * 10 + (int) c2 - '0';
++str2;
} while( str2 != aString2.end() && wxIsdigit( *str2 ) );
2007-12-01 05:53:52 +00:00
if( nb1 < nb2 )
return -1;
if( nb1 > nb2 )
2007-12-01 05:53:52 +00:00
return 1;
c1 = ( str1 != aString1.end() ) ? *str1 : wxUniChar( 0 );
c2 = ( str2 != aString2.end() ) ? *str2 : wxUniChar( 0 );
2007-12-01 05:53:52 +00:00
}
// Any numerical comparisons to here are identical.
if( aIgnoreCase )
2007-12-01 05:53:52 +00:00
{
if( wxToupper( c1 ) < wxToupper( c2 ) )
return -1;
if( wxToupper( c1 ) > wxToupper( c2 ) )
return 1;
}
else
{
if( c1 < c2 )
2007-12-01 05:53:52 +00:00
return -1;
if( c1 > c2 )
2007-12-01 05:53:52 +00:00
return 1;
}
if( str1 != aString1.end() )
++str1;
if( str2 != aString2.end() )
++str2;
}
if( str1 == aString1.end() && str2 != aString2.end() )
{
return -1; // Identical to here but aString1 is longer.
}
else if( str1 != aString1.end() && str2 == aString2.end() )
{
return 1; // Identical to here but aString2 is longer.
2007-12-01 05:53:52 +00:00
}
return 0;
2007-05-06 16:03:28 +00:00
}
bool WildCompareString( const wxString& pattern, const wxString& string_to_tst,
bool case_sensitive )
2007-05-06 16:03:28 +00:00
{
2007-12-01 05:53:52 +00:00
const wxChar* cp = NULL, * mp = NULL;
const wxChar* wild, * string;
wxString _pattern, _string_to_tst;
if( case_sensitive )
{
wild = pattern.GetData();
string = string_to_tst.GetData();
2007-12-01 05:53:52 +00:00
}
else
{
_pattern = pattern;
_pattern.MakeUpper();
_string_to_tst = string_to_tst;
_string_to_tst.MakeUpper();
wild = _pattern.GetData();
string = _string_to_tst.GetData();
2007-12-01 05:53:52 +00:00
}
while( ( *string ) && ( *wild != '*' ) )
2007-12-01 05:53:52 +00:00
{
if( ( *wild != *string ) && ( *wild != '?' ) )
return false;
2007-12-01 05:53:52 +00:00
wild++; string++;
}
while( *string )
{
if( *wild == '*' )
{
if( !*++wild )
2007-12-01 05:53:52 +00:00
return 1;
mp = wild;
cp = string + 1;
}
else if( ( *wild == *string ) || ( *wild == '?' ) )
2007-12-01 05:53:52 +00:00
{
wild++;
string++;
}
else
{
wild = mp;
string = cp++;
}
}
while( *wild == '*' )
{
wild++;
}
return !*wild;
2007-05-06 16:03:28 +00:00
}
bool ApplyModifier( double& value, const wxString& aString )
{
static const wxString modifiers( wxT( "pnumkKM" ) );
if( !aString.length() )
return false;
wxChar modifier;
wxString units;
if( modifiers.Find( aString[ 0 ] ) >= 0 )
{
modifier = aString[ 0 ];
units = aString.Mid( 1 ).Trim();
}
else
{
modifier = ' ';
units = aString.Mid( 0 ).Trim();
}
if( units.length()
&& !units.CmpNoCase( wxT( "F" ) )
&& !units.CmpNoCase( wxT( "hz" ) )
&& !units.CmpNoCase( wxT( "W" ) )
&& !units.CmpNoCase( wxT( "V" ) )
&& !units.CmpNoCase( wxT( "H" ) ) )
return false;
if( modifier == 'p' )
value *= 1.0e-12;
if( modifier == 'n' )
value *= 1.0e-9;
else if( modifier == 'u' )
value *= 1.0e-6;
else if( modifier == 'm' )
value *= 1.0e-3;
else if( modifier == 'k' || modifier == 'K' )
value *= 1.0e3;
else if( modifier == 'M' )
value *= 1.0e6;
else if( modifier == 'G' )
value *= 1.0e9;
return true;
}
int ValueStringCompare( wxString strFWord, wxString strSWord )
{
// Compare unescaped text
strFWord = UnescapeString( strFWord );
strSWord = UnescapeString( strSWord );
// The different sections of the two strings
wxString strFWordBeg, strFWordMid, strFWordEnd;
wxString strSWordBeg, strSWordMid, strSWordEnd;
// Split the two strings into separate parts
SplitString( strFWord, &strFWordBeg, &strFWordMid, &strFWordEnd );
SplitString( strSWord, &strSWordBeg, &strSWordMid, &strSWordEnd );
// Compare the Beginning section of the strings
int isEqual = strFWordBeg.CmpNoCase( strSWordBeg );
if( isEqual > 0 )
return 1;
else if( isEqual < 0 )
return -1;
else
{
// If the first sections are equal compare their digits
double lFirstNumber = 0;
double lSecondNumber = 0;
bool endingIsModifier = false;
strFWordMid.ToDouble( &lFirstNumber );
strSWordMid.ToDouble( &lSecondNumber );
endingIsModifier |= ApplyModifier( lFirstNumber, strFWordEnd );
endingIsModifier |= ApplyModifier( lSecondNumber, strSWordEnd );
if( lFirstNumber > lSecondNumber )
return 1;
else if( lFirstNumber < lSecondNumber )
return -1;
// If the first two sections are equal and the endings are modifiers then compare them
else if( !endingIsModifier )
return strFWordEnd.CmpNoCase( strSWordEnd );
// Ran out of things to compare; they must match
else
return 0;
}
}
int SplitString( wxString strToSplit,
wxString* strBeginning,
wxString* strDigits,
wxString* strEnd )
{
static const wxString separators( wxT( ".," ) );
// Clear all the return strings
strBeginning->Empty();
strDigits->Empty();
strEnd->Empty();
// There no need to do anything if the string is empty
if( strToSplit.length() == 0 )
return 0;
// Starting at the end of the string look for the first digit
int ii;
for( ii = (strToSplit.length() - 1); ii >= 0; ii-- )
{
if( wxIsdigit( strToSplit[ii] ) )
break;
}
// If there were no digits then just set the single string
if( ii < 0 )
{
*strBeginning = strToSplit;
}
else
{
// Since there is at least one digit this is the trailing string
*strEnd = strToSplit.substr( ii + 1 );
// Go to the end of the digits
int position = ii + 1;
for( ; ii >= 0; ii-- )
{
if( !wxIsdigit( strToSplit[ii] ) && separators.Find( strToSplit[ii] ) < 0 )
break;
}
// If all that was left was digits, then just set the digits string
if( ii < 0 )
*strDigits = strToSplit.substr( 0, position );
/* We were only looking for the last set of digits everything else is
* part of the preamble */
else
{
*strDigits = strToSplit.substr( ii + 1, position - ii - 1 );
*strBeginning = strToSplit.substr( 0, ii + 1 );
}
}
return 0;
}
int GetTrailingInt( const wxString& aStr )
{
int number = 0;
int base = 1;
// Trim and extract the trailing numeric part
int index = aStr.Len() - 1;
while( index >= 0 )
{
const char chr = aStr.GetChar( index );
if( chr < '0' || chr > '9' )
break;
number += ( chr - '0' ) * base;
base *= 10;
index--;
}
return number;
}
wxString GetIllegalFileNameWxChars()
{
return FROM_UTF8( illegalFileNameChars );
}
bool ReplaceIllegalFileNameChars( std::string* aName, int aReplaceChar )
{
bool changed = false;
std::string result;
result.reserve( aName->length() );
for( std::string::iterator it = aName->begin(); it != aName->end(); ++it )
{
if( strchr( illegalFileNameChars, *it ) )
{
if( aReplaceChar )
StrPrintf( &result, "%c", aReplaceChar );
else
StrPrintf( &result, "%%%02x", *it );
changed = true;
}
else
{
result += *it;
}
}
if( changed )
*aName = result;
return changed;
}
bool ReplaceIllegalFileNameChars( wxString& aName, int aReplaceChar )
{
bool changed = false;
wxString result;
result.reserve( aName.Length() );
wxString illWChars = GetIllegalFileNameWxChars();
for( wxString::iterator it = aName.begin(); it != aName.end(); ++it )
{
if( illWChars.Find( *it ) != wxNOT_FOUND )
{
if( aReplaceChar )
result += aReplaceChar;
else
result += wxString::Format( "%%%02x", *it );
changed = true;
}
else
{
result += *it;
}
}
if( changed )
aName = result;
return changed;
}
2020-10-24 02:06:42 +00:00
void wxStringSplit( const wxString& aText, wxArrayString& aStrings, wxChar aSplitter )
{
wxString tmp;
for( unsigned ii = 0; ii < aText.Length(); ii++ )
{
if( aText[ii] == aSplitter )
{
aStrings.Add( tmp );
tmp.Clear();
}
else
tmp << aText[ii];
}
if( !tmp.IsEmpty() )
{
aStrings.Add( tmp );
}
}
void StripTrailingZeros( wxString& aStringValue, unsigned aTrailingZeroAllowed )
{
struct lconv* lc = localeconv();
char sep = lc->decimal_point[0];
unsigned sep_pos = aStringValue.Find( sep );
if( sep_pos > 0 )
{
// We want to keep at least aTrailingZeroAllowed digits after the separator
unsigned min_len = sep_pos + aTrailingZeroAllowed + 1;
while( aStringValue.Len() > min_len )
{
if( aStringValue.Last() == '0' )
aStringValue.RemoveLast();
else
break;
}
}
}
std::string Double2Str( double aValue )
{
char buf[50];
int len;
if( aValue != 0.0 && std::fabs( aValue ) <= 0.0001 )
{
// For these small values, %f works fine,
// and %g gives an exponent
len = sprintf( buf, "%.16f", aValue );
while( --len > 0 && buf[len] == '0' )
buf[len] = '\0';
if( buf[len] == '.' )
buf[len] = '\0';
else
++len;
}
else
{
// For these values, %g works fine, and sometimes %f
// gives a bad value (try aValue = 1.222222222222, with %.16f format!)
len = sprintf( buf, "%.10g", aValue );
}
return std::string( buf, len );
}
wxString AngleToStringDegrees( double aAngle )
{
wxString text;
text.Printf( wxT( "%.3f" ), aAngle / 10.0 );
StripTrailingZeros( text, 1 );
return text;
}