|
|
/*
* This program source code file is part of KiCad, a free EDA CAD application. * * Copyright (C) 2016 CERN * Copyright (C) 2016-2018 KiCad Developers, see AUTHORS.txt for contributors. * * @author Wayne Stambaugh <stambaughw@gmail.com> * * 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, see <http://www.gnu.org/licenses/>.
*/
#include <ctype.h>
#include <algorithm>
#include <wx/mstream.h>
#include <wx/filename.h>
#include <wx/tokenzr.h>
#include <pgm_base.h>
#include <draw_graphic_text.h>
#include <kiway.h>
#include <kicad_string.h>
#include <richio.h>
#include <core/typeinfo.h>
#include <properties.h>
#include <trace_helpers.h>
#include <general.h>
#include <sch_bitmap.h>
#include <sch_bus_entry.h>
#include <sch_component.h>
#include <sch_junction.h>
#include <sch_line.h>
#include <sch_marker.h>
#include <sch_no_connect.h>
#include <sch_text.h>
#include <sch_sheet.h>
#include <sch_legacy_plugin.h>
#include <template_fieldnames.h>
#include <sch_screen.h>
#include <class_libentry.h>
#include <class_library.h>
#include <lib_arc.h>
#include <lib_bezier.h>
#include <lib_circle.h>
#include <lib_field.h>
#include <lib_pin.h>
#include <lib_polyline.h>
#include <lib_rectangle.h>
#include <lib_text.h>
#include <eeschema_id.h> // for MAX_UNIT_COUNT_PER_PACKAGE definition
#include <symbol_lib_table.h> // for PropPowerSymsOnly definintion.
#include <confirm.h>
// Must be the first line of part library document (.dcm) files.
#define DOCFILE_IDENT "EESchema-DOCLIB Version 2.0"
#define SCH_PARSE_ERROR( text, reader, pos ) \
THROW_PARSE_ERROR( text, reader.GetSource(), reader.Line(), \ reader.LineNumber(), pos - reader.Line() )
// Token delimiters.
const char* delims = " \t\r\n";
// Tokens to read/save graphic lines style
#define T_STYLE "style"
#define T_COLOR "rgb" // cannot be modifed (used by wxWidgets)
#define T_COLORA "rgba" // cannot be modifed (used by wxWidgets)
#define T_WIDTH "width"
static bool is_eol( char c ){ // The default file eol character used internally by KiCad.
// |
// | Possible eol if someone edited the file by hand on certain platforms.
// | |
// | | May have gone past eol with strtok().
// | | |
if( c == '\n' || c == '\r' || c == 0 ) return true;
return false;}
/**
* Compare \a aString to the string starting at \a aLine and advances the character point to * the end of \a String and returns the new pointer position in \a aOutput if it is not NULL. * * @param aString - A pointer to the string to compare. * @param aLine - A pointer to string to begin the comparison. * @param aOutput - A pointer to a string pointer to the end of the comparison if not NULL. * @return true if \a aString was found starting at \a aLine. Otherwise false. */static bool strCompare( const char* aString, const char* aLine, const char** aOutput = NULL ){ size_t len = strlen( aString ); bool retv = ( strncasecmp( aLine, aString, len ) == 0 ) && ( isspace( aLine[ len ] ) || aLine[ len ] == 0 );
if( retv && aOutput ) { const char* tmp = aLine;
// Move past the end of the token.
tmp += len;
// Move to the beginning of the next token.
while( *tmp && isspace( *tmp ) ) tmp++;
*aOutput = tmp; }
return retv;}
/**
* Parse an ASCII integer string with possible leading whitespace into * an integer and updates the pointer at \a aOutput if it is not NULL, just * like "man strtol()". * * @param aReader - The line reader used to generate exception throw information. * @param aLine - A pointer the current position in a string. * @param aOutput - The pointer to a string pointer to copy the string pointer position when * the parsing is complete. * @return A valid integer value. * @throw An #IO_ERROR on an unexpected end of line. * @throw A #PARSE_ERROR if the parsed token is not a valid integer. */static int parseInt( FILE_LINE_READER& aReader, const char* aLine, const char** aOutput = NULL ){ if( !*aLine ) SCH_PARSE_ERROR( _( "unexpected end of line" ), aReader, aLine );
// Clear errno before calling strtol() in case some other crt call set it.
errno = 0;
long retv = strtol( aLine, (char**) aOutput, 10 );
// Make sure no error occurred when calling strtol().
if( errno == ERANGE ) SCH_PARSE_ERROR( "invalid integer value", aReader, aLine );
// strtol does not strip off whitespace before the next token.
if( aOutput ) { const char* next = *aOutput;
while( *next && isspace( *next ) ) next++;
*aOutput = next; }
return (int) retv;}
/**
* Parse an ASCII hex integer string with possible leading whitespace into * a long integer and updates the pointer at \a aOutput if it is not NULL, just * like "man strtol". * * @param aReader - The line reader used to generate exception throw information. * @param aLine - A pointer the current position in a string. * @param aOutput - The pointer to a string pointer to copy the string pointer position when * the parsing is complete. * @return A valid integer value. * @throw IO_ERROR on an unexpected end of line. * @throw PARSE_ERROR if the parsed token is not a valid integer. */static unsigned long parseHex( FILE_LINE_READER& aReader, const char* aLine, const char** aOutput = NULL ){ if( !*aLine ) SCH_PARSE_ERROR( _( "unexpected end of line" ), aReader, aLine );
unsigned long retv;
// Clear errno before calling strtoul() in case some other crt call set it.
errno = 0; retv = strtoul( aLine, (char**) aOutput, 16 );
// Make sure no error occurred when calling strtoul().
if( errno == ERANGE ) SCH_PARSE_ERROR( "invalid hexadecimal number", aReader, aLine );
// Strip off whitespace before the next token.
if( aOutput ) { // const char* next = aLine + strlen( token );
const char* next = *aOutput;
while( *next && isspace( *next ) ) next++;
*aOutput = next; }
return retv;}
/**
* Parses an ASCII point string with possible leading whitespace into a double precision * floating point number and updates the pointer at \a aOutput if it is not NULL, just * like "man strtod". * * @param aReader - The line reader used to generate exception throw information. * @param aLine - A pointer the current position in a string. * @param aOutput - The pointer to a string pointer to copy the string pointer position when * the parsing is complete. * @return A valid double value. * @throw IO_ERROR on an unexpected end of line. * @throw PARSE_ERROR if the parsed token is not a valid integer. */static double parseDouble( FILE_LINE_READER& aReader, const char* aLine, const char** aOutput = NULL ){ if( !*aLine ) SCH_PARSE_ERROR( _( "unexpected end of line" ), aReader, aLine );
// Clear errno before calling strtod() in case some other crt call set it.
errno = 0;
double retv = strtod( aLine, (char**) aOutput );
// Make sure no error occurred when calling strtod().
if( errno == ERANGE ) SCH_PARSE_ERROR( "invalid floating point number", aReader, aLine );
// strtod does not strip off whitespace before the next token.
if( aOutput ) { const char* next = *aOutput;
while( *next && isspace( *next ) ) next++;
*aOutput = next; }
return retv;}
/**
* Parse a single ASCII character and updates the pointer at \a aOutput if it is not NULL. * * @param aReader - The line reader used to generate exception throw information. * @param aCurrentToken - A pointer the current position in a string. * @param aNextToken - The pointer to a string pointer to copy the string pointer position when * the parsing is complete. * @return A valid ASCII character. * @throw IO_ERROR on an unexpected end of line. * @throw PARSE_ERROR if the parsed token is not a a single character token. */static char parseChar( FILE_LINE_READER& aReader, const char* aCurrentToken, const char** aNextToken = NULL ){ while( *aCurrentToken && isspace( *aCurrentToken ) ) aCurrentToken++;
if( !*aCurrentToken ) SCH_PARSE_ERROR( _( "unexpected end of line" ), aReader, aCurrentToken );
if( !isspace( *( aCurrentToken + 1 ) ) ) SCH_PARSE_ERROR( "expected single character token", aReader, aCurrentToken );
if( aNextToken ) { const char* next = aCurrentToken + 2;
while( *next && isspace( *next ) ) next++;
*aNextToken = next; }
return *aCurrentToken;}
/**
* Parse an unquoted utf8 string and updates the pointer at \a aOutput if it is not NULL. * * The parsed string must be a continuous string with no white space. * * @param aString - A reference to the parsed string. * @param aReader - The line reader used to generate exception throw information. * @param aCurrentToken - A pointer the current position in a string. * @param aNextToken - The pointer to a string pointer to copy the string pointer position when * the parsing is complete. * @param aCanBeEmpty - True if the parsed string is optional. False if it is mandatory. * @throw IO_ERROR on an unexpected end of line. * @throw PARSE_ERROR if the \a aCanBeEmpty is false and no string was parsed. */static void parseUnquotedString( wxString& aString, FILE_LINE_READER& aReader, const char* aCurrentToken, const char** aNextToken = NULL, bool aCanBeEmpty = false ){ if( !*aCurrentToken ) { if( aCanBeEmpty ) return; else SCH_PARSE_ERROR( _( "unexpected end of line" ), aReader, aCurrentToken ); }
const char* tmp = aCurrentToken;
while( *tmp && isspace( *tmp ) ) tmp++;
if( !*tmp ) { if( aCanBeEmpty ) return; else SCH_PARSE_ERROR( _( "unexpected end of line" ), aReader, aCurrentToken ); }
std::string utf8;
while( *tmp && !isspace( *tmp ) ) utf8 += *tmp++;
aString = FROM_UTF8( utf8.c_str() );
if( aString.IsEmpty() && !aCanBeEmpty ) SCH_PARSE_ERROR( _( "expected unquoted string" ), aReader, aCurrentToken );
if( aNextToken ) { const char* next = tmp;
while( *next && isspace( *next ) ) next++;
*aNextToken = next; }}
/**
* Parse an quoted ASCII utf8 and updates the pointer at \a aOutput if it is not NULL. * * The parsed string must be contained within a single line. There are no multi-line * quoted strings in the legacy schematic file format. * * @param aString - A reference to the parsed string. * @param aReader - The line reader used to generate exception throw information. * @param aCurrentToken - A pointer the current position in a string. * @param aNextToken - The pointer to a string pointer to copy the string pointer position when * the parsing is complete. * @param aCanBeEmpty - True if the parsed string is optional. False if it is mandatory. * @throw IO_ERROR on an unexpected end of line. * @throw PARSE_ERROR if the \a aCanBeEmpty is false and no string was parsed. */static void parseQuotedString( wxString& aString, FILE_LINE_READER& aReader, const char* aCurrentToken, const char** aNextToken = NULL, bool aCanBeEmpty = false ){ if( !*aCurrentToken ) { if( aCanBeEmpty ) return; else SCH_PARSE_ERROR( _( "unexpected end of line" ), aReader, aCurrentToken ); }
const char* tmp = aCurrentToken;
while( *tmp && isspace( *tmp ) ) tmp++;
if( !*tmp ) { if( aCanBeEmpty ) return; else SCH_PARSE_ERROR( _( "unexpected end of line" ), aReader, aCurrentToken ); }
// Verify opening quote.
if( *tmp != '"' ) SCH_PARSE_ERROR( "expecting opening quote", aReader, aCurrentToken );
tmp++;
std::string utf8; // utf8 without escapes and quotes.
// Fetch everything up to closing quote.
while( *tmp ) { if( *tmp == '\\' ) { tmp++;
if( !*tmp ) SCH_PARSE_ERROR( _( "unexpected end of line" ), aReader, aCurrentToken );
// Do not copy the escape byte if it is followed by \ or "
if( *tmp != '"' && *tmp != '\\' ) utf8 += '\\';
utf8 += *tmp; } else if( *tmp == '"' ) // Closing double quote.
{ break; } else { utf8 += *tmp; }
tmp++; }
aString = FROM_UTF8( utf8.c_str() );
if( aString.IsEmpty() && !aCanBeEmpty ) SCH_PARSE_ERROR( "expected quoted string", aReader, aCurrentToken );
if( *tmp && *tmp != '"' ) SCH_PARSE_ERROR( "no closing quote for string found", aReader, tmp );
// Move past the closing quote.
tmp++;
if( aNextToken ) { const char* next = tmp;
while( *next && *next == ' ' ) next++;
*aNextToken = next; }}
/**
* A cache assistant for the part library portion of the #SCH_PLUGIN API, and only for the * #SCH_LEGACY_PLUGIN, so therefore is private to this implementation file, i.e. not placed * into a header. */class SCH_LEGACY_PLUGIN_CACHE{ static int m_modHash; // Keep track of the modification status of the library.
wxString m_fileName; // Absolute path and file name.
wxFileName m_libFileName; // Absolute path and file name is required here.
wxDateTime m_fileModTime; LIB_ALIAS_MAP m_aliases; // Map of names of LIB_ALIAS pointers.
bool m_isWritable; bool m_isModified; int m_versionMajor; int m_versionMinor; int m_libType; // Is this cache a component or symbol library.
LIB_PART* loadPart( FILE_LINE_READER& aReader ); void loadHeader( FILE_LINE_READER& aReader ); void loadAliases( std::unique_ptr< LIB_PART >& aPart, FILE_LINE_READER& aReader ); void loadField( std::unique_ptr< LIB_PART >& aPart, FILE_LINE_READER& aReader ); void loadDrawEntries( std::unique_ptr< LIB_PART >& aPart, FILE_LINE_READER& aReader ); void loadFootprintFilters( std::unique_ptr< LIB_PART >& aPart, FILE_LINE_READER& aReader ); void loadDocs(); LIB_ARC* loadArc( std::unique_ptr< LIB_PART >& aPart, FILE_LINE_READER& aReader ); LIB_CIRCLE* loadCircle( std::unique_ptr< LIB_PART >& aPart, FILE_LINE_READER& aReader ); LIB_TEXT* loadText( std::unique_ptr< LIB_PART >& aPart, FILE_LINE_READER& aReader ); LIB_RECTANGLE* loadRectangle( std::unique_ptr< LIB_PART >& aPart, FILE_LINE_READER& aReader ); LIB_PIN* loadPin( std::unique_ptr< LIB_PART >& aPart, FILE_LINE_READER& aReader ); LIB_POLYLINE* loadPolyLine( std::unique_ptr< LIB_PART >& aPart, FILE_LINE_READER& aReader ); LIB_BEZIER* loadBezier( std::unique_ptr< LIB_PART >& aPart, FILE_LINE_READER& aReader );
FILL_T parseFillMode( FILE_LINE_READER& aReader, const char* aLine, const char** aOutput ); bool checkForDuplicates( wxString& aAliasName ); LIB_ALIAS* removeAlias( LIB_ALIAS* aAlias );
void saveDocFile(); void saveSymbol( LIB_PART* aSymbol, std::unique_ptr< FILE_OUTPUTFORMATTER >& aFormatter ); void saveArc( LIB_ARC* aArc, std::unique_ptr< FILE_OUTPUTFORMATTER >& aFormatter ); void saveBezier( LIB_BEZIER* aBezier, std::unique_ptr< FILE_OUTPUTFORMATTER >& aFormatter ); void saveCircle( LIB_CIRCLE* aCircle, std::unique_ptr< FILE_OUTPUTFORMATTER >& aFormatter ); void saveField( LIB_FIELD* aField, std::unique_ptr< FILE_OUTPUTFORMATTER >& aFormatter ); void savePin( LIB_PIN* aPin, std::unique_ptr< FILE_OUTPUTFORMATTER >& aFormatter ); void savePolyLine( LIB_POLYLINE* aPolyLine, std::unique_ptr< FILE_OUTPUTFORMATTER >& aFormatter ); void saveRectangle( LIB_RECTANGLE* aRectangle, std::unique_ptr< FILE_OUTPUTFORMATTER >& aFormatter ); void saveText( LIB_TEXT* aText, std::unique_ptr< FILE_OUTPUTFORMATTER >& aFormatter );
friend SCH_LEGACY_PLUGIN;
public: SCH_LEGACY_PLUGIN_CACHE( const wxString& aLibraryPath ); ~SCH_LEGACY_PLUGIN_CACHE();
int GetModifyHash() const { return m_modHash; }
// Most all functions in this class throw IO_ERROR exceptions. There are no
// error codes nor user interface calls from here, nor in any SCH_PLUGIN objects.
// Catch these exceptions higher up please.
/// Save the entire library to file m_libFileName;
void Save( bool aSaveDocFile = true );
void Load();
void AddSymbol( const LIB_PART* aPart );
void DeleteAlias( const wxString& aAliasName );
void DeleteSymbol( const wxString& aAliasName );
// If m_libFileName is a symlink follow it to the real source file
wxFileName GetRealFile() const;
wxDateTime GetLibModificationTime();
bool IsFile( const wxString& aFullPathAndFileName ) const;
bool IsFileChanged() const;
void SetModified( bool aModified = true ) { m_isModified = aModified; }
wxString GetLogicalName() const { return m_libFileName.GetName(); }
void SetFileName( const wxString& aFileName ) { m_libFileName = aFileName; }
wxString GetFileName() const { return m_libFileName.GetFullPath(); }};
SCH_LEGACY_PLUGIN::SCH_LEGACY_PLUGIN(){ init( NULL );}
SCH_LEGACY_PLUGIN::~SCH_LEGACY_PLUGIN(){ delete m_cache;}
void SCH_LEGACY_PLUGIN::init( KIWAY* aKiway, const PROPERTIES* aProperties ){ m_version = 0; m_rootSheet = NULL; m_props = aProperties; m_kiway = aKiway; m_cache = NULL; m_out = NULL;}
SCH_SHEET* SCH_LEGACY_PLUGIN::Load( const wxString& aFileName, KIWAY* aKiway, SCH_SHEET* aAppendToMe, const PROPERTIES* aProperties ){ wxASSERT( !aFileName || aKiway != NULL );
LOCALE_IO toggle; // toggles on, then off, the C locale.
SCH_SHEET* sheet;
wxFileName fn = aFileName;
// Unfortunately child sheet file names the legacy schematic file format are not fully
// qualified and are always appended to the project path. The aFileName attribute must
// always be an absolute path so the project path can be used for load child sheet files.
wxASSERT( fn.IsAbsolute() );
if( aAppendToMe ) { wxLogTrace( traceSchLegacyPlugin, "Append \"%s\" to sheet \"%s\".", aFileName, aAppendToMe->GetFileName() );
wxFileName normedFn = aAppendToMe->GetFileName();
if( !normedFn.IsAbsolute() ) { if( aFileName.Right( normedFn.GetFullPath().Length() ) == normedFn.GetFullPath() ) m_path = aFileName.Left( aFileName.Length() - normedFn.GetFullPath().Length() ); }
if( m_path.IsEmpty() ) m_path = aKiway->Prj().GetProjectPath();
wxLogTrace( traceSchLegacyPlugin, "Normalized append path \"%s\".", m_path ); } else { m_path = aKiway->Prj().GetProjectPath(); }
m_currentPath.push( m_path ); init( aKiway, aProperties );
if( aAppendToMe == NULL ) { // Clean up any allocated memory if an exception occurs loading the schematic.
std::unique_ptr< SCH_SHEET > newSheet( new SCH_SHEET ); newSheet->SetFileName( aFileName ); m_rootSheet = newSheet.get(); loadHierarchy( newSheet.get() );
// If we got here, the schematic loaded successfully.
sheet = newSheet.release(); } else { m_rootSheet = aAppendToMe->GetRootSheet(); wxASSERT( m_rootSheet != NULL ); sheet = aAppendToMe; loadHierarchy( sheet ); }
wxASSERT( m_currentPath.size() == 1 ); // only the project path should remain
return sheet;}
// Everything below this comment is recursive. Modify with care.
void SCH_LEGACY_PLUGIN::loadHierarchy( SCH_SHEET* aSheet ){ SCH_SCREEN* screen = NULL;
if( !aSheet->GetScreen() ) { // SCH_SCREEN objects store the full path and file name where the SCH_SHEET object only
// stores the file name and extension. Add the project path to the file name and
// extension to compare when calling SCH_SHEET::SearchHierarchy().
wxFileName fileName = aSheet->GetFileName();
if( !fileName.IsAbsolute() ) fileName.MakeAbsolute( m_currentPath.top() );
// Save the current path so that it gets restored when decending and ascending the
// sheet hierarchy which allows for sheet schematic files to be nested in folders
// relative to the last path a schematic was loaded from.
wxLogTrace( traceSchLegacyPlugin, "Saving path \"%s\"", m_currentPath.top() ); m_currentPath.push( fileName.GetPath() ); wxLogTrace( traceSchLegacyPlugin, "Current path \"%s\"", m_currentPath.top() ); wxLogTrace( traceSchLegacyPlugin, "Loading \"%s\"", fileName.GetFullPath() );
m_rootSheet->SearchHierarchy( fileName.GetFullPath(), &screen );
if( screen ) { aSheet->SetScreen( screen );
// Do not need to load the sub-sheets - this has already been done.
} else { aSheet->SetScreen( new SCH_SCREEN( m_kiway ) ); aSheet->GetScreen()->SetFileName( fileName.GetFullPath() );
try { loadFile( fileName.GetFullPath(), aSheet->GetScreen() );
EDA_ITEM* item = aSheet->GetScreen()->GetDrawItems();
while( item ) { if( item->Type() == SCH_SHEET_T ) { SCH_SHEET* sheet = (SCH_SHEET*) item;
// Set the parent to aSheet. This effectively creates a method to find
// the root sheet from any sheet so a pointer to the root sheet does not
// need to be stored globally. Note: this is not the same as a hierarchy.
// Complex hierarchies can have multiple copies of a sheet. This only
// provides a simple tree to find the root sheet.
sheet->SetParent( aSheet );
// Recursion starts here.
loadHierarchy( sheet ); }
item = item->Next(); } } catch( const IO_ERROR& ioe ) { // If there is a problem loading the root sheet, there is no recovery.
if( aSheet == m_rootSheet ) throw( ioe );
// For all subsheets, queue up the error message for the caller.
if( !m_error.IsEmpty() ) m_error += "\n";
m_error += ioe.What(); } }
m_currentPath.pop(); wxLogTrace( traceSchLegacyPlugin, "Restoring path \"%s\"", m_currentPath.top() ); }}
void SCH_LEGACY_PLUGIN::loadFile( const wxString& aFileName, SCH_SCREEN* aScreen ){ FILE_LINE_READER reader( aFileName );
loadHeader( reader, aScreen );
while( reader.ReadLine() ) { char* line = reader.Line();
while( *line && *line == ' ' ) line++;
// Either an object will be loaded properly or the file load will fail and raise
// an exception.
if( strCompare( "$Descr", line ) ) loadPageSettings( reader, aScreen ); else if( strCompare( "$Comp", line ) ) aScreen->Append( loadComponent( reader ) ); else if( strCompare( "$Sheet", line ) ) aScreen->Append( loadSheet( reader ) ); else if( strCompare( "$Bitmap", line ) ) aScreen->Append( loadBitmap( reader ) ); else if( strCompare( "Connection", line ) ) aScreen->Append( loadJunction( reader ) ); else if( strCompare( "NoConn", line ) ) aScreen->Append( loadNoConnect( reader ) ); else if( strCompare( "Wire", line ) ) aScreen->Append( loadWire( reader ) ); else if( strCompare( "Entry", line ) ) aScreen->Append( loadBusEntry( reader ) ); else if( strCompare( "Text", line ) ) aScreen->Append( loadText( reader ) ); else if( strCompare( "$EndSCHEMATC", line ) ) return; }
// Unfortunately schematic files prior to version 2 are not terminated with $EndSCHEMATC
// so checking for it's existance will fail so just exit here and take our chances. :(
if( m_version > 1 ) THROW_IO_ERROR( "'$EndSCHEMATC' not found" );}
void SCH_LEGACY_PLUGIN::loadHeader( FILE_LINE_READER& aReader, SCH_SCREEN* aScreen ){ const char* line = aReader.ReadLine();
if( !line || !strCompare( "Eeschema Schematic File Version", line, &line ) ) { m_error.Printf( _( "\"%s\" does not appear to be an Eeschema file" ), GetChars( aScreen->GetFileName() ) ); THROW_IO_ERROR( m_error ); }
// get the file version here.
m_version = parseInt( aReader, line, &line );
// The next lines are the lib list section, and are mainly comments, like:
// LIBS:power
// the lib list is not used, but is in schematic file just in case.
// It is usually not empty, but we accept empty list.
// If empty, there is a legacy section, not used
// EELAYER i j
// and the last line is
// EELAYER END
// Skip all lines until the end of header "EELAYER END" is found
while( aReader.ReadLine() ) { line = aReader.Line();
while( *line == ' ' ) line++;
if( strCompare( "EELAYER END", line ) ) return; }
THROW_IO_ERROR( _( "Missing 'EELAYER END'" ) );}
void SCH_LEGACY_PLUGIN::loadPageSettings( FILE_LINE_READER& aReader, SCH_SCREEN* aScreen ){ wxASSERT( aScreen != NULL );
wxString buf; const char* line = aReader.Line();
PAGE_INFO pageInfo; TITLE_BLOCK tb;
wxCHECK_RET( strCompare( "$Descr", line, &line ), "Invalid sheet description" );
parseUnquotedString( buf, aReader, line, &line );
if( !pageInfo.SetType( buf ) ) SCH_PARSE_ERROR( "invalid page size", aReader, line );
int pagew = parseInt( aReader, line, &line ); int pageh = parseInt( aReader, line, &line );
if( buf == PAGE_INFO::Custom ) { pageInfo.SetWidthMils( pagew ); pageInfo.SetHeightMils( pageh ); } else { wxString orientation;
// Non custom size, set portrait if its present. Can be empty string which defaults
// to landscape.
parseUnquotedString( orientation, aReader, line, &line, true );
if( orientation == "portrait" ) pageInfo.SetPortrait( true ); }
aScreen->SetPageSettings( pageInfo );
while( line != NULL ) { buf.clear();
if( !aReader.ReadLine() ) SCH_PARSE_ERROR( _( "unexpected end of file" ), aReader, line );
line = aReader.Line();
if( strCompare( "Sheet", line, &line ) ) { aScreen->m_ScreenNumber = parseInt( aReader, line, &line ); aScreen->m_NumberOfScreens = parseInt( aReader, line, &line ); } else if( strCompare( "Title", line, &line ) ) { parseQuotedString( buf, aReader, line, &line, true ); tb.SetTitle( buf ); } else if( strCompare( "Date", line, &line ) ) { parseQuotedString( buf, aReader, line, &line, true ); tb.SetDate( buf ); } else if( strCompare( "Rev", line, &line ) ) { parseQuotedString( buf, aReader, line, &line, true ); tb.SetRevision( buf ); } else if( strCompare( "Comp", line, &line ) ) { parseQuotedString( buf, aReader, line, &line, true ); tb.SetCompany( buf ); } else if( strCompare( "Comment1", line, &line ) ) { parseQuotedString( buf, aReader, line, &line, true ); tb.SetComment1( buf ); } else if( strCompare( "Comment2", line, &line ) ) { parseQuotedString( buf, aReader, line, &line, true ); tb.SetComment2( buf ); } else if( strCompare( "Comment3", line, &line ) ) { parseQuotedString( buf, aReader, line, &line, true ); tb.SetComment3( buf ); } else if( strCompare( "Comment4", line, &line ) ) { parseQuotedString( buf, aReader, line, &line, true ); tb.SetComment4( buf ); } else if( strCompare( "$EndDescr", line ) ) { aScreen->SetTitleBlock( tb ); return; } }
SCH_PARSE_ERROR( "missing 'EndDescr'", aReader, line );}
SCH_SHEET* SCH_LEGACY_PLUGIN::loadSheet( FILE_LINE_READER& aReader ){ std::unique_ptr< SCH_SHEET > sheet( new SCH_SHEET() );
sheet->SetTimeStamp( GetNewTimeStamp() );
const char* line = aReader.ReadLine();
while( line != NULL ) { if( strCompare( "S", line, &line ) ) // Sheet dimensions.
{ wxPoint position;
position.x = parseInt( aReader, line, &line ); position.y = parseInt( aReader, line, &line ); sheet->SetPosition( position );
wxSize size;
size.SetWidth( parseInt( aReader, line, &line ) ); size.SetHeight( parseInt( aReader, line, &line ) ); sheet->SetSize( size ); } else if( strCompare( "U", line, &line ) ) // Sheet time stamp.
{ sheet->SetTimeStamp( parseHex( aReader, line ) ); } else if( *line == 'F' ) // Sheet field.
{ line++;
wxString text; int size; int fieldId = parseInt( aReader, line, &line );
if( fieldId == 0 || fieldId == 1 ) // Sheet name and file name.
{ parseQuotedString( text, aReader, line, &line ); size = parseInt( aReader, line, &line );
if( fieldId == 0 ) { sheet->SetName( text ); sheet->SetSheetNameSize( size ); } else { sheet->SetFileName( text ); sheet->SetFileNameSize( size ); } } else // Sheet pin.
{ std::unique_ptr< SCH_SHEET_PIN > sheetPin( new SCH_SHEET_PIN( sheet.get() ) );
sheetPin->SetNumber( fieldId );
// Can be empty fields.
parseQuotedString( text, aReader, line, &line, true );
sheetPin->SetText( text );
if( line == NULL ) THROW_IO_ERROR( _( "unexpected end of line" ) );
switch( parseChar( aReader, line, &line ) ) { case 'I': sheetPin->SetShape( NET_INPUT ); break;
case 'O': sheetPin->SetShape( NET_OUTPUT ); break;
case 'B': sheetPin->SetShape( NET_BIDI ); break;
case 'T': sheetPin->SetShape( NET_TRISTATE ); break;
case 'U': sheetPin->SetShape( NET_UNSPECIFIED ); break; default: SCH_PARSE_ERROR( "invalid sheet pin type", aReader, line ); }
switch( parseChar( aReader, line, &line ) ) { case 'R': /* pin on right side */ sheetPin->SetEdge( SCH_SHEET_PIN::SHEET_RIGHT_SIDE ); break;
case 'T': /* pin on top side */ sheetPin->SetEdge( SCH_SHEET_PIN::SHEET_TOP_SIDE ); break;
case 'B': /* pin on bottom side */ sheetPin->SetEdge( SCH_SHEET_PIN::SHEET_BOTTOM_SIDE ); break;
case 'L': /* pin on left side */ sheetPin->SetEdge( SCH_SHEET_PIN::SHEET_LEFT_SIDE ); break; default: SCH_PARSE_ERROR( "invalid sheet pin side", aReader, line ); }
wxPoint position;
position.x = parseInt( aReader, line, &line ); position.y = parseInt( aReader, line, &line ); sheetPin->SetPosition( position );
size = parseInt( aReader, line, &line );
sheetPin->SetTextSize( wxSize( size, size ) );
sheet->AddPin( sheetPin.release() ); } } else if( strCompare( "$EndSheet", line ) ) return sheet.release();
line = aReader.ReadLine(); }
SCH_PARSE_ERROR( "missing '$EndSheet`", aReader, line );
return NULL; // Prevents compiler warning. Should never get here.
}
SCH_BITMAP* SCH_LEGACY_PLUGIN::loadBitmap( FILE_LINE_READER& aReader ){ std::unique_ptr< SCH_BITMAP > bitmap( new SCH_BITMAP );
const char* line = aReader.Line();
wxCHECK( strCompare( "$Bitmap", line, &line ), NULL );
line = aReader.ReadLine();
while( line != NULL ) { if( strCompare( "Pos", line, &line ) ) { wxPoint position;
position.x = parseInt( aReader, line, &line ); position.y = parseInt( aReader, line, &line ); bitmap->SetPosition( position ); } else if( strCompare( "Scale", line, &line ) ) { /// @todo Make m_scale private and add accessors.
bitmap->GetImage()->SetScale( parseDouble( aReader, line, &line ) ); } else if( strCompare( "Data", line, &line ) ) { wxMemoryOutputStream stream;
while( line ) { if( !aReader.ReadLine() ) SCH_PARSE_ERROR( _( "Unexpected end of file" ), aReader, line );
line = aReader.Line();
if( strCompare( "EndData", line ) ) { // all the PNG date is read.
// We expect here m_image and m_bitmap are void
wxImage* image = new wxImage(); wxMemoryInputStream istream( stream ); image->LoadFile( istream, wxBITMAP_TYPE_PNG ); bitmap->GetImage()->SetImage( image ); bitmap->GetImage()->SetBitmap( new wxBitmap( *image ) ); break; }
// Read PNG data, stored in hexadecimal,
// each byte = 2 hexadecimal digits and a space between 2 bytes
// and put it in memory stream buffer
int len = strlen( line );
for( ; len > 0 && !isspace( *line ); len -= 3, line += 3 ) { int value = 0;
if( sscanf( line, "%X", &value ) == 1 ) stream.PutC( (char) value ); else THROW_IO_ERROR( "invalid PNG data" ); } }
if( line == NULL ) THROW_IO_ERROR( _( "unexpected end of file" ) ); } else if( strCompare( "$EndBitmap", line ) ) return bitmap.release();
line = aReader.ReadLine(); }
THROW_IO_ERROR( _( "unexpected end of file" ) );}
SCH_JUNCTION* SCH_LEGACY_PLUGIN::loadJunction( FILE_LINE_READER& aReader ){ std::unique_ptr< SCH_JUNCTION > junction( new SCH_JUNCTION );
const char* line = aReader.Line();
wxCHECK( strCompare( "Connection", line, &line ), NULL );
wxString name;
parseUnquotedString( name, aReader, line, &line );
wxPoint position;
position.x = parseInt( aReader, line, &line ); position.y = parseInt( aReader, line, &line ); junction->SetPosition( position );
return junction.release();}
SCH_NO_CONNECT* SCH_LEGACY_PLUGIN::loadNoConnect( FILE_LINE_READER& aReader ){ std::unique_ptr< SCH_NO_CONNECT > no_connect( new SCH_NO_CONNECT );
const char* line = aReader.Line();
wxCHECK( strCompare( "NoConn", line, &line ), NULL );
wxString name;
parseUnquotedString( name, aReader, line, &line );
wxPoint position;
position.x = parseInt( aReader, line, &line ); position.y = parseInt( aReader, line, &line ); no_connect->SetPosition( position );
return no_connect.release();}
SCH_LINE* SCH_LEGACY_PLUGIN::loadWire( FILE_LINE_READER& aReader ){ std::unique_ptr< SCH_LINE > wire( new SCH_LINE );
const char* line = aReader.Line();
wxCHECK( strCompare( "Wire", line, &line ), NULL );
if( strCompare( "Wire", line, &line ) ) wire->SetLayer( LAYER_WIRE ); else if( strCompare( "Bus", line, &line ) ) wire->SetLayer( LAYER_BUS ); else if( strCompare( "Notes", line, &line ) ) wire->SetLayer( LAYER_NOTES ); else SCH_PARSE_ERROR( "invalid line type", aReader, line );
if( !strCompare( "Line", line, &line ) ) SCH_PARSE_ERROR( "invalid wire definition", aReader, line );
// Since Sept 15, 2017, a line style is alloved (width, style, color)
// Only non default values are stored
while( !is_eol( *line ) ) { wxString buf;
parseUnquotedString( buf, aReader, line, &line );
if( buf == ")" ) continue;
else if( buf == T_WIDTH ) { int size = parseInt( aReader, line, &line ); wire->SetLineWidth( size ); } else if( buf == T_STYLE ) { parseUnquotedString( buf, aReader, line, &line ); int style = SCH_LINE::GetLineStyleInternalId( buf ); wire->SetLineStyle( style ); } else // should be the color parameter.
{ // The color param is something like rgb(150, 40, 191)
// and because there is no space between ( and 150
// the first param is inside buf.
// So break keyword and the first param into 2 separate strings.
wxString prm, keyword; keyword = buf.BeforeLast( '(', &prm );
if( ( keyword == T_COLOR ) || ( keyword == T_COLORA ) ) { long color[4] = { 0 };
int ii = 0;
if( !prm.IsEmpty() ) { prm.ToLong( &color[ii] ); ii++; }
int prm_count = ( keyword == T_COLORA ) ? 4 : 3; // fix opacity to 1.0 or 255, when not exists in file
color[3] = 255;
for(; ii < prm_count && !is_eol( *line ); ii++ ) color[ii] = parseInt( aReader, line, &line );
wire->SetLineColor( color[0]/255.0, color[1]/255.0, color[2]/255.0,color[3]/255.0 ); } } }
// Read the segment en points coordinates:
line = aReader.ReadLine();
wxPoint begin, end;
begin.x = parseInt( aReader, line, &line ); begin.y = parseInt( aReader, line, &line ); end.x = parseInt( aReader, line, &line ); end.y = parseInt( aReader, line, &line );
wire->SetStartPoint( begin ); wire->SetEndPoint( end );
return wire.release();}
SCH_BUS_ENTRY_BASE* SCH_LEGACY_PLUGIN::loadBusEntry( FILE_LINE_READER& aReader ){ const char* line = aReader.Line();
wxCHECK( strCompare( "Entry", line, &line ), NULL );
std::unique_ptr< SCH_BUS_ENTRY_BASE > busEntry;
if( strCompare( "Wire", line, &line ) ) { busEntry.reset( new SCH_BUS_WIRE_ENTRY );
if( !strCompare( "Line", line, &line ) ) SCH_PARSE_ERROR( "invalid bus entry definition expected 'Line'", aReader, line ); } else if( strCompare( "Bus", line, &line ) ) { busEntry.reset( new SCH_BUS_BUS_ENTRY );
if( !strCompare( "Bus", line, &line ) ) SCH_PARSE_ERROR( "invalid bus entry definition expected 'Bus'", aReader, line ); } else SCH_PARSE_ERROR( "invalid bus entry type", aReader, line );
line = aReader.ReadLine();
wxPoint pos; wxSize size;
pos.x = parseInt( aReader, line, &line ); pos.y = parseInt( aReader, line, &line ); size.x = parseInt( aReader, line, &line ); size.y = parseInt( aReader, line, &line );
size.x -= pos.x; size.y -= pos.y;
busEntry->SetPosition( pos ); busEntry->SetSize( size );
return busEntry.release();}
SCH_TEXT* SCH_LEGACY_PLUGIN::loadText( FILE_LINE_READER& aReader ){ const char* line = aReader.Line();
wxCHECK( strCompare( "Text", line, &line ), NULL );
std::unique_ptr< SCH_TEXT> text;
if( strCompare( "Notes", line, &line ) ) text.reset( new SCH_TEXT ); else if( strCompare( "Label", line, &line ) ) text.reset( new SCH_LABEL ); else if( strCompare( "HLabel", line, &line ) ) text.reset( new SCH_HIERLABEL ); else if( strCompare( "GLabel", line, &line ) ) { // Prior to version 2, the SCH_GLOBALLABEL object did not exist.
if( m_version == 1 ) text.reset( new SCH_HIERLABEL ); else text.reset( new SCH_GLOBALLABEL ); } else SCH_PARSE_ERROR( "unknown Text type", aReader, line );
// Parse the parameters common to all text objects.
wxPoint position;
position.x = parseInt( aReader, line, &line ); position.y = parseInt( aReader, line, &line ); text->SetPosition( position ); text->SetLabelSpinStyle( parseInt( aReader, line, &line ) );
int size = parseInt( aReader, line, &line );
text->SetTextSize( wxSize( size, size ) );
// Parse the global and hierarchical label type.
if( text->Type() == SCH_HIERARCHICAL_LABEL_T || text->Type() == SCH_GLOBAL_LABEL_T ) { if( strCompare( SheetLabelType[NET_INPUT], line, &line ) ) text->SetShape( NET_INPUT ); else if( strCompare( SheetLabelType[NET_OUTPUT], line, &line ) ) text->SetShape( NET_OUTPUT ); else if( strCompare( SheetLabelType[NET_BIDI], line, &line ) ) text->SetShape( NET_BIDI ); else if( strCompare( SheetLabelType[NET_TRISTATE], line, &line ) ) text->SetShape( NET_TRISTATE ); else if( strCompare( SheetLabelType[NET_UNSPECIFIED], line, &line ) ) text->SetShape( NET_UNSPECIFIED ); else SCH_PARSE_ERROR( "invalid label type", aReader, line ); }
int thickness = 0;
// The following tokens do not exist in version 1 schematic files,
// and not always in version 2 for HLabels and GLabels
if( m_version > 1 ) { if( m_version > 2 || *line >= ' ' ) { if( strCompare( "Italic", line, &line ) ) text->SetItalic( true ); else if( !strCompare( "~", line, &line ) ) SCH_PARSE_ERROR( _( "expected 'Italics' or '~'" ), aReader, line ); }
// The thickness token does not exist in older versions of the schematic file format
// so calling parseInt will be made only if the EOL is not reached.
if( *line >= ' ' ) thickness = parseInt( aReader, line, &line ); }
text->SetBold( thickness != 0 ); text->SetThickness( thickness != 0 ? GetPenSizeForBold( size ) : 0 );
// Read the text string for the text.
char* tmp = aReader.ReadLine();
tmp = strtok( tmp, "\r\n" ); wxString val = FROM_UTF8( tmp );
for( ; ; ) { int i = val.find( wxT( "\\n" ) );
if( i == wxNOT_FOUND ) break;
val.erase( i, 2 ); val.insert( i, wxT( "\n" ) ); }
text->SetText( val );
return text.release();}
SCH_COMPONENT* SCH_LEGACY_PLUGIN::loadComponent( FILE_LINE_READER& aReader ){ const char* line = aReader.Line();
wxCHECK( strCompare( "$Comp", line, &line ), NULL );
std::unique_ptr< SCH_COMPONENT > component( new SCH_COMPONENT() );
line = aReader.ReadLine();
while( line != NULL ) { if( strCompare( "L", line, &line ) ) { wxString libName;
parseUnquotedString( libName, aReader, line, &line ); libName.Replace( "~", " " );
LIB_ID libId;
// Prior to schematic version 4, library IDs did not have a library nickname so
// parsing the symbol name with LIB_ID::Parse() would break symbol library links
// that contained '/' and ':' characters.
if( m_version > 3 ) libId.Parse( libName, LIB_ID::ID_SCH, true ); else libId.SetLibItemName( libName, false );
component->SetLibId( libId );
wxString refDesignator;
parseUnquotedString( refDesignator, aReader, line, &line ); refDesignator.Replace( "~", " " );
wxString prefix = refDesignator;
while( prefix.Length() ) { if( ( prefix.Last() < '0' || prefix.Last() > '9') && prefix.Last() != '?' ) break;
prefix.RemoveLast(); }
// Avoid a prefix containing trailing/leading spaces
prefix.Trim( true ); prefix.Trim( false );
if( prefix.IsEmpty() ) component->SetPrefix( wxString( "U" ) ); else component->SetPrefix( prefix ); } else if( strCompare( "U", line, &line ) ) { // This fixes a potentially buggy files caused by unit being set to zero which
// causes netlist issues. See https://bugs.launchpad.net/kicad/+bug/1677282.
int unit = parseInt( aReader, line, &line );
if( unit == 0 ) { unit = 1;
// Set the file as modified so the user can be warned.
if( m_rootSheet && m_rootSheet->GetScreen() ) m_rootSheet->GetScreen()->SetModify(); }
component->SetUnit( unit ); component->SetConvert( parseInt( aReader, line, &line ) ); component->SetTimeStamp( parseHex( aReader, line, &line ) ); } else if( strCompare( "P", line, &line ) ) { wxPoint pos;
pos.x = parseInt( aReader, line, &line ); pos.y = parseInt( aReader, line, &line ); component->SetPosition( pos ); } else if( strCompare( "AR", line, &line ) ) { const char* strCompare = "Path="; int len = strlen( strCompare );
if( strncasecmp( strCompare, line, len ) != 0 ) SCH_PARSE_ERROR( "missing 'Path=' token", aReader, line );
line += len; wxString path, reference, unit;
parseQuotedString( path, aReader, line, &line );
strCompare = "Ref="; len = strlen( strCompare );
if( strncasecmp( strCompare, line, len ) != 0 ) SCH_PARSE_ERROR( "missing 'Ref=' token", aReader, line );
line+= len; parseQuotedString( reference, aReader, line, &line );
strCompare = "Part="; len = strlen( strCompare );
if( strncasecmp( strCompare, line, len ) != 0 ) SCH_PARSE_ERROR( "missing 'Part=' token", aReader, line );
line+= len; parseQuotedString( unit, aReader, line, &line );
long tmp;
if( !unit.ToLong( &tmp, 10 ) ) SCH_PARSE_ERROR( "expected integer value", aReader, line );
if( tmp < 0 || tmp > MAX_UNIT_COUNT_PER_PACKAGE ) SCH_PARSE_ERROR( "unit value out of range", aReader, line );
component->AddHierarchicalReference( path, reference, (int)tmp ); component->GetField( REFERENCE )->SetText( reference );
} else if( strCompare( "F", line, &line ) ) { int index = parseInt( aReader, line, &line );
wxString text, name;
parseQuotedString( text, aReader, line, &line, true );
char orientation = parseChar( aReader, line, &line ); wxPoint pos; pos.x = parseInt( aReader, line, &line ); pos.y = parseInt( aReader, line, &line ); int size = parseInt( aReader, line, &line ); int attributes = parseHex( aReader, line, &line );
if( index >= component->GetFieldCount() ) { // The first MANDATOR_FIELDS _must_ be constructed within
// the SCH_COMPONENT constructor. This assert is simply here
// to guard against a change in that constructor.
wxASSERT( component->GetFieldCount() >= MANDATORY_FIELDS );
// Ignore the _supplied_ fieldNdx. It is not important anymore
// if within the user defined fields region (i.e. >= MANDATORY_FIELDS).
// We freely renumber the index to fit the next available field slot.
index = component->GetFieldCount(); // new has this index after insertion
SCH_FIELD field( wxPoint( 0, 0 ), -1, component.get(), name ); component->AddField( field ); }
// Prior to version 2 of the schematic file format, none of the following existed.
if( m_version > 1 ) { wxString textAttrs; char hjustify = parseChar( aReader, line, &line );
parseUnquotedString( textAttrs, aReader, line, &line );
// The name of the field is optional.
parseQuotedString( name, aReader, line, &line, true );
if( hjustify == 'L' ) component->GetField( index )->SetHorizJustify( GR_TEXT_HJUSTIFY_LEFT ); else if( hjustify == 'R' ) component->GetField( index )->SetHorizJustify( GR_TEXT_HJUSTIFY_RIGHT ); else if( hjustify != 'C' ) SCH_PARSE_ERROR( "component field text horizontal justification must be " "L, R, or C", aReader, line );
// We are guaranteed to have a least one character here for older file formats
// otherwise an exception would have been raised..
if( textAttrs[0] == 'T' ) component->GetField( index )->SetVertJustify( GR_TEXT_VJUSTIFY_TOP ); else if( textAttrs[0] == 'B' ) component->GetField( index )->SetVertJustify( GR_TEXT_VJUSTIFY_BOTTOM ); else if( textAttrs[0] != 'C' ) SCH_PARSE_ERROR( "component field text vertical justification must be " "B, T, or C", aReader, line );
// Newer file formats include the bold and italics text attribute.
if( textAttrs.Length() > 1 ) { if( textAttrs.Length() != 3 ) SCH_PARSE_ERROR( _( "component field text attributes must be 3 characters wide" ), aReader, line );
if( textAttrs[1] == 'I' ) component->GetField( index )->SetItalic( true ); else if( textAttrs[1] != 'N' ) SCH_PARSE_ERROR( "component field text italics indicator must be I or N", aReader, line );
if( textAttrs[2] == 'B' ) component->GetField( index )->SetBold( true ); else if( textAttrs[2] != 'N' ) SCH_PARSE_ERROR( "component field text bold indicator must be B or N", aReader, line ); } }
component->GetField( index )->SetText( text ); component->GetField( index )->SetTextPos( pos ); component->GetField( index )->SetVisible( !attributes ); component->GetField( index )->SetTextSize( wxSize( size, size ) );
if( orientation == 'H' ) component->GetField( index )->SetTextAngle( TEXT_ANGLE_HORIZ ); else if( orientation == 'V' ) component->GetField( index )->SetTextAngle( TEXT_ANGLE_VERT ); else SCH_PARSE_ERROR( "component field orientation must be H or V", aReader, line );
if( name.IsEmpty() ) name = TEMPLATE_FIELDNAME::GetDefaultFieldName( index );
component->GetField( index )->SetName( name ); } else if( strCompare( "$EndComp", line ) ) { // Ensure all flags (some are set by previous initializations) are reset:
component->ClearFlags(); return component.release(); } else { // There are two lines that begin with a tab or spaces that includes a line with the
// redundant position information and the transform matrix settings.
// Parse the redundant position information just the same to check for formatting
// errors.
parseInt( aReader, line, &line ); // Always 1.
parseInt( aReader, line, &line ); // The X coordinate.
parseInt( aReader, line, &line ); // The Y coordinate.
line = aReader.ReadLine();
TRANSFORM transform;
transform.x1 = parseInt( aReader, line, &line );
if( transform.x1 < -1 || transform.x1 > 1 ) SCH_PARSE_ERROR( "invalid component X1 transform value", aReader, line );
transform.y1 = parseInt( aReader, line, &line );
if( transform.y1 < -1 || transform.y1 > 1 ) SCH_PARSE_ERROR( "invalid component Y1 transform value", aReader, line );
transform.x2 = parseInt( aReader, line, &line );
if( transform.x2 < -1 || transform.x2 > 1 ) SCH_PARSE_ERROR( "invalid component X2 transform value", aReader, line );
transform.y2 = parseInt( aReader, line, &line );
if( transform.y2 < -1 || transform.y2 > 1 ) SCH_PARSE_ERROR( "invalid component Y2 transform value", aReader, line );
component->SetTransform( transform ); }
line = aReader.ReadLine(); }
SCH_PARSE_ERROR( "invalid component line", aReader, line );
return NULL; // Prevents compiler warning. Should never get here.
}
void SCH_LEGACY_PLUGIN::Save( const wxString& aFileName, SCH_SCREEN* aScreen, KIWAY* aKiway, const PROPERTIES* aProperties ){ wxCHECK_RET( aScreen != NULL, "NULL SCH_SCREEN object." ); wxCHECK_RET( !aFileName.IsEmpty(), "No schematic file name defined." );
LOCALE_IO toggle; // toggles on, then off, the C locale, to write floating point values.
init( aKiway, aProperties );
wxFileName fn = aFileName;
// File names should be absolute. Don't assume everything relative to the project path
// works properly.
wxASSERT( fn.IsAbsolute() );
FILE_OUTPUTFORMATTER formatter( fn.GetFullPath() );
m_out = &formatter; // no ownership
Format( aScreen );}
void SCH_LEGACY_PLUGIN::Format( SCH_SCREEN* aScreen ){ wxCHECK_RET( aScreen != NULL, "NULL SCH_SCREEN* object." ); wxCHECK_RET( m_kiway != NULL, "NULL KIWAY* object." );
// Write the header
m_out->Print( 0, "%s %s %d\n", "EESchema", SCHEMATIC_HEAD_STRING, EESCHEMA_VERSION );
// Write the project libraries.
for( const PART_LIB& lib : *m_kiway->Prj().SchLibs() ) m_out->Print( 0, "LIBS:%s\n", TO_UTF8( lib.GetName() ) );
// This section is not used, but written for file compatibility
m_out->Print( 0, "EELAYER %d %d\n", SCH_LAYER_ID_COUNT, 0 ); m_out->Print( 0, "EELAYER END\n" );
/* Write page info, ScreenNumber and NumberOfScreen; not very meaningful for
* SheetNumber and Sheet Count in a complex hierarchy, but useful in * simple hierarchy and flat hierarchy. Used also to search the root * sheet ( ScreenNumber = 1 ) within the files */ const TITLE_BLOCK& tb = aScreen->GetTitleBlock(); const PAGE_INFO& page = aScreen->GetPageSettings();
m_out->Print( 0, "$Descr %s %d %d%s\n", TO_UTF8( page.GetType() ), page.GetWidthMils(), page.GetHeightMils(), !page.IsCustom() && page.IsPortrait() ? " portrait" : "" ); m_out->Print( 0, "encoding utf-8\n" ); m_out->Print( 0, "Sheet %d %d\n", aScreen->m_ScreenNumber, aScreen->m_NumberOfScreens ); m_out->Print( 0, "Title %s\n", EscapedUTF8( tb.GetTitle() ).c_str() ); m_out->Print( 0, "Date %s\n", EscapedUTF8( tb.GetDate() ).c_str() ); m_out->Print( 0, "Rev %s\n", EscapedUTF8( tb.GetRevision() ).c_str() ); m_out->Print( 0, "Comp %s\n", EscapedUTF8( tb.GetCompany() ).c_str() ); m_out->Print( 0, "Comment1 %s\n", EscapedUTF8( tb.GetComment1() ).c_str() ); m_out->Print( 0, "Comment2 %s\n", EscapedUTF8( tb.GetComment2() ).c_str() ); m_out->Print( 0, "Comment3 %s\n", EscapedUTF8( tb.GetComment3() ).c_str() ); m_out->Print( 0, "Comment4 %s\n", EscapedUTF8( tb.GetComment4() ).c_str() ); m_out->Print( 0, "$EndDescr\n" );
for( SCH_ITEM* item = aScreen->GetDrawItems(); item; item = item->Next() ) { switch( item->Type() ) { case SCH_COMPONENT_T: saveComponent( static_cast< SCH_COMPONENT* >( item ) ); break; case SCH_BITMAP_T: saveBitmap( static_cast< SCH_BITMAP* >( item ) ); break; case SCH_SHEET_T: saveSheet( static_cast< SCH_SHEET* >( item ) ); break; case SCH_JUNCTION_T: saveJunction( static_cast< SCH_JUNCTION* >( item ) ); break; case SCH_NO_CONNECT_T: saveNoConnect( static_cast< SCH_NO_CONNECT* >( item ) ); break; case SCH_BUS_WIRE_ENTRY_T: case SCH_BUS_BUS_ENTRY_T: saveBusEntry( static_cast< SCH_BUS_ENTRY_BASE* >( item ) ); break; case SCH_LINE_T: saveLine( static_cast< SCH_LINE* >( item ) ); break; case SCH_TEXT_T: case SCH_LABEL_T: case SCH_GLOBAL_LABEL_T: case SCH_HIERARCHICAL_LABEL_T: saveText( static_cast< SCH_TEXT* >( item ) ); break; default: wxASSERT( "Unexpected schematic object type in SCH_LEGACY_PLUGIN::Format()" ); } }
m_out->Print( 0, "$EndSCHEMATC\n" );}
void SCH_LEGACY_PLUGIN::saveComponent( SCH_COMPONENT* aComponent ){ std::string name1; std::string name2; wxArrayString reference_fields;
static wxString delimiters( wxT( " " ) );
// This is redundant with the AR entries below, but it makes the files backwards-compatible.
if( aComponent->GetPathsAndReferences().GetCount() > 0 ) { reference_fields = wxStringTokenize( aComponent->GetPathsAndReferences()[0], delimiters ); name1 = toUTFTildaText( reference_fields[1] ); } else { if( aComponent->GetField( REFERENCE )->GetText().IsEmpty() ) name1 = toUTFTildaText( aComponent->GetPrefix() ); else name1 = toUTFTildaText( aComponent->GetField( REFERENCE )->GetText() ); }
wxString part_name = aComponent->GetLibId().Format();
if( part_name.size() ) { name2 = toUTFTildaText( part_name ); } else { name2 = "_NONAME_"; }
m_out->Print( 0, "$Comp\n" ); m_out->Print( 0, "L %s %s\n", name2.c_str(), name1.c_str() );
// Generate unit number, convert and time stamp
m_out->Print( 0, "U %d %d %8.8lX\n", aComponent->GetUnit(), aComponent->GetConvert(), (unsigned long)aComponent->GetTimeStamp() );
// Save the position
m_out->Print( 0, "P %d %d\n", aComponent->GetPosition().x, aComponent->GetPosition().y );
/* If this is a complex hierarchy; save hierarchical references.
* but for simple hierarchies it is not necessary. * the reference inf is already saved * this is useful for old Eeschema version compatibility */ if( aComponent->GetPathsAndReferences().GetCount() > 1 ) { for( unsigned int ii = 0; ii < aComponent->GetPathsAndReferences().GetCount(); ii++ ) { /*format:
* AR Path="/140/2" Ref="C99" Part="1" * where 140 is the uid of the containing sheet * and 2 is the timestamp of this component. * (timestamps are actually 8 hex chars) * Ref is the conventional component reference for this 'path' * Part is the conventional component part selection for this 'path' */ reference_fields = wxStringTokenize( aComponent->GetPathsAndReferences()[ii], delimiters );
m_out->Print( 0, "AR Path=\"%s\" Ref=\"%s\" Part=\"%s\" \n", TO_UTF8( reference_fields[0] ), TO_UTF8( reference_fields[1] ), TO_UTF8( reference_fields[2] ) ); } }
// update the ugly field index, which I would like to see go away someday soon.
for( int i = 0; i < aComponent->GetFieldCount(); ++i ) aComponent->GetField( i )->SetId( i );
// Fixed fields:
// Save mandatory fields even if they are blank,
// because the visibility, size and orientation are set from libary editor.
for( unsigned i = 0; i < MANDATORY_FIELDS; ++i ) saveField( aComponent->GetField( i ) );
// User defined fields:
// The *policy* about which user defined fields are part of a symbol is now
// only in the dialog editors. No policy should be enforced here, simply
// save all the user defined fields, they are present because a dialog editor
// thought they should be. If you disagree, go fix the dialog editors.
for( int i = MANDATORY_FIELDS; i < aComponent->GetFieldCount(); ++i ) saveField( aComponent->GetField( i ) );
// Unit number, position, box ( old standard )
m_out->Print( 0, "\t%-4d %-4d %-4d\n", aComponent->GetUnit(), aComponent->GetPosition().x, aComponent->GetPosition().y );
TRANSFORM transform = aComponent->GetTransform();
m_out->Print( 0, "\t%-4d %-4d %-4d %-4d\n", transform.x1, transform.y1, transform.x2, transform.y2 ); m_out->Print( 0, "$EndComp\n" );}
void SCH_LEGACY_PLUGIN::saveField( SCH_FIELD* aField ){ char hjustify = 'C';
if( aField->GetHorizJustify() == GR_TEXT_HJUSTIFY_LEFT ) hjustify = 'L'; else if( aField->GetHorizJustify() == GR_TEXT_HJUSTIFY_RIGHT ) hjustify = 'R';
char vjustify = 'C';
if( aField->GetVertJustify() == GR_TEXT_VJUSTIFY_BOTTOM ) vjustify = 'B'; else if( aField->GetVertJustify() == GR_TEXT_VJUSTIFY_TOP ) vjustify = 'T';
m_out->Print( 0, "F %d %s %c %-3d %-3d %-3d %4.4X %c %c%c%c", aField->GetId(), EscapedUTF8( aField->GetText() ).c_str(), // wraps in quotes too
aField->GetTextAngle() == TEXT_ANGLE_HORIZ ? 'H' : 'V', aField->GetLibPosition().x, aField->GetLibPosition().y, aField->GetTextWidth(), !aField->IsVisible(), hjustify, vjustify, aField->IsItalic() ? 'I' : 'N', aField->IsBold() ? 'B' : 'N' );
// Save field name, if the name is user definable
if( aField->GetId() >= FIELD1 ) { m_out->Print( 0, " %s", EscapedUTF8( aField->GetName() ).c_str() ); }
m_out->Print( 0, "\n" );}
void SCH_LEGACY_PLUGIN::saveBitmap( SCH_BITMAP* aBitmap ){ wxCHECK_RET( aBitmap != NULL, "SCH_BITMAP* is NULL" );
wxImage* image = aBitmap->GetImage()->GetImageData();
wxCHECK_RET( image != NULL, "wxImage* is NULL" );
m_out->Print( 0, "$Bitmap\n" ); m_out->Print( 0, "Pos %-4d %-4d\n", aBitmap->GetPosition().x, aBitmap->GetPosition().y ); m_out->Print( 0, "Scale %f\n", aBitmap->GetImage()->GetScale() ); m_out->Print( 0, "Data\n" );
wxMemoryOutputStream stream;
image->SaveFile( stream, wxBITMAP_TYPE_PNG );
// Write binary data in hexadecimal form (ASCII)
wxStreamBuffer* buffer = stream.GetOutputStreamBuffer(); char* begin = (char*) buffer->GetBufferStart();
for( int ii = 0; begin < buffer->GetBufferEnd(); begin++, ii++ ) { if( ii >= 32 ) { ii = 0;
m_out->Print( 0, "\n" ); }
m_out->Print( 0, "%2.2X ", *begin & 0xFF ); }
m_out->Print( 0, "\nEndData\n" ); m_out->Print( 0, "$EndBitmap\n" );}
void SCH_LEGACY_PLUGIN::saveSheet( SCH_SHEET* aSheet ){ wxCHECK_RET( aSheet != NULL, "SCH_SHEET* is NULL" );
m_out->Print( 0, "$Sheet\n" ); m_out->Print( 0, "S %-4d %-4d %-4d %-4d\n", aSheet->GetPosition().x, aSheet->GetPosition().y, aSheet->GetSize().x, aSheet->GetSize().y );
m_out->Print( 0, "U %8.8lX\n", (unsigned long) aSheet->GetTimeStamp() );
if( !aSheet->GetName().IsEmpty() ) m_out->Print( 0, "F0 %s %d\n", EscapedUTF8( aSheet->GetName() ).c_str(), aSheet->GetSheetNameSize() );
if( !aSheet->GetFileName().IsEmpty() ) m_out->Print( 0, "F1 %s %d\n", EscapedUTF8( aSheet->GetFileName() ).c_str(), aSheet->GetFileNameSize() );
for( const SCH_SHEET_PIN& pin : aSheet->GetPins() ) { int type, side;
if( pin.GetText().IsEmpty() ) break;
switch( pin.GetEdge() ) { default: case SCH_SHEET_PIN::SHEET_LEFT_SIDE: side = 'L'; break;
case SCH_SHEET_PIN::SHEET_RIGHT_SIDE: side = 'R'; break;
case SCH_SHEET_PIN::SHEET_TOP_SIDE: side = 'T'; break;
case SCH_SHEET_PIN::SHEET_BOTTOM_SIDE: side = 'B'; break; }
switch( pin.GetShape() ) { case NET_INPUT: type = 'I'; break;
case NET_OUTPUT: type = 'O'; break;
case NET_BIDI: type = 'B'; break;
case NET_TRISTATE: type = 'T'; break;
default: case NET_UNSPECIFIED: type = 'U'; break; }
m_out->Print( 0, "F%d %s %c %c %-3d %-3d %-3d\n", pin.GetNumber(), EscapedUTF8( pin.GetText() ).c_str(), // supplies wrapping quotes
type, side, pin.GetPosition().x, pin.GetPosition().y, pin.GetTextWidth() ); }
m_out->Print( 0, "$EndSheet\n" );}
void SCH_LEGACY_PLUGIN::saveJunction( SCH_JUNCTION* aJunction ){ wxCHECK_RET( aJunction != NULL, "SCH_JUNCTION* is NULL" );
m_out->Print( 0, "Connection ~ %-4d %-4d\n", aJunction->GetPosition().x, aJunction->GetPosition().y );}
void SCH_LEGACY_PLUGIN::saveNoConnect( SCH_NO_CONNECT* aNoConnect ){ wxCHECK_RET( aNoConnect != NULL, "SCH_NOCONNECT* is NULL" );
m_out->Print( 0, "NoConn ~ %-4d %-4d\n", aNoConnect->GetPosition().x, aNoConnect->GetPosition().y );}
void SCH_LEGACY_PLUGIN::saveBusEntry( SCH_BUS_ENTRY_BASE* aBusEntry ){ wxCHECK_RET( aBusEntry != NULL, "SCH_BUS_ENTRY_BASE* is NULL" );
if( aBusEntry->GetLayer() == LAYER_WIRE ) m_out->Print( 0, "Entry Wire Line\n\t%-4d %-4d %-4d %-4d\n", aBusEntry->GetPosition().x, aBusEntry->GetPosition().y, aBusEntry->m_End().x, aBusEntry->m_End().y ); else m_out->Print( 0, "Entry Bus Bus\n\t%-4d %-4d %-4d %-4d\n", aBusEntry->GetPosition().x, aBusEntry->GetPosition().y, aBusEntry->m_End().x, aBusEntry->m_End().y );}
void SCH_LEGACY_PLUGIN::saveLine( SCH_LINE* aLine ){ wxCHECK_RET( aLine != NULL, "SCH_LINE* is NULL" );
const char* layer = "Notes"; const char* width = "Line";
if( aLine->GetLayer() == LAYER_WIRE ) layer = "Wire"; else if( aLine->GetLayer() == LAYER_BUS ) layer = "Bus";
m_out->Print( 0, "Wire %s %s", layer, width );
// Write line style (width, type, color) only for non default values
if( aLine->GetLayer() == LAYER_NOTES ) { if( aLine->GetPenSize() != aLine->GetDefaultWidth() ) m_out->Print( 0, " %s %d", T_WIDTH, aLine->GetLineSize() );
if( aLine->GetLineStyle() != aLine->GetDefaultStyle() ) m_out->Print( 0, " %s %s", T_STYLE, SCH_LINE::GetLineStyleName( aLine->GetLineStyle() ) );
if( aLine->GetLineColor() != aLine->GetDefaultColor() ) m_out->Print( 0, " %s", TO_UTF8( aLine->GetLineColor().ToColour().GetAsString( wxC2S_CSS_SYNTAX ) ) ); }
m_out->Print( 0, "\n" );
m_out->Print( 0, "\t%-4d %-4d %-4d %-4d", aLine->GetStartPoint().x, aLine->GetStartPoint().y, aLine->GetEndPoint().x, aLine->GetEndPoint().y );
m_out->Print( 0, "\n");}
void SCH_LEGACY_PLUGIN::saveText( SCH_TEXT* aText ){ wxCHECK_RET( aText != NULL, "SCH_TEXT* is NULL" );
const char* italics = "~"; const char* textType = "Notes";
if( aText->IsItalic() ) italics = "Italic";
wxString text = aText->GetText();
SCH_LAYER_ID layer = aText->GetLayer();
if( layer == LAYER_NOTES || layer == LAYER_LOCLABEL ) { if( layer == LAYER_NOTES ) { // For compatibility reasons, the text must be saved in only one text line
// so replace all EOLs with \\n
text.Replace( wxT( "\n" ), wxT( "\\n" ) );
// Here we should have no CR or LF character in line
// This is not always the case if a multiline text was copied (using a copy/paste
// function) from a text that uses E.O.L characters that differs from the current
// EOL format. This is mainly the case under Linux using LF symbol when copying
// a text from Windows (using CRLF symbol) so we must just remove the extra CR left
// (or LF left under MacOSX)
for( unsigned ii = 0; ii < text.Len(); ) { if( text[ii] == 0x0A || text[ii] == 0x0D ) text.erase( ii, 1 ); else ii++; } } else { textType = "Label"; }
m_out->Print( 0, "Text %s %-4d %-4d %-4d %-4d %s %d\n%s\n", textType, aText->GetPosition().x, aText->GetPosition().y, aText->GetLabelSpinStyle(), aText->GetTextWidth(), italics, aText->GetThickness(), TO_UTF8( text ) ); } else if( layer == LAYER_GLOBLABEL || layer == LAYER_HIERLABEL ) { textType = ( layer == LAYER_GLOBLABEL ) ? "GLabel" : "HLabel";
m_out->Print( 0, "Text %s %-4d %-4d %-4d %-4d %s %s %d\n%s\n", textType, aText->GetPosition().x, aText->GetPosition().y, aText->GetLabelSpinStyle(), aText->GetTextWidth(), SheetLabelType[aText->GetShape()], italics, aText->GetThickness(), TO_UTF8( text ) ); }}
int SCH_LEGACY_PLUGIN_CACHE::m_modHash = 1; // starts at 1 and goes up
SCH_LEGACY_PLUGIN_CACHE::SCH_LEGACY_PLUGIN_CACHE( const wxString& aFullPathAndFileName ) : m_fileName( aFullPathAndFileName ), m_libFileName( aFullPathAndFileName ), m_isWritable( true ), m_isModified( false ){ m_versionMajor = -1; m_versionMinor = -1; m_libType = LIBRARY_TYPE_EESCHEMA;}
SCH_LEGACY_PLUGIN_CACHE::~SCH_LEGACY_PLUGIN_CACHE(){ // When the cache is destroyed, all of the alias objects on the heap should be deleted.
for( LIB_ALIAS_MAP::iterator it = m_aliases.begin(); it != m_aliases.end(); ++it ) { wxLogTrace( traceSchLegacyPlugin, wxT( "Removing alias %s from library %s." ), GetChars( it->second->GetName() ), GetChars( GetLogicalName() ) ); LIB_PART* part = it->second->GetPart(); LIB_ALIAS* alias = it->second; delete alias;
// When the last alias of a part is destroyed, the part is no longer required and it
// too is destroyed.
if( part && part->GetAliasCount() == 0 ) delete part; }
m_aliases.clear();}
// If m_libFileName is a symlink follow it to the real source file
wxFileName SCH_LEGACY_PLUGIN_CACHE::GetRealFile() const{ wxFileName fn( m_libFileName );
#ifndef __WINDOWS__
if( fn.Exists( wxFILE_EXISTS_SYMLINK ) ) { char buffer[ PATH_MAX + 1 ]; ssize_t pathLen = readlink( TO_UTF8( fn.GetFullPath() ), buffer, PATH_MAX );
if( pathLen > 0 ) { buffer[ pathLen ] = '\0'; fn.Assign( fn.GetPath() + wxT( "/" ) + wxString::FromUTF8( buffer ) ); fn.Normalize(); } }#endif
return fn;}
wxDateTime SCH_LEGACY_PLUGIN_CACHE::GetLibModificationTime(){ wxFileName fn = GetRealFile();
// update the writable flag while we have a wxFileName, in a network this
// is possibly quite dynamic anyway.
m_isWritable = fn.IsFileWritable();
return fn.GetModificationTime();}
bool SCH_LEGACY_PLUGIN_CACHE::IsFile( const wxString& aFullPathAndFileName ) const{ return m_fileName == aFullPathAndFileName;}
bool SCH_LEGACY_PLUGIN_CACHE::IsFileChanged() const{ wxFileName fn = GetRealFile();
if( m_fileModTime.IsValid() && fn.IsOk() && fn.FileExists() ) return fn.GetModificationTime() != m_fileModTime;
return false;}
LIB_ALIAS* SCH_LEGACY_PLUGIN_CACHE::removeAlias( LIB_ALIAS* aAlias ){ wxCHECK_MSG( aAlias != NULL, NULL, "NULL pointer cannot be removed from library." );
LIB_ALIAS_MAP::iterator it = m_aliases.find( aAlias->GetName() );
if( it == m_aliases.end() ) return NULL;
// If the entry pointer doesn't match the name it is mapped to in the library, we
// have done something terribly wrong.
wxCHECK_MSG( *it->second == aAlias, NULL, "Pointer mismatch while attempting to remove alias entry <" + aAlias->GetName() + "> from library cache <" + m_libFileName.GetName() + ">." );
LIB_ALIAS* alias = aAlias; LIB_PART* part = alias->GetPart();
alias = part->RemoveAlias( alias );
if( !alias ) { delete part;
if( m_aliases.size() > 1 ) { LIB_ALIAS_MAP::iterator next = it; next++;
if( next == m_aliases.end() ) next = m_aliases.begin();
alias = next->second; } }
m_aliases.erase( it ); m_isModified = true; ++m_modHash; return alias;}
void SCH_LEGACY_PLUGIN_CACHE::AddSymbol( const LIB_PART* aPart ){ // aPart is cloned in PART_LIB::AddPart(). The cache takes ownership of aPart.
wxArrayString aliasNames = aPart->GetAliasNames();
for( size_t i = 0; i < aliasNames.size(); i++ ) { LIB_ALIAS_MAP::iterator it = m_aliases.find( aliasNames[i] );
if( it != m_aliases.end() ) removeAlias( it->second );
LIB_ALIAS* alias = const_cast< LIB_PART* >( aPart )->GetAlias( aliasNames[i] );
wxASSERT_MSG( alias != NULL, "No alias <" + aliasNames[i] + "> found in symbol <" + aPart->GetName() +">." );
m_aliases[ aliasNames[i] ] = alias; }
m_isModified = true; ++m_modHash;}
void SCH_LEGACY_PLUGIN_CACHE::Load(){ if( !m_libFileName.FileExists() ) { wxString msg = wxString::Format( _( "Library file \"%s\" not found.\n\n" "Use the Manage Symbol Libraries dialog to fix the " "path (or remove the library)." ), m_libFileName.GetFullPath() ); KIDIALOG dlg( Pgm().App().GetTopWindow(), msg, KIDIALOG::KD_ERROR ); dlg.DoNotShowCheckbox( __FILE__, __LINE__ ); dlg.ShowModal(); return; }
wxCHECK_RET( m_libFileName.IsAbsolute(), wxString::Format( "Cannot use relative file paths in legacy plugin to " "open library \"%s\".", m_libFileName.GetFullPath() ) );
wxLogTrace( traceSchLegacyPlugin, "Loading legacy symbol file \"%s\"", m_libFileName.GetFullPath() );
FILE_LINE_READER reader( m_libFileName.GetFullPath() );
if( !reader.ReadLine() ) THROW_IO_ERROR( _( "unexpected end of file" ) );
const char* line = reader.Line();
if( !strCompare( "EESchema-LIBRARY Version", line, &line ) ) { // Old .sym files (which are libraries with only one symbol, used to store and reuse shapes)
// EESchema-LIB Version x.x SYMBOL. They are valid files.
if( !strCompare( "EESchema-LIB Version", line, &line ) ) SCH_PARSE_ERROR( "file is not a valid component or symbol library file", reader, line ); }
m_versionMajor = parseInt( reader, line, &line );
if( *line != '.' ) SCH_PARSE_ERROR( "invalid file version formatting in header", reader, line );
line++;
m_versionMinor = parseInt( reader, line, &line );
if( m_versionMajor < 1 || m_versionMinor < 0 || m_versionMinor > 99 ) SCH_PARSE_ERROR( "invalid file version in header", reader, line );
// Check if this is a symbol library which is the same as a component library but without
// any alias, documentation, footprint filters, etc.
if( strCompare( "SYMBOL", line, &line ) ) { // Symbol files add date and time stamp info to the header.
m_libType = LIBRARY_TYPE_SYMBOL;
/// @todo Probably should check for a valid date and time stamp even though it's not used.
} else { m_libType = LIBRARY_TYPE_EESCHEMA; }
while( reader.ReadLine() ) { line = reader.Line();
if( *line == '#' || isspace( *line ) ) // Skip comments and blank lines.
continue;
// Headers where only supported in older library file formats.
if( m_libType == LIBRARY_TYPE_EESCHEMA && strCompare( "$HEADER", line ) ) loadHeader( reader );
if( strCompare( "DEF", line ) ) { // Read one DEF/ENDDEF part entry from library:
loadPart( reader ); } }
++m_modHash;
// Remember the file modification time of library file when the
// cache snapshot was made, so that in a networked environment we will
// reload the cache as needed.
m_fileModTime = GetLibModificationTime();
if( USE_OLD_DOC_FILE_FORMAT( m_versionMajor, m_versionMinor ) ) loadDocs();}
void SCH_LEGACY_PLUGIN_CACHE::loadDocs(){ const char* line; wxString text; wxString aliasName; wxFileName fn = m_libFileName; LIB_ALIAS* alias = NULL;;
fn.SetExt( DOC_EXT );
// Not all libraries will have a document file.
if( !fn.FileExists() ) return;
if( !fn.IsFileReadable() ) THROW_IO_ERROR( wxString::Format( _( "user does not have permission to read library " "document file \"%s\"" ), fn.GetFullPath() ) );
FILE_LINE_READER reader( fn.GetFullPath() );
line = reader.ReadLine();
if( !line ) THROW_IO_ERROR( _( "symbol document library file is empty" ) );
if( !strCompare( DOCFILE_IDENT, line, &line ) ) SCH_PARSE_ERROR( "invalid document library file version formatting in header", reader, line );
while( reader.ReadLine() ) { line = reader.Line();
if( *line == '#' ) // Comment line.
continue;
if( !strCompare( "$CMP", line, &line ) != 0 ) SCH_PARSE_ERROR( "$CMP command expected", reader, line );
parseUnquotedString( aliasName, reader, line, &line ); // Alias name.
aliasName = LIB_ID::FixIllegalChars( aliasName, LIB_ID::ID_SCH ); LIB_ALIAS_MAP::iterator it = m_aliases.find( aliasName );
if( it == m_aliases.end() ) wxLogWarning( "Alias '%s' not found in library:\n\n" "'%s'\n\nat line %d offset %d", aliasName, fn.GetFullPath(), reader.LineNumber(), (int) (line - reader.Line() ) ); else alias = it->second;
// Read the curent alias associated doc.
// if the alias does not exist, just skip the description
// (Can happen if a .dcm is not synchronized with the corresponding .lib file)
while( reader.ReadLine() ) { line = reader.Line();
if( !line ) SCH_PARSE_ERROR( "unexpected end of file", reader, line );
if( strCompare( "$ENDCMP", line, &line ) ) break;
text = FROM_UTF8( line + 2 ); // Remove spaces at eol, and eol chars:
text = text.Trim();
switch( line[0] ) { case 'D': if( alias ) alias->SetDescription( text ); break;
case 'K': if( alias ) alias->SetKeyWords( text ); break;
case 'F': if( alias ) alias->SetDocFileName( text ); break;
case 0: case '\n': case '\r': case '#': // Empty line or commment
break;
default: SCH_PARSE_ERROR( "expected token in symbol definition", reader, line ); } } }}
void SCH_LEGACY_PLUGIN_CACHE::loadHeader( FILE_LINE_READER& aReader ){ const char* line = aReader.Line();
wxASSERT( strCompare( "$HEADER", line, &line ) );
while( aReader.ReadLine() ) { line = (char*) aReader;
// The time stamp saved in old library files is not used or saved in the latest
// library file version.
if( strCompare( "TimeStamp", line, &line ) ) continue; else if( strCompare( "$ENDHEADER", line, &line ) ) return; }
SCH_PARSE_ERROR( "$ENDHEADER not found", aReader, line );}
LIB_PART* SCH_LEGACY_PLUGIN_CACHE::loadPart( FILE_LINE_READER& aReader ){ const char* line = aReader.Line();
wxCHECK( strCompare( "DEF", line, &line ), NULL );
// Read DEF line:
char yes_no = 0;
std::unique_ptr< LIB_PART > part( new LIB_PART( wxEmptyString ) );
wxString name, prefix;
parseUnquotedString( name, aReader, line, &line ); // Part name.
parseUnquotedString( prefix, aReader, line, &line ); // Prefix name
parseInt( aReader, line, &line ); // NumOfPins, unused.
part->SetPinNameOffset( parseInt( aReader, line, &line ) ); // Pin name offset.
yes_no = parseChar( aReader, line, &line ); // Show pin numbers.
if( !( yes_no == 'Y' || yes_no == 'N') ) SCH_PARSE_ERROR( "expected Y or N", aReader, line );
part->SetShowPinNumbers( ( yes_no == 'N' ) ? false : true );
yes_no = parseChar( aReader, line, &line ); // Show pin numbers.
if( !( yes_no == 'Y' || yes_no == 'N') ) SCH_PARSE_ERROR( "expected Y or N", aReader, line );
part->SetShowPinNames( ( yes_no == 'N' ) ? false : true ); // Show pin names.
part->SetUnitCount( parseInt( aReader, line, &line ) ); // Number of units.
// Ensure m_unitCount is >= 1. Could be read as 0 in old libraries.
if( part->GetUnitCount() < 1 ) part->SetUnitCount( 1 );
// Copy part name and prefix.
// The root alias is added to the alias list by SetName() which is called by SetText().
if( name.IsEmpty() ) { part->SetName( "~" ); } else if( name[0] != '~' ) { part->SetName( name ); } else { part->SetName( name.Right( name.Length() - 1 ) ); part->GetValueField().SetVisible( false ); }
// Don't set the library alias, this is determined by the symbol library table.
part->SetLibId( LIB_ID( wxEmptyString, part->GetName() ) );
// There are some code paths in SetText() that do not set the root alias to the
// alias list so add it here if it didn't get added by SetText().
if( !part->HasAlias( part->GetName() ) ) part->AddAlias( part->GetName() );
LIB_FIELD& reference = part->GetReferenceField();
if( prefix == "~" ) { reference.Empty(); reference.SetVisible( false ); } else { reference.SetText( prefix ); }
// In version 2.2 and earlier, this parameter was a '0' which was just a place holder.
// The was no concept of interchangeable multiple unit symbols.
if( LIB_VERSION( m_versionMajor, m_versionMinor ) <= LIB_VERSION( 2, 2 ) ) { // Nothing needs to be set since the default setting for symbols with multiple
// units were never interchangeable. Just parse the 0 an move on.
parseInt( aReader, line, &line ); } else { char locked = parseChar( aReader, line, &line );
if( locked == 'L' ) part->LockUnits( true ); else if( locked == 'F' || locked == '0' ) part->LockUnits( false ); else SCH_PARSE_ERROR( "expected L, F, or 0", aReader, line ); }
// There is the optional power component flag.
if( *line ) { char power = parseChar( aReader, line, &line );
if( power == 'P' ) part->SetPower(); else if( power == 'N' ) part->SetNormal(); else SCH_PARSE_ERROR( "expected P or N", aReader, line ); }
line = aReader.ReadLine();
// Read lines until "ENDDEF" is found.
while( line ) { if( *line == '#' ) // Comment
; else if( strCompare( "Ti", line, &line ) ) // Modification date is ignored.
continue; else if( strCompare( "ALIAS", line, &line ) ) // Aliases
loadAliases( part, aReader ); else if( *line == 'F' ) // Fields
loadField( part, aReader ); else if( strCompare( "DRAW", line, &line ) ) // Drawing objects.
loadDrawEntries( part, aReader ); else if( strCompare( "$FPLIST", line, &line ) ) // Footprint filter list
loadFootprintFilters( part, aReader ); else if( strCompare( "ENDDEF", line, &line ) ) // End of part description
{ // Add aliases
for( size_t ii = 0; ii < part->GetAliasCount(); ++ii ) { LIB_ALIAS* alias = part->GetAlias( ii ); const wxString& aliasName = alias->GetName(); auto it = m_aliases.find( aliasName );
if( it != m_aliases.end() ) { // Find a new name for the alias
wxString newName; int idx = 0; LIB_ALIAS_MAP::const_iterator jt;
do { newName = wxString::Format( "%s_%d", aliasName, idx ); jt = m_aliases.find( newName ); ++idx; } while( jt != m_aliases.end() );
wxLogWarning( "Symbol name conflict in library:\n%s\n" "'%s' has been renamed to '%s'", m_fileName, aliasName, newName );
if( alias->IsRoot() ) part->SetName( newName ); else alias->SetName( newName );
m_aliases[newName] = alias; } else { m_aliases[aliasName] = alias; } }
return part.release(); }
line = aReader.ReadLine(); }
SCH_PARSE_ERROR( "missing ENDDEF", aReader, line );}
bool SCH_LEGACY_PLUGIN_CACHE::checkForDuplicates( wxString& aAliasName ){ wxCHECK_MSG( !aAliasName.IsEmpty(), false, "alias name cannot be empty" );
// The alias name is not a duplicate so don't change it.
if( m_aliases.find( aAliasName ) == m_aliases.end() ) return false;
int dupCounter = 1; wxString newAlias = aAliasName;
// If the alias is already loaded, the library is broken. It may have been possible in
// the past that this could happen so we assign a new alias name to prevent any conflicts
// rather than throw an exception.
while( m_aliases.find( newAlias ) != m_aliases.end() ) { newAlias = aAliasName << dupCounter; dupCounter++; }
aAliasName = newAlias;
return true;}
void SCH_LEGACY_PLUGIN_CACHE::loadAliases( std::unique_ptr< LIB_PART >& aPart, FILE_LINE_READER& aReader ){ wxString newAlias; const char* line = aReader.Line();
wxCHECK_RET( strCompare( "ALIAS", line, &line ), "Invalid ALIAS section" );
// Parse the ALIAS list.
wxString alias; parseUnquotedString( alias, aReader, line, &line );
while( !alias.IsEmpty() ) { newAlias = alias; checkForDuplicates( newAlias ); aPart->AddAlias( newAlias ); alias.clear(); parseUnquotedString( alias, aReader, line, &line, true ); }}
void SCH_LEGACY_PLUGIN_CACHE::loadField( std::unique_ptr< LIB_PART >& aPart, FILE_LINE_READER& aReader ){ const char* line = aReader.Line();
wxCHECK_RET( *line == 'F', "Invalid field line" );
int id;
if( sscanf( line + 1, "%d", &id ) != 1 || id < 0 ) SCH_PARSE_ERROR( "invalid field ID", aReader, line + 1 );
LIB_FIELD* field;
if( (unsigned) id < MANDATORY_FIELDS ) { field = aPart->GetField( id );
// this will fire only if somebody broke a constructor or editor.
// MANDATORY_FIELDS are always present in ram resident components, no
// exceptions, and they always have their names set, even fixed fields.
wxASSERT( field ); } else { field = new LIB_FIELD( aPart.get(), id ); aPart->AddDrawItem( field ); }
// Skip to the first double quote.
while( *line != '"' && *line != 0 ) line++;
if( *line == 0 ) SCH_PARSE_ERROR( _( "unexpected end of line" ), aReader, line );
parseQuotedString( field->m_Text, aReader, line, &line, true );
// Doctor the *.lib file field which has a "~" in blank fields. New saves will
// not save like this.
if( field->m_Text.size() == 1 && field->m_Text[0] == '~' ) field->m_Text.clear();
wxPoint pos;
pos.x = parseInt( aReader, line, &line ); pos.y = parseInt( aReader, line, &line ); field->SetPosition( pos );
wxSize textSize;
textSize.x = textSize.y = parseInt( aReader, line, &line ); field->SetTextSize( textSize );
char textOrient = parseChar( aReader, line, &line );
if( textOrient == 'H' ) field->SetTextAngle( TEXT_ANGLE_HORIZ ); else if( textOrient == 'V' ) field->SetTextAngle( TEXT_ANGLE_VERT ); else SCH_PARSE_ERROR( "invalid field text orientation parameter", aReader, line );
char textVisible = parseChar( aReader, line, &line );
if( textVisible == 'V' ) field->SetVisible( true ); else if ( textVisible == 'I' ) field->SetVisible( false ); else SCH_PARSE_ERROR( "invalid field text visibility parameter", aReader, line );
// It may be technically correct to use the library version to determine if the field text
// attributes are present. If anyone knows if that is valid and what version that would be,
// please change this to test the library version rather than an EOL or the quoted string
// of the field name.
if( *line != 0 && *line != '"' ) { char textHJustify = parseChar( aReader, line, &line );
if( textHJustify == 'C' ) field->SetHorizJustify( GR_TEXT_HJUSTIFY_CENTER ); else if( textHJustify == 'L' ) field->SetHorizJustify( GR_TEXT_HJUSTIFY_LEFT ); else if( textHJustify == 'R' ) field->SetHorizJustify( GR_TEXT_HJUSTIFY_RIGHT ); else SCH_PARSE_ERROR( "invalid field text horizontal justification", aReader, line );
wxString attributes;
parseUnquotedString( attributes, aReader, line, &line );
size_t attrSize = attributes.size();
if( !(attrSize == 3 || attrSize == 1 ) ) SCH_PARSE_ERROR( "invalid field text attributes size", aReader, line );
switch( (wxChar) attributes[0] ) { case 'C': field->SetVertJustify( GR_TEXT_VJUSTIFY_CENTER ); break; case 'B': field->SetVertJustify( GR_TEXT_VJUSTIFY_BOTTOM ); break; case 'T': field->SetVertJustify( GR_TEXT_VJUSTIFY_TOP ); break; default: SCH_PARSE_ERROR( "invalid field text vertical justification", aReader, line ); }
if( attrSize == 3 ) { wxChar attr_1 = attributes[1]; wxChar attr_2 = attributes[2];
if( attr_1 == 'I' ) // Italic
field->SetItalic( true ); else if( attr_1 != 'N' ) // No italics is default, check for error.
SCH_PARSE_ERROR( "invalid field text italic parameter", aReader, line );
if ( attr_2 == 'B' ) // Bold
field->SetBold( true ); else if( attr_2 != 'N' ) // No bold is default, check for error.
SCH_PARSE_ERROR( "invalid field text bold parameter", aReader, line ); } }
// Fields in RAM must always have names.
if( (unsigned) id < MANDATORY_FIELDS ) { // Fields in RAM must always have names, because we are trying to get
// less dependent on field ids and more dependent on names.
// Plus assumptions are made in the field editors.
field->m_name = TEMPLATE_FIELDNAME::GetDefaultFieldName( id );
// Ensure the VALUE field = the part name (can be not the case
// with malformed libraries: edited by hand, or converted from other tools)
if( id == VALUE ) field->m_Text = aPart->GetName(); } else { parseQuotedString( field->m_name, aReader, line, &line, true ); // Optional.
}}
void SCH_LEGACY_PLUGIN_CACHE::loadDrawEntries( std::unique_ptr< LIB_PART >& aPart, FILE_LINE_READER& aReader ){ const char* line = aReader.Line();
wxCHECK_RET( strCompare( "DRAW", line, &line ), "Invalid DRAW section" );
line = aReader.ReadLine();
while( line ) { if( strCompare( "ENDDRAW", line, &line ) ) return;
switch( line[0] ) { case 'A': // Arc
aPart->AddDrawItem( loadArc( aPart, aReader ) ); break;
case 'C': // Circle
aPart->AddDrawItem( loadCircle( aPart, aReader ) ); break;
case 'T': // Text
aPart->AddDrawItem( loadText( aPart, aReader ) ); break;
case 'S': // Square
aPart->AddDrawItem( loadRectangle( aPart, aReader ) ); break;
case 'X': // Pin Description
aPart->AddDrawItem( loadPin( aPart, aReader ) ); break;
case 'P': // Polyline
aPart->AddDrawItem( loadPolyLine( aPart, aReader ) ); break;
case 'B': // Bezier Curves
aPart->AddDrawItem( loadBezier( aPart, aReader ) ); break;
case '#': // Comment
case '\n': // Empty line
case '\r': case 0: break;
default: SCH_PARSE_ERROR( "undefined DRAW entry", aReader, line ); }
line = aReader.ReadLine(); }
SCH_PARSE_ERROR( "file ended prematurely loading component draw element", aReader, line );}
FILL_T SCH_LEGACY_PLUGIN_CACHE::parseFillMode( FILE_LINE_READER& aReader, const char* aLine, const char** aOutput ){ switch( parseChar( aReader, aLine, aOutput ) ) { case 'F': return FILLED_SHAPE; case 'f': return FILLED_WITH_BG_BODYCOLOR; case 'N': return NO_FILL; default: SCH_PARSE_ERROR( "invalid fill type, expected f, F, or N", aReader, aLine ); }}
LIB_ARC* SCH_LEGACY_PLUGIN_CACHE::loadArc( std::unique_ptr< LIB_PART >& aPart, FILE_LINE_READER& aReader ){ const char* line = aReader.Line();
wxCHECK_MSG( strCompare( "A", line, &line ), NULL, "Invalid LIB_ARC definition" );
LIB_ARC* arc = new LIB_ARC( aPart.get() );
wxPoint center;
center.x = parseInt( aReader, line, &line ); center.y = parseInt( aReader, line, &line );
arc->SetPosition( center ); arc->SetRadius( parseInt( aReader, line, &line ) );
int angle1 = parseInt( aReader, line, &line ); int angle2 = parseInt( aReader, line, &line );
NORMALIZE_ANGLE_POS( angle1 ); NORMALIZE_ANGLE_POS( angle2 ); arc->SetFirstRadiusAngle( angle1 ); arc->SetSecondRadiusAngle( angle2 );
arc->SetUnit( parseInt( aReader, line, &line ) ); arc->SetConvert( parseInt( aReader, line, &line ) ); arc->SetWidth( parseInt( aReader, line, &line ) );
// Old libraries (version <= 2.2) do not have always this FILL MODE param
// when fill mode is no fill (default mode).
if( *line != 0 ) arc->SetFillMode( parseFillMode( aReader, line, &line ) );
// Actual Coordinates of arc ends are read from file
if( *line != 0 ) { wxPoint arcStart, arcEnd;
arcStart.x = parseInt( aReader, line, &line ); arcStart.y = parseInt( aReader, line, &line ); arcEnd.x = parseInt( aReader, line, &line ); arcEnd.y = parseInt( aReader, line, &line );
arc->SetStart( arcStart ); arc->SetEnd( arcEnd ); } else { // Actual Coordinates of arc ends are not read from file
// (old library), calculate them
wxPoint arcStart( arc->GetRadius(), 0 ); wxPoint arcEnd( arc->GetRadius(), 0 );
RotatePoint( &arcStart.x, &arcStart.y, -angle1 ); arcStart += arc->GetPosition(); arc->SetStart( arcStart ); RotatePoint( &arcEnd.x, &arcEnd.y, -angle2 ); arcEnd += arc->GetPosition(); arc->SetEnd( arcEnd ); }
return arc;}
LIB_CIRCLE* SCH_LEGACY_PLUGIN_CACHE::loadCircle( std::unique_ptr< LIB_PART >& aPart, FILE_LINE_READER& aReader ){ const char* line = aReader.Line();
wxCHECK_MSG( strCompare( "C", line, &line ), NULL, "Invalid LIB_CIRCLE definition" );
LIB_CIRCLE* circle = new LIB_CIRCLE( aPart.get() );
wxPoint center;
center.x = parseInt( aReader, line, &line ); center.y = parseInt( aReader, line, &line );
circle->SetPosition( center ); circle->SetRadius( parseInt( aReader, line, &line ) ); circle->SetUnit( parseInt( aReader, line, &line ) ); circle->SetConvert( parseInt( aReader, line, &line ) ); circle->SetWidth( parseInt( aReader, line, &line ) );
if( *line != 0 ) circle->SetFillMode( parseFillMode( aReader, line, &line ) );
return circle;}
LIB_TEXT* SCH_LEGACY_PLUGIN_CACHE::loadText( std::unique_ptr< LIB_PART >& aPart, FILE_LINE_READER& aReader ){ const char* line = aReader.Line();
wxCHECK_MSG( strCompare( "T", line, &line ), NULL, "Invalid LIB_TEXT definition" );
LIB_TEXT* text = new LIB_TEXT( aPart.get() );
text->SetTextAngle( (double) parseInt( aReader, line, &line ) );
wxPoint center;
center.x = parseInt( aReader, line, &line ); center.y = parseInt( aReader, line, &line ); text->SetPosition( center );
wxSize size;
size.x = size.y = parseInt( aReader, line, &line ); text->SetTextSize( size ); text->SetVisible( !parseInt( aReader, line, &line ) ); text->SetUnit( parseInt( aReader, line, &line ) ); text->SetConvert( parseInt( aReader, line, &line ) );
wxString str;
// If quoted string loading fails, load as not quoted string.
if( *line == '"' ) parseQuotedString( str, aReader, line, &line ); else { parseUnquotedString( str, aReader, line, &line );
// In old libs, "spaces" are replaced by '~' in unquoted strings:
str.Replace( "~", " " ); }
if( !str.IsEmpty() ) { // convert two apostrophes back to double quote
str.Replace( "''", "\"" ); }
text->SetText( str );
// Here things are murky and not well defined. At some point it appears the format
// was changed to add text properties. However rather than add the token to the end of
// the text definition, it was added after the string and no mention if the file
// verion was bumped or not so this code make break on very old component libraries.
//
// Update: apparently even in the latest version this can be different so added a test
// for end of line before checking for the text properties.
if( LIB_VERSION( m_versionMajor, m_versionMinor ) > LIB_VERSION( 2, 0 ) && !is_eol( *line ) ) { if( strCompare( "Italic", line, &line ) ) text->SetItalic( true ); else if( !strCompare( "Normal", line, &line ) ) SCH_PARSE_ERROR( "invalid text stype, expected 'Normal' or 'Italic'", aReader, line );
if( parseInt( aReader, line, &line ) > 0 ) text->SetBold( true );
// Some old libaries version > 2.0 do not have these options for text justification:
if( !is_eol( *line ) ) { switch( parseChar( aReader, line, &line ) ) { case 'L': text->SetHorizJustify( GR_TEXT_HJUSTIFY_LEFT ); break; case 'C': text->SetHorizJustify( GR_TEXT_HJUSTIFY_CENTER ); break; case 'R': text->SetHorizJustify( GR_TEXT_HJUSTIFY_RIGHT ); break; default: SCH_PARSE_ERROR( "invalid horizontal text justication; expected L, C, or R", aReader, line ); }
switch( parseChar( aReader, line, &line ) ) { case 'T': text->SetVertJustify( GR_TEXT_VJUSTIFY_TOP ); break; case 'C': text->SetVertJustify( GR_TEXT_VJUSTIFY_CENTER ); break; case 'B': text->SetVertJustify( GR_TEXT_VJUSTIFY_BOTTOM ); break; default: SCH_PARSE_ERROR( "invalid vertical text justication; expected T, C, or B", aReader, line ); } } }
return text;}
LIB_RECTANGLE* SCH_LEGACY_PLUGIN_CACHE::loadRectangle( std::unique_ptr< LIB_PART >& aPart, FILE_LINE_READER& aReader ){ const char* line = aReader.Line();
wxCHECK_MSG( strCompare( "S", line, &line ), NULL, "Invalid LIB_RECTANGLE definition" );
LIB_RECTANGLE* rectangle = new LIB_RECTANGLE( aPart.get() );
wxPoint pos;
pos.x = parseInt( aReader, line, &line ); pos.y = parseInt( aReader, line, &line ); rectangle->SetPosition( pos );
wxPoint end;
end.x = parseInt( aReader, line, &line ); end.y = parseInt( aReader, line, &line ); rectangle->SetEnd( end );
rectangle->SetUnit( parseInt( aReader, line, &line ) ); rectangle->SetConvert( parseInt( aReader, line, &line ) ); rectangle->SetWidth( parseInt( aReader, line, &line ) );
if( *line != 0 ) rectangle->SetFillMode( parseFillMode( aReader, line, &line ) );
return rectangle;}
LIB_PIN* SCH_LEGACY_PLUGIN_CACHE::loadPin( std::unique_ptr< LIB_PART >& aPart, FILE_LINE_READER& aReader ){ const char* line = aReader.Line();
wxCHECK_MSG( strCompare( "X", line, &line ), NULL, "Invalid LIB_PIN definition" );
LIB_PIN* pin = new LIB_PIN( aPart.get() );
parseUnquotedString( pin->m_name, aReader, line, &line ); parseUnquotedString( pin->m_number, aReader, line, &line );
wxPoint pos; pos.x = parseInt( aReader, line, &line ); pos.y = parseInt( aReader, line, &line ); pin->m_position = pos;
pin->m_length = parseInt( aReader, line, &line ); pin->m_orientation = parseChar( aReader, line, &line ); pin->m_numTextSize = parseInt( aReader, line, &line ); pin->m_nameTextSize = parseInt( aReader, line, &line ); pin->m_Unit = parseInt( aReader, line, &line ); pin->m_Convert = parseInt( aReader, line, &line );
char type = parseChar( aReader, line, &line );
wxString attributes;
// Optional
parseUnquotedString( attributes, aReader, line, &line, true );
switch( type ) { case 'I': pin->m_type = PIN_INPUT; break; case 'O': pin->m_type = PIN_OUTPUT; break; case 'B': pin->m_type = PIN_BIDI; break; case 'T': pin->m_type = PIN_TRISTATE; break; case 'P': pin->m_type = PIN_PASSIVE; break; case 'U': pin->m_type = PIN_UNSPECIFIED; break; case 'W': pin->m_type = PIN_POWER_IN; break; case 'w': pin->m_type = PIN_POWER_OUT; break; case 'C': pin->m_type = PIN_OPENCOLLECTOR; break; case 'E': pin->m_type = PIN_OPENEMITTER; break; case 'N': pin->m_type = PIN_NC; break; default: SCH_PARSE_ERROR( "unknown pin type", aReader, line ); }
if( !attributes.IsEmpty() ) /* Special Symbol defined */ { enum { INVERTED = 1 << 0, CLOCK = 1 << 1, LOWLEVEL_IN = 1 << 2, LOWLEVEL_OUT = 1 << 3, FALLING_EDGE = 1 << 4, NONLOGIC = 1 << 5 };
int flags = 0;
for( int j = attributes.size(); j > 0; ) { switch( attributes[--j].GetValue() ) { case '~': break; case 'N': pin->m_attributes |= PIN_INVISIBLE; break; case 'I': flags |= INVERTED; break; case 'C': flags |= CLOCK; break; case 'L': flags |= LOWLEVEL_IN; break; case 'V': flags |= LOWLEVEL_OUT; break; case 'F': flags |= FALLING_EDGE; break; case 'X': flags |= NONLOGIC; break; default: SCH_PARSE_ERROR( "unknown pin attribute", aReader, line ); } }
switch( flags ) { case 0: pin->m_shape = PINSHAPE_LINE; break; case INVERTED: pin->m_shape = PINSHAPE_INVERTED; break; case CLOCK: pin->m_shape = PINSHAPE_CLOCK; break; case INVERTED | CLOCK: pin->m_shape = PINSHAPE_INVERTED_CLOCK; break; case LOWLEVEL_IN: pin->m_shape = PINSHAPE_INPUT_LOW; break; case LOWLEVEL_IN | CLOCK: pin->m_shape = PINSHAPE_CLOCK_LOW; break; case LOWLEVEL_OUT: pin->m_shape = PINSHAPE_OUTPUT_LOW; break; case FALLING_EDGE: pin->m_shape = PINSHAPE_FALLING_EDGE_CLOCK; break; case NONLOGIC: pin->m_shape = PINSHAPE_NONLOGIC; break; default: SCH_PARSE_ERROR( "pin attributes do not define a valid pin shape", aReader, line ); } }
return pin;}
LIB_POLYLINE* SCH_LEGACY_PLUGIN_CACHE::loadPolyLine( std::unique_ptr< LIB_PART >& aPart, FILE_LINE_READER& aReader ){ const char* line = aReader.Line();
wxCHECK_MSG( strCompare( "P", line, &line ), NULL, "Invalid LIB_POLYLINE definition" );
LIB_POLYLINE* polyLine = new LIB_POLYLINE( aPart.get() );
int points = parseInt( aReader, line, &line ); polyLine->SetUnit( parseInt( aReader, line, &line ) ); polyLine->SetConvert( parseInt( aReader, line, &line ) ); polyLine->SetWidth( parseInt( aReader, line, &line ) ); polyLine->Reserve( points );
wxPoint pt;
for( int i = 0; i < points; i++ ) { pt.x = parseInt( aReader, line, &line ); pt.y = parseInt( aReader, line, &line ); polyLine->AddPoint( pt ); }
if( *line != 0 ) polyLine->SetFillMode( parseFillMode( aReader, line, &line ) );
return polyLine;}
LIB_BEZIER* SCH_LEGACY_PLUGIN_CACHE::loadBezier( std::unique_ptr< LIB_PART >& aPart, FILE_LINE_READER& aReader ){ const char* line = aReader.Line();
wxCHECK_MSG( strCompare( "B", line, &line ), NULL, "Invalid LIB_BEZIER definition" );
LIB_BEZIER* bezier = new LIB_BEZIER( aPart.get() );
int points = parseInt( aReader, line, &line ); bezier->SetUnit( parseInt( aReader, line, &line ) ); bezier->SetConvert( parseInt( aReader, line, &line ) ); bezier->SetWidth( parseInt( aReader, line, &line ) );
wxPoint pt; bezier->Reserve( points );
for( int i = 0; i < points; i++ ) { pt.x = parseInt( aReader, line, &line ); pt.y = parseInt( aReader, line, &line ); bezier->AddPoint( pt ); }
if( *line != 0 ) bezier->SetFillMode( parseFillMode( aReader, line, &line ) );
return bezier;}
void SCH_LEGACY_PLUGIN_CACHE::loadFootprintFilters( std::unique_ptr< LIB_PART >& aPart, FILE_LINE_READER& aReader ){ const char* line = aReader.Line();
wxCHECK_RET( strCompare( "$FPLIST", line, &line ), "Invalid footprint filter list" );
line = aReader.ReadLine();
while( line ) { if( strCompare( "$ENDFPLIST", line, &line ) ) return;
wxString footprint;
parseUnquotedString( footprint, aReader, line, &line ); aPart->GetFootprints().Add( footprint ); line = aReader.ReadLine(); }
SCH_PARSE_ERROR( "file ended prematurely while loading footprint filters", aReader, line );}
void SCH_LEGACY_PLUGIN_CACHE::Save( bool aSaveDocFile ){ if( !m_isModified ) return;
// Write through symlinks, don't replace them
wxFileName fn = GetRealFile();
std::unique_ptr< FILE_OUTPUTFORMATTER > formatter( new FILE_OUTPUTFORMATTER( fn.GetFullPath() ) ); formatter->Print( 0, "%s %d.%d\n", LIBFILE_IDENT, LIB_VERSION_MAJOR, LIB_VERSION_MINOR ); formatter->Print( 0, "#encoding utf-8\n");
for( LIB_ALIAS_MAP::iterator it = m_aliases.begin(); it != m_aliases.end(); it++ ) { if( !it->second->IsRoot() ) continue;
saveSymbol( it->second->GetPart(), formatter ); }
formatter->Print( 0, "#\n#End Library\n" ); formatter.reset();
m_fileModTime = fn.GetModificationTime(); m_isModified = false;
if( aSaveDocFile ) saveDocFile();}
void SCH_LEGACY_PLUGIN_CACHE::saveSymbol( LIB_PART* aSymbol, std::unique_ptr< FILE_OUTPUTFORMATTER >& aFormatter ){ wxCHECK_RET( aSymbol, "Invalid LIB_PART pointer." );
LIB_FIELD& value = aSymbol->GetValueField();
// First line: it s a comment (component name for readers)
aFormatter->Print( 0, "#\n# %s\n#\n", TO_UTF8( value.GetText() ) );
// Save data
aFormatter->Print( 0, "DEF" );
if( value.IsVisible() ) { aFormatter->Print( 0, " %s", TO_UTF8( value.GetText() ) ); } else { aFormatter->Print( 0, " ~%s", TO_UTF8( value.GetText() ) ); }
LIB_FIELD& reference = aSymbol->GetReferenceField();
if( !reference.GetText().IsEmpty() ) { aFormatter->Print( 0, " %s", TO_UTF8( reference.GetText() ) ); } else { aFormatter->Print( 0, " ~" ); }
aFormatter->Print( 0, " %d %d %c %c %d %c %c\n", 0, aSymbol->GetPinNameOffset(), aSymbol->ShowPinNumbers() ? 'Y' : 'N', aSymbol->ShowPinNames() ? 'Y' : 'N', aSymbol->GetUnitCount(), aSymbol->UnitsLocked() ? 'L' : 'F', aSymbol->IsPower() ? 'P' : 'N' );
timestamp_t dateModified = aSymbol->GetDateLastEdition();
if( dateModified != 0 ) { int sec = dateModified & 63; int min = ( dateModified >> 6 ) & 63; int hour = ( dateModified >> 12 ) & 31; int day = ( dateModified >> 17 ) & 31; int mon = ( dateModified >> 22 ) & 15; int year = ( dateModified >> 26 ) + 1990;
aFormatter->Print( 0, "Ti %d/%d/%d %d:%d:%d\n", year, mon, day, hour, min, sec ); }
LIB_FIELDS fields; aSymbol->GetFields( fields );
// Mandatory fields:
// may have their own save policy so there is a separate loop for them.
// Empty fields are saved, because the user may have set visibility,
// size and orientation
for( int i = 0; i < MANDATORY_FIELDS; ++i ) { saveField( &fields[i], aFormatter ); }
// User defined fields:
// may have their own save policy so there is a separate loop for them.
int fieldId = MANDATORY_FIELDS; // really wish this would go away.
for( unsigned i = MANDATORY_FIELDS; i < fields.size(); ++i ) { // There is no need to save empty fields, i.e. no reason to preserve field
// names now that fields names come in dynamically through the template
// fieldnames.
if( !fields[i].GetText().IsEmpty() ) { fields[i].SetId( fieldId++ ); saveField( &fields[i], aFormatter ); } }
// Save the alias list: a line starting by "ALIAS". The first alias is the root
// and has the same name as the component. In the old library file format this
// alias does not get added to the alias list.
if( aSymbol->GetAliasCount() > 1 ) { wxArrayString aliases = aSymbol->GetAliasNames();
aFormatter->Print( 0, "ALIAS" );
for( unsigned i = 1; i < aliases.size(); i++ ) { aFormatter->Print( 0, " %s", TO_UTF8( aliases[i] ) ); }
aFormatter->Print( 0, "\n" ); }
wxArrayString footprints = aSymbol->GetFootprints();
// Write the footprint filter list
if( footprints.GetCount() != 0 ) { aFormatter->Print( 0, "$FPLIST\n" );
for( unsigned i = 0; i < footprints.GetCount(); i++ ) { aFormatter->Print( 0, " %s\n", TO_UTF8( footprints[i] ) ); }
aFormatter->Print( 0, "$ENDFPLIST\n" ); }
// Save graphics items (including pins)
if( !aSymbol->GetDrawItems().empty() ) { // Sort the draw items in order to editing a file editing by hand.
aSymbol->GetDrawItems().sort();
aFormatter->Print( 0, "DRAW\n" );
for( LIB_ITEM& item : aSymbol->GetDrawItems() ) { switch( item.Type() ) { case LIB_FIELD_T: // Fields have already been saved above.
continue;
case LIB_ARC_T: saveArc( (LIB_ARC*) &item, aFormatter ); break;
case LIB_BEZIER_T: saveBezier( (LIB_BEZIER*) &item, aFormatter ); break;
case LIB_CIRCLE_T: saveCircle( ( LIB_CIRCLE* ) &item, aFormatter ); break;
case LIB_PIN_T: savePin( (LIB_PIN* ) &item, aFormatter ); break;
case LIB_POLYLINE_T: savePolyLine( ( LIB_POLYLINE* ) &item, aFormatter ); break;
case LIB_RECTANGLE_T: saveRectangle( ( LIB_RECTANGLE* ) &item, aFormatter ); break;
case LIB_TEXT_T: saveText( ( LIB_TEXT* ) &item, aFormatter ); break;
default: ; } }
aFormatter->Print( 0, "ENDDRAW\n" ); }
aFormatter->Print( 0, "ENDDEF\n" );}
void SCH_LEGACY_PLUGIN_CACHE::saveArc( LIB_ARC* aArc, std::unique_ptr< FILE_OUTPUTFORMATTER >& aFormatter ){ wxCHECK_RET( aArc && aArc->Type() == LIB_ARC_T, "Invalid LIB_ARC object." );
int x1 = aArc->GetFirstRadiusAngle();
if( x1 > 1800 ) x1 -= 3600;
int x2 = aArc->GetSecondRadiusAngle();
if( x2 > 1800 ) x2 -= 3600;
aFormatter->Print( 0, "A %d %d %d %d %d %d %d %d %c %d %d %d %d\n", aArc->GetPosition().x, aArc->GetPosition().y, aArc->GetRadius(), x1, x2, aArc->GetUnit(), aArc->GetConvert(), aArc->GetWidth(), fill_tab[aArc->GetFillMode()], aArc->GetStart().x, aArc->GetStart().y, aArc->GetEnd().x, aArc->GetEnd().y );}
void SCH_LEGACY_PLUGIN_CACHE::saveBezier( LIB_BEZIER* aBezier, std::unique_ptr< FILE_OUTPUTFORMATTER >& aFormatter ){ wxCHECK_RET( aBezier && aBezier->Type() == LIB_BEZIER_T, "Invalid LIB_BEZIER object." );
aFormatter->Print( 0, "B %u %d %d %d", (unsigned)aBezier->GetPoints().size(), aBezier->GetUnit(), aBezier->GetConvert(), aBezier->GetWidth() );
for( const auto& pt : aBezier->GetPoints() ) aFormatter->Print( 0, " %d %d", pt.x, pt.y );
aFormatter->Print( 0, " %c\n", fill_tab[aBezier->GetFillMode()] );}
void SCH_LEGACY_PLUGIN_CACHE::saveCircle( LIB_CIRCLE* aCircle, std::unique_ptr< FILE_OUTPUTFORMATTER >& aFormatter ){ wxCHECK_RET( aCircle && aCircle->Type() == LIB_CIRCLE_T, "Invalid LIB_CIRCLE object." );
aFormatter->Print( 0, "C %d %d %d %d %d %d %c\n", aCircle->GetPosition().x, aCircle->GetPosition().y, aCircle->GetRadius(), aCircle->GetUnit(), aCircle->GetConvert(), aCircle->GetWidth(), fill_tab[aCircle->GetFillMode()] );}
void SCH_LEGACY_PLUGIN_CACHE::saveField( LIB_FIELD* aField, std::unique_ptr< FILE_OUTPUTFORMATTER >& aFormatter ){ wxCHECK_RET( aField && aField->Type() == LIB_FIELD_T, "Invalid LIB_FIELD object." );
int hjustify, vjustify; int id = aField->GetId(); wxString text = aField->m_Text;
hjustify = 'C';
if( aField->GetHorizJustify() == GR_TEXT_HJUSTIFY_LEFT ) hjustify = 'L'; else if( aField->GetHorizJustify() == GR_TEXT_HJUSTIFY_RIGHT ) hjustify = 'R';
vjustify = 'C';
if( aField->GetVertJustify() == GR_TEXT_VJUSTIFY_BOTTOM ) vjustify = 'B'; else if( aField->GetVertJustify() == GR_TEXT_VJUSTIFY_TOP ) vjustify = 'T';
aFormatter->Print( 0, "F%d %s %d %d %d %c %c %c %c%c%c", id, EscapedUTF8( text ).c_str(), // wraps in quotes
aField->GetTextPos().x, aField->GetTextPos().y, aField->GetTextWidth(), aField->GetTextAngle() == 0 ? 'H' : 'V', aField->IsVisible() ? 'V' : 'I', hjustify, vjustify, aField->IsItalic() ? 'I' : 'N', aField->IsBold() ? 'B' : 'N' );
/* Save field name, if necessary
* Field name is saved only if it is not the default name. * Just because default name depends on the language and can change from * a country to another */ wxString defName = TEMPLATE_FIELDNAME::GetDefaultFieldName( id );
if( id >= FIELD1 && !aField->m_name.IsEmpty() && aField->m_name != defName ) aFormatter->Print( 0, " %s", EscapedUTF8( aField->m_name ).c_str() );
aFormatter->Print( 0, "\n" );}
void SCH_LEGACY_PLUGIN_CACHE::savePin( LIB_PIN* aPin, std::unique_ptr< FILE_OUTPUTFORMATTER >& aFormatter ){ wxCHECK_RET( aPin && aPin->Type() == LIB_PIN_T, "Invalid LIB_PIN object." );
int Etype;
switch( aPin->GetType() ) { default: case PIN_INPUT: Etype = 'I'; break;
case PIN_OUTPUT: Etype = 'O'; break;
case PIN_BIDI: Etype = 'B'; break;
case PIN_TRISTATE: Etype = 'T'; break;
case PIN_PASSIVE: Etype = 'P'; break;
case PIN_UNSPECIFIED: Etype = 'U'; break;
case PIN_POWER_IN: Etype = 'W'; break;
case PIN_POWER_OUT: Etype = 'w'; break;
case PIN_OPENCOLLECTOR: Etype = 'C'; break;
case PIN_OPENEMITTER: Etype = 'E'; break;
case PIN_NC: Etype = 'N'; break; }
if( !aPin->GetName().IsEmpty() ) aFormatter->Print( 0, "X %s", TO_UTF8( aPin->GetName() ) ); else aFormatter->Print( 0, "X ~" );
aFormatter->Print( 0, " %s %d %d %d %c %d %d %d %d %c", aPin->GetNumber().IsEmpty() ? "~" : TO_UTF8( aPin->GetNumber() ), aPin->GetPosition().x, aPin->GetPosition().y, (int) aPin->GetLength(), (int) aPin->GetOrientation(), aPin->GetNumberTextSize(), aPin->GetNameTextSize(), aPin->GetUnit(), aPin->GetConvert(), Etype );
if( aPin->GetShape() || !aPin->IsVisible() ) aFormatter->Print( 0, " " );
if( !aPin->IsVisible() ) aFormatter->Print( 0, "N" );
switch( aPin->GetShape() ) { case PINSHAPE_LINE: break;
case PINSHAPE_INVERTED: aFormatter->Print( 0, "I" ); break;
case PINSHAPE_CLOCK: aFormatter->Print( 0, "C" ); break;
case PINSHAPE_INVERTED_CLOCK: aFormatter->Print( 0, "IC" ); break;
case PINSHAPE_INPUT_LOW: aFormatter->Print( 0, "L" ); break;
case PINSHAPE_CLOCK_LOW: aFormatter->Print( 0, "CL" ); break;
case PINSHAPE_OUTPUT_LOW: aFormatter->Print( 0, "V" ); break;
case PINSHAPE_FALLING_EDGE_CLOCK: aFormatter->Print( 0, "F" ); break;
case PINSHAPE_NONLOGIC: aFormatter->Print( 0, "X" ); break;
default: assert( !"Invalid pin shape" ); }
aFormatter->Print( 0, "\n" );
aPin->ClearFlags( IS_CHANGED );}
void SCH_LEGACY_PLUGIN_CACHE::savePolyLine( LIB_POLYLINE* aPolyLine, std::unique_ptr< FILE_OUTPUTFORMATTER >& aFormatter ){ wxCHECK_RET( aPolyLine && aPolyLine->Type() == LIB_POLYLINE_T, "Invalid LIB_POLYLINE object." );
int ccount = aPolyLine->GetCornerCount();
aFormatter->Print( 0, "P %d %d %d %d", ccount, aPolyLine->GetUnit(), aPolyLine->GetConvert(), aPolyLine->GetWidth() );
for( const auto& pt : aPolyLine->GetPolyPoints() ) { aFormatter->Print( 0, " %d %d", pt.x, pt.y ); }
aFormatter->Print( 0, " %c\n", fill_tab[aPolyLine->GetFillMode()] );}
void SCH_LEGACY_PLUGIN_CACHE::saveRectangle( LIB_RECTANGLE* aRectangle, std::unique_ptr< FILE_OUTPUTFORMATTER >& aFormatter ){ wxCHECK_RET( aRectangle && aRectangle->Type() == LIB_RECTANGLE_T, "Invalid LIB_RECTANGLE object." );
aFormatter->Print( 0, "S %d %d %d %d %d %d %d %c\n", aRectangle->GetPosition().x, aRectangle->GetPosition().y, aRectangle->GetEnd().x, aRectangle->GetEnd().y, aRectangle->GetUnit(), aRectangle->GetConvert(), aRectangle->GetWidth(), fill_tab[aRectangle->GetFillMode()] );}
void SCH_LEGACY_PLUGIN_CACHE::saveText( LIB_TEXT* aText, std::unique_ptr< FILE_OUTPUTFORMATTER >& aFormatter ){ wxCHECK_RET( aText && aText->Type() == LIB_TEXT_T, "Invalid LIB_TEXT object." );
wxString text = aText->GetText();
if( text.Contains( wxT( " " ) ) || text.Contains( wxT( "~" ) ) || text.Contains( wxT( "\"" ) ) ) { // convert double quote to similar-looking two apostrophes
text.Replace( wxT( "\"" ), wxT( "''" ) ); text = wxT( "\"" ) + text + wxT( "\"" ); }
aFormatter->Print( 0, "T %g %d %d %d %d %d %d %s", aText->GetTextAngle(), aText->GetTextPos().x, aText->GetTextPos().y, aText->GetTextWidth(), !aText->IsVisible(), aText->GetUnit(), aText->GetConvert(), TO_UTF8( text ) );
aFormatter->Print( 0, " %s %d", aText->IsItalic() ? "Italic" : "Normal", aText->IsBold() );
char hjustify = 'C';
if( aText->GetHorizJustify() == GR_TEXT_HJUSTIFY_LEFT ) hjustify = 'L'; else if( aText->GetHorizJustify() == GR_TEXT_HJUSTIFY_RIGHT ) hjustify = 'R';
char vjustify = 'C';
if( aText->GetVertJustify() == GR_TEXT_VJUSTIFY_BOTTOM ) vjustify = 'B'; else if( aText->GetVertJustify() == GR_TEXT_VJUSTIFY_TOP ) vjustify = 'T';
aFormatter->Print( 0, " %c %c\n", hjustify, vjustify );}
void SCH_LEGACY_PLUGIN_CACHE::saveDocFile(){ wxFileName fileName = m_libFileName;
fileName.SetExt( DOC_EXT ); FILE_OUTPUTFORMATTER formatter( fileName.GetFullPath() );
formatter.Print( 0, "%s\n", DOCFILE_IDENT );
for( LIB_ALIAS_MAP::iterator it = m_aliases.begin(); it != m_aliases.end(); it++ ) { wxString description = it->second->GetDescription(); wxString keyWords = it->second->GetKeyWords(); wxString docFileName = it->second->GetDocFileName();
if( description.IsEmpty() && keyWords.IsEmpty() && docFileName.IsEmpty() ) continue;
formatter.Print( 0, "#\n$CMP %s\n", TO_UTF8( it->second->GetName() ) );
if( !description.IsEmpty() ) formatter.Print( 0, "D %s\n", TO_UTF8( description ) );
if( !keyWords.IsEmpty() ) formatter.Print( 0, "K %s\n", TO_UTF8( keyWords ) );
if( !docFileName.IsEmpty() ) formatter.Print( 0, "F %s\n", TO_UTF8( docFileName ) );
formatter.Print( 0, "$ENDCMP\n" ); }
formatter.Print( 0, "#\n#End Doc Library\n" );}
void SCH_LEGACY_PLUGIN_CACHE::DeleteAlias( const wxString& aAliasName ){ LIB_ALIAS_MAP::iterator it = m_aliases.find( aAliasName );
if( it == m_aliases.end() ) THROW_IO_ERROR( wxString::Format( _( "library %s does not contain an alias %s" ), m_libFileName.GetFullName(), aAliasName ) );
LIB_ALIAS* alias = it->second; LIB_PART* part = alias->GetPart();
alias = part->RemoveAlias( alias );
if( !alias ) { delete part;
if( m_aliases.size() > 1 ) { LIB_ALIAS_MAP::iterator next = it; next++;
if( next == m_aliases.end() ) next = m_aliases.begin();
alias = next->second; } }
m_aliases.erase( it ); ++m_modHash; m_isModified = true;}
void SCH_LEGACY_PLUGIN_CACHE::DeleteSymbol( const wxString& aAliasName ){ LIB_ALIAS_MAP::iterator it = m_aliases.find( aAliasName );
if( it == m_aliases.end() ) THROW_IO_ERROR( wxString::Format( _( "library %s does not contain an alias %s" ), m_libFileName.GetFullName(), aAliasName ) );
LIB_ALIAS* alias = it->second; LIB_PART* part = alias->GetPart();
wxArrayString aliasNames = part->GetAliasNames();
// Deleting all of the aliases deletes the symbol from the library.
for( size_t i = 0; i < aliasNames.Count(); i++ ) DeleteAlias( aliasNames[i] );}
void SCH_LEGACY_PLUGIN::cacheLib( const wxString& aLibraryFileName ){ if( !m_cache || !m_cache->IsFile( aLibraryFileName ) || m_cache->IsFileChanged() ) { // a spectacular episode in memory management:
delete m_cache; m_cache = new SCH_LEGACY_PLUGIN_CACHE( aLibraryFileName );
// Because m_cache is rebuilt, increment PART_LIBS::s_modify_generation
// to modify the hash value that indicate component to symbol links
// must be updated.
PART_LIBS::s_modify_generation++;
if( !isBuffering( m_props ) ) m_cache->Load(); }}
bool SCH_LEGACY_PLUGIN::writeDocFile( const PROPERTIES* aProperties ){ std::string propName( SCH_LEGACY_PLUGIN::PropNoDocFile );
if( aProperties && aProperties->find( propName ) != aProperties->end() ) return false;
return true;}
bool SCH_LEGACY_PLUGIN::isBuffering( const PROPERTIES* aProperties ){ return ( aProperties && aProperties->Exists( SCH_LEGACY_PLUGIN::PropBuffering ) );}
int SCH_LEGACY_PLUGIN::GetModifyHash() const{ if( m_cache ) return m_cache->GetModifyHash();
// If the cache hasn't been loaded, it hasn't been modified.
return 0;}
size_t SCH_LEGACY_PLUGIN::GetSymbolLibCount( const wxString& aLibraryPath, const PROPERTIES* aProperties ){ LOCALE_IO toggle;
m_props = aProperties;
cacheLib( aLibraryPath );
return m_cache->m_aliases.size();}
void SCH_LEGACY_PLUGIN::EnumerateSymbolLib( wxArrayString& aAliasNameList, const wxString& aLibraryPath, const PROPERTIES* aProperties ){ LOCALE_IO toggle; // toggles on, then off, the C locale.
m_props = aProperties;
bool powerSymbolsOnly = ( aProperties && aProperties->find( SYMBOL_LIB_TABLE::PropPowerSymsOnly ) != aProperties->end() ); cacheLib( aLibraryPath );
const LIB_ALIAS_MAP& aliases = m_cache->m_aliases;
for( LIB_ALIAS_MAP::const_iterator it = aliases.begin(); it != aliases.end(); ++it ) { if( !powerSymbolsOnly || it->second->GetPart()->IsPower() ) aAliasNameList.Add( it->first ); }}
void SCH_LEGACY_PLUGIN::EnumerateSymbolLib( std::vector<LIB_ALIAS*>& aAliasList, const wxString& aLibraryPath, const PROPERTIES* aProperties ){ LOCALE_IO toggle; // toggles on, then off, the C locale.
m_props = aProperties;
bool powerSymbolsOnly = ( aProperties && aProperties->find( SYMBOL_LIB_TABLE::PropPowerSymsOnly ) != aProperties->end() ); cacheLib( aLibraryPath );
const LIB_ALIAS_MAP& aliases = m_cache->m_aliases;
for( LIB_ALIAS_MAP::const_iterator it = aliases.begin(); it != aliases.end(); ++it ) { if( !powerSymbolsOnly || it->second->GetPart()->IsPower() ) aAliasList.push_back( it->second ); }}
LIB_ALIAS* SCH_LEGACY_PLUGIN::LoadSymbol( const wxString& aLibraryPath, const wxString& aAliasName, const PROPERTIES* aProperties ){ LOCALE_IO toggle; // toggles on, then off, the C locale.
m_props = aProperties;
cacheLib( aLibraryPath );
LIB_ALIAS_MAP::const_iterator it = m_cache->m_aliases.find( aAliasName );
if( it == m_cache->m_aliases.end() ) return NULL;
return it->second;}
void SCH_LEGACY_PLUGIN::SaveSymbol( const wxString& aLibraryPath, const LIB_PART* aSymbol, const PROPERTIES* aProperties ){ m_props = aProperties;
cacheLib( aLibraryPath );
m_cache->AddSymbol( aSymbol );
if( !isBuffering( aProperties ) ) m_cache->Save( writeDocFile( aProperties ) );}
void SCH_LEGACY_PLUGIN::DeleteAlias( const wxString& aLibraryPath, const wxString& aAliasName, const PROPERTIES* aProperties ){ m_props = aProperties;
cacheLib( aLibraryPath );
m_cache->DeleteAlias( aAliasName );
if( !isBuffering( aProperties ) ) m_cache->Save( writeDocFile( aProperties ) );}
void SCH_LEGACY_PLUGIN::DeleteSymbol( const wxString& aLibraryPath, const wxString& aAliasName, const PROPERTIES* aProperties ){ m_props = aProperties;
cacheLib( aLibraryPath );
m_cache->DeleteSymbol( aAliasName );
if( !isBuffering( aProperties ) ) m_cache->Save( writeDocFile( aProperties ) );}
void SCH_LEGACY_PLUGIN::CreateSymbolLib( const wxString& aLibraryPath, const PROPERTIES* aProperties ){ if( wxFileExists( aLibraryPath ) ) { THROW_IO_ERROR( wxString::Format( _( "symbol library \"%s\" already exists, cannot create a new library" ), aLibraryPath.GetData() ) ); }
LOCALE_IO toggle;
m_props = aProperties;
delete m_cache; m_cache = new SCH_LEGACY_PLUGIN_CACHE( aLibraryPath ); m_cache->SetModified(); m_cache->Save( writeDocFile( aProperties ) ); m_cache->Load(); // update m_writable and m_mod_time
}
bool SCH_LEGACY_PLUGIN::DeleteSymbolLib( const wxString& aLibraryPath, const PROPERTIES* aProperties ){ wxFileName fn = aLibraryPath;
if( !fn.FileExists() ) return false;
// Some of the more elaborate wxRemoveFile() crap puts up its own wxLog dialog
// we don't want that. we want bare metal portability with no UI here.
if( wxRemove( aLibraryPath ) ) { THROW_IO_ERROR( wxString::Format( _( "library \"%s\" cannot be deleted" ), aLibraryPath.GetData() ) ); }
if( m_cache && m_cache->IsFile( aLibraryPath ) ) { delete m_cache; m_cache = 0; }
return true;}
void SCH_LEGACY_PLUGIN::SaveLibrary( const wxString& aLibraryPath, const PROPERTIES* aProperties ){ if( !m_cache ) m_cache = new SCH_LEGACY_PLUGIN_CACHE( aLibraryPath );
wxString oldFileName = m_cache->GetFileName();
if( !m_cache->IsFile( aLibraryPath ) ) { m_cache->SetFileName( aLibraryPath ); }
// This is a forced save.
m_cache->SetModified(); m_cache->Save( writeDocFile( aProperties ) ); m_cache->SetFileName( oldFileName );}
bool SCH_LEGACY_PLUGIN::CheckHeader( const wxString& aFileName ){ // Open file and check first line
wxTextFile tempFile;
tempFile.Open( aFileName ); wxString firstline; // read the first line
firstline = tempFile.GetFirstLine(); tempFile.Close();
return firstline.StartsWith( "EESchema" );}
bool SCH_LEGACY_PLUGIN::IsSymbolLibWritable( const wxString& aLibraryPath ){ return wxFileName::IsFileWritable( aLibraryPath );}
const char* SCH_LEGACY_PLUGIN::PropBuffering = "buffering";const char* SCH_LEGACY_PLUGIN::PropNoDocFile = "no_doc_file";
|