kicad/common/string.cpp

323 lines
6.9 KiB
C++

/*********************************************/
/* string.cpp */
/* some useful functions to handle strings */
/*********************************************/
#include "fctsys.h"
#include "macros.h"
#include "kicad_string.h"
/* read a double-quote delimited text from source and put it in in dest,
* read NbMaxChar bytes max
* return the char count read from source
*/
int ReadDelimitedText( char* dest, char* source, int NbMaxChar )
{
int ii, jj, flag = 0;
for( ii = 0, jj = 0; ii < NbMaxChar - 1; jj++, source++ )
{
if( *source == 0 )
break; /* E.O.L. */
if( *source == '"' ) /* delimiter is " */
{
if( flag )
break; /* End of delimited text */
flag = 1; /* First delimiter found. */
}
else if( flag )
{
*dest = *source; dest++; ii++;
}
}
*dest = 0; /* Null terminated */
return jj;
}
/* Remove training spaces in text
* return a pointer on the first non space char in text
*/
char* StrPurge( char* text )
{
char* ptspace;
if( text == NULL )
return NULL;
while( ( *text <= ' ' ) && *text )
text++;
ptspace = text + strlen( text ) - 1;
while( ( *ptspace <= ' ' ) && *ptspace && ( ptspace >= text ) )
{
*ptspace = 0; ptspace--;
}
return text;
}
/* Read lines from File
* Skip void lines and comments (starting by #)
* return the first non void line.
* increments *LineNum for each line
*/
char* GetLine( FILE* File, char* Line, int* LineNum, int SizeLine )
{
do {
if( fgets( Line, SizeLine, File ) == NULL )
return NULL;
if( LineNum )
*LineNum += 1;
} while( Line[0] == '#' || Line[0] == '\n' || Line[0] == '\r'
|| Line[0] == 0 );
strtok( Line, "\n\r" );
return Line;
}
/* return in aBuffer the date and time
* time is the local time.
*/
char* DateAndTime( char* aBuffer )
{
wxString datetime;
datetime = DateAndTime();
strcpy( aBuffer, CONV_TO_UTF8( datetime ) );
return aBuffer;
}
/* return the date and time in a wxString
* note: does the same thing than strftime()
* time is the local time.
*/
wxString DateAndTime()
{
wxString Line;
wxDateTime datetime = wxDateTime::Now();
datetime.SetCountry( wxDateTime::Country_Default );
Line = datetime.Format( wxDefaultDateTimeFormat, wxDateTime::Local );
return Line;
}
/*
* sort() function
* Same as strncmp() but numbers in strings
* are compared according to the value, not the ascii value of each digit
*/
int StrLenNumCmp( const wxChar* str1, const wxChar* str2, int NbMax )
{
int i;
int nb1 = 0, nb2 = 0;
if( ( str1 == NULL ) || ( str2 == NULL ) )
return 0;
for( i = 0; i < NbMax; i++ )
{
if( isdigit( *str1 ) && isdigit( *str2 ) ) /* digit found */
{
nb1 = 0; nb2 = 0;
while( isdigit( *str1 ) )
{
nb1 = nb1 * 10 + *str1 - '0'; str1++;
}
while( isdigit( *str2 ) )
{
nb2 = nb2 * 10 + *str2 - '0'; str2++;
}
if( nb1 < nb2 )
return -1;
if( nb1 > nb2 )
return 1;
}
if( *str1 < *str2 )
return -1;
if( *str1 > *str2 )
return 1;
if( ( *str1 == 0 ) && ( *str2 == 0 ) )
return 0;
str1++; str2++;
}
return 0;
}
/*
* sort() function
* Same as stricmp() but numbers in strings
* are compared according to the value, not the ascii value of each digit
*/
int StrNumICmp( const wxChar* str1, const wxChar* str2 )
{
return StrLenNumICmp( str1, str2, 32735 );
}
/*
* sort() function
* Same as strnicmp() but numbers in strings
* are compared according to the value, not the ascii value of each digit
*/
int StrLenNumICmp( const wxChar* str1, const wxChar* str2, int NbMax )
{
int i;
int nb1 = 0, nb2 = 0;
if( ( str1 == NULL ) || ( str2 == NULL ) )
return 0;
for( i = 0; i < NbMax; i++ )
{
if( isdigit( *str1 ) && isdigit( *str2 ) ) /* find number */
{
nb1 = 0; nb2 = 0;
while( isdigit( *str1 ) )
{
nb1 = nb1 * 10 + *str1 - '0'; str1++;
}
while( isdigit( *str2 ) )
{
nb2 = nb2 * 10 + *str2 - '0'; str2++;
}
if( nb1 < nb2 )
return -1;
if( nb1 > nb2 )
return 1;
}
if( toupper( *str1 ) < toupper( *str2 ) )
return -1;
if( toupper( *str1 ) > toupper( *str2 ) )
return 1;
if( (*str1 == 0 ) && ( *str2 == 0 ) )
return 0;
str1++; str2++;
}
return 0;
}
/* compare a string to a pattern
* ( usual chars * and ? allowed).
* if case_sensitive == true, comparison is case sensitive
* return true if match else false
*/
bool WildCompareString( const wxString& pattern, const wxString& string_to_tst,
bool case_sensitive )
{
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();
}
else
{
_pattern = pattern;
_pattern.MakeUpper();
_string_to_tst = string_to_tst;
_string_to_tst.MakeUpper();
wild = _pattern.GetData();
string = _string_to_tst.GetData();
}
while( ( *string ) && ( *wild != '*' ) )
{
if( ( *wild != *string ) && ( *wild != '?' ) )
return FALSE;
wild++; string++;
}
while( *string )
{
if( *wild == '*' )
{
if( !*++wild )
return 1;
mp = wild;
cp = string + 1;
}
else if( ( *wild == *string ) || ( *wild == '?' ) )
{
wild++;
string++;
}
else
{
wild = mp;
string = cp++;
}
}
while( *wild == '*' )
{
wild++;
}
return !*wild;
}
/* Converts a string used to compensate for internalization of printf().
* Generated floats with a comma instead of point.
* Obsolete: use SetLocaleTo_C_standard instead
*/
char* to_point( char* Text )
{
char* line = Text;
if( Text == NULL )
return NULL;
for( ; *Text != 0; Text++ )
{
if( *Text == ',' )
*Text = '.';
}
return line;
}
/* Convert string to upper case.
* Returns pointer to the converted string.
*/
char* strupper( char* Text )
{
char* code = Text;
if( Text )
{
while( *code )
{
if( ( *code >= 'a' ) && ( *code <= 'z' ) )
*code += 'A' - 'a';
code++;
}
}
return Text;
}