Commit ed8c021f authored by Dick Hollenbeck's avatar Dick Hollenbeck

++all:

  * TokenList2DsnLexer.cmake now supports comments, which start with a leading
    # character, and may be either on their own line or on a line after a token.
  * DSNLEXER::PopReader() now pops even the last LINE_READER* and returns it.
++pcbnew:
  * SPECCTRA_DB now inherits from new class SPECCTRA_LEXER, which led to a great
    deal of simplification and code factoring.
  * Moved specctra keywords into specctra.keywords.
parent 42b9e0e6
......@@ -17,6 +17,8 @@ install_manifest.txt
Documentation/doxygen
*.cmake
*.bak
pcbnew/specctra_keywords.cpp
pcbnew/specctra_lexer.h
new/html
new/sch_lib_table_keywords.cpp
new/sch_lib_table_lexer.h
......
......@@ -4,6 +4,17 @@ KiCad ChangeLog 2010
Please add newer entries at the top, list the date and your name with
email address.
2011-Jan-19 UPDATE Dick Hollenbeck <dick@softplc.com>
================================================================================
++all:
* TokenList2DsnLexer.cmake now supports comments, which start with a leading
# character, and may be either on their own line or on a line after a token.
* DSNLEXER::PopReader() now pops even the last LINE_READER* and returns it.
++pcbnew:
* SPECCTRA_DB now inherits from new class SPECCTRA_LEXER, which led to a great
deal of simplification and code factoring.
* Moved specctra keywords into specctra.keywords.
2011-Jan-17 UPDATE Dick Hollenbeck <dick@softplc.com>
================================================================================
++all:
......
#
# This program source code file is part of KICAD, a free EDA CAD application.
#
# Copyright (C) 2010 Wayne Stambaugh <stambaughw@verizon.net>
......@@ -160,26 +160,33 @@ const KEYWORD ${LEXERCLASS}::keywords[] = {
"
)
file( STRINGS ${inputFile} tmpTokens NO_HEX_CONVERSION )
file( STRINGS ${inputFile} lines NO_HEX_CONVERSION )
foreach( tmpToken ${tmpTokens} )
foreach( line ${lines} )
math( EXPR lineCount "${lineCount} + 1" )
string( STRIP tmpToken "${tmpToken}" )
# strip any comment from # to end of line
string( REGEX REPLACE "#.*$" "" tmpToken "${line}" )
string( STRIP "${tmpToken}" token )
# Ignore empty lines.
if( tmpToken )
if( NOT token STREQUAL "" ) # if token is "off" simple if( token) does not work
# Make sure token is valid.
string( REGEX MATCH "[a-z][_0-9a-z]*[0-9a-z]$" validToken "${tmpToken}" )
if( validToken STREQUAL tmpToken )
#message( "token=${token}" )
string( REGEX MATCH "[a-z][_0-9a-z]*" validToken "${token}" )
#message( "validToken=${validToken}" )
if( validToken STREQUAL token )
list( APPEND tokens "${validToken}" )
else( validToken STREQUAL tmpToken )
else()
message( FATAL_ERROR
"Invalid token string \"${tmpToken}\" at line ${lineCount} in file "
"<${inputFile}>." )
endif( validToken STREQUAL tmpToken )
endif( tmpToken )
endforeach( tmpToken ${tmpTokens} )
endif()
endif()
endforeach()
list( SORT tokens )
......@@ -190,7 +197,7 @@ list( LENGTH tokens tokensAfter )
if( NOT ( tokensBefore EQUAL tokensAfter ) )
message( FATAL_ERROR "Duplicate tokens found in file <${inputFile}>." )
endif( NOT ( tokensBefore EQUAL tokensAfter ) )
endif()
file( WRITE "${outHeaderFile}" "${includeFileHeader}" )
file( WRITE "${outCppFile}" "${sourceFileHeader}" )
......@@ -214,7 +221,7 @@ foreach( token ${tokens} )
file( APPEND "${outCppFile}" ",\n" )
endif( lineCount EQUAL tokensAfter )
math( EXPR lineCount "${lineCount} + 1" )
endforeach( token ${tokens} )
endforeach()
file( APPEND "${outHeaderFile}"
" };
......@@ -275,6 +282,12 @@ public:
{
}
/**
* Function TokenName
* returns the name of the token in ASCII form.
*/
static const char* TokenName( ${enum}::T aTok );
/**
* Function NextTok
* returns the next token found in the input file or T_EOF when reaching
......@@ -356,5 +369,23 @@ file( APPEND "${outCppFile}"
const unsigned ${LEXERCLASS}::keyword_count = unsigned( sizeof( ${LEXERCLASS}::keywords )/sizeof( ${LEXERCLASS}::keywords[0] ) );
const char* ${LEXERCLASS}::TokenName( T aTok )
{
const char* ret;
if( (unsigned) aTok < keyword_count )
{
ret = keywords[aTok].name;
}
else if( aTok < 0 )
{
return DSNLEXER::Syntax( aTok );
}
else
ret = \"token too big\";
return ret;
}
"
)
......@@ -91,6 +91,7 @@ DSNLEXER::DSNLEXER( const KEYWORD* aKeywordTable, unsigned aKeywordCount,
keywords( aKeywordTable ),
keywordCount( aKeywordCount )
{
if( aLineReader )
PushReader( aLineReader );
init();
}
......@@ -119,22 +120,33 @@ void DSNLEXER::PushReader( LINE_READER* aLineReader )
}
bool DSNLEXER::PopReader()
LINE_READER* DSNLEXER::PopReader()
{
// the very last reader cannot be popped, for that would screw up limit and next.
if( readerStack.size() >= 2 )
LINE_READER* ret = 0;
if( readerStack.size() )
{
ret = reader;
readerStack.pop_back();
if( readerStack.size() )
{
reader = readerStack.back();
start = (const char*) (*reader);
start = reader->Line();
// force a new readLine() as first thing.
limit = start;
next = start;
return true;
}
return false;
else
{
reader = 0;
start = dummy;
limit = dummy;
limit = dummy;
}
}
return ret;
}
......@@ -198,7 +210,7 @@ const char* DSNLEXER::Syntax( int aTok )
ret = "quoted string";
break;
case DSN_EOF:
ret = "end of file";
ret = "end of input";
break;
default:
ret = "???";
......@@ -257,10 +269,10 @@ void DSNLEXER::Expecting( int aTok ) throw( IO_ERROR )
}
void DSNLEXER::Expecting( const wxString& text ) throw( IO_ERROR )
void DSNLEXER::Expecting( const char* text ) throw( IO_ERROR )
{
wxString errText( _("Expecting") );
errText << wxT(" '") << text << wxT("'");
errText << wxT(" '") << wxString::FromUTF8( text ) << wxT("'");
THROW_PARSE_ERROR( errText, CurSource(), CurLine(), CurLineNumber(), CurOffset() );
}
......@@ -272,6 +284,7 @@ void DSNLEXER::Unexpected( int aTok ) throw( IO_ERROR )
THROW_PARSE_ERROR( errText, CurSource(), CurLine(), CurLineNumber(), CurOffset() );
}
void DSNLEXER::Duplicate( int aTok ) throw( IO_ERROR )
{
wxString errText;
......@@ -280,10 +293,11 @@ void DSNLEXER::Duplicate( int aTok ) throw( IO_ERROR )
THROW_PARSE_ERROR( errText, CurSource(), CurLine(), CurLineNumber(), CurOffset() );
}
void DSNLEXER::Unexpected( const wxString& text ) throw( IO_ERROR )
void DSNLEXER::Unexpected( const char* text ) throw( IO_ERROR )
{
wxString errText( _("Unexpected") );
errText << wxT(" '") << text << wxT("'");
errText << wxT(" '") << wxString::FromUTF8( text ) << wxT("'");
THROW_PARSE_ERROR( errText, CurSource(), CurLine(), CurLineNumber(), CurOffset() );
}
......@@ -317,7 +331,7 @@ int DSNLEXER::NeedSYMBOLorNUMBER() throw( IO_ERROR )
{
int tok = NextTok();
if( !IsSymbol( tok ) && tok!=DSN_NUMBER )
Expecting( _("symbol|number") );
Expecting( "symbol|number" );
return tok;
}
......
......@@ -79,7 +79,7 @@ void TEMPLATE_FIELDNAME::Parse( TEMPLATE_FIELDNAMES_LEXER* in ) throw( IO_ERROR
break;
default:
in->Expecting( wxT( "value|visible" ) );
in->Expecting( "value|visible" );
break;
}
}
......@@ -129,7 +129,7 @@ void TEMPLATES::Parse( TEMPLATE_FIELDNAMES_LEXER* in ) throw( IO_ERROR )
break;
default:
in->Unexpected( CONV_FROM_UTF8( in->CurText() ) );
in->Unexpected( in->CurText() );
break;
}
}
......
......@@ -76,10 +76,12 @@ enum DSN_SYNTAX_T {
*/
class DSNLEXER
{
protected:
bool iOwnReaders; ///< on readerStack, should I delete them?
const char* start;
const char* next;
const char* limit;
char dummy[1]; ///< when there is no reader.
typedef std::vector<LINE_READER*> READER_STACK;
......@@ -102,18 +104,22 @@ class DSNLEXER
void init();
int readLine() throw( IO_ERROR )
{
if( reader )
{
unsigned len = reader->ReadLine();
// start may have changed in ReadLine(), which can resize and
// relocate reader's line buffer.
start = (*reader);
start = reader->Line();
next = start;
limit = next + len;
return len;
}
return 0;
}
/**
......@@ -220,10 +226,11 @@ public:
* possible if there are at least 2 LINE_READERs on the stack, since popping
* the last one is not supported.
*
* @return bool - true if there was at least two readers on the stack and
* therefore the pop succeeded, else false and the pop failed.
* @return LINE_READER* - is the one that was in use before the pop, or NULL
* if there was not at least two readers on the stack and therefore the
* pop failed.
*/
bool PopReader();
LINE_READER* PopReader();
// Some functions whose return value is best overloaded to return an enum
// in a derived class.
......@@ -341,11 +348,11 @@ public:
/**
* Function Expecting
* throws an IO_ERROR exception with an input file specific error message.
* @param aErrorMsg is the token/keyword type which was expected at the
* current input location.
* @param aTokenList is the token/keyword type which was expected at the
* current input location, e.g. "pin|graphic|property"
* @throw IO_ERROR with the location within the input file of the problem.
*/
void Expecting( const wxString& aErrorMsg ) throw( IO_ERROR );
void Expecting( const char* aTokenList ) throw( IO_ERROR );
/**
* Function Unexpected
......@@ -369,11 +376,11 @@ public:
/**
* Function Unexpected
* throws an IO_ERROR exception with an input file specific error message.
* @param aErrorMsg is the token/keyword type which was not expected at the
* @param aToken is the token which was not expected at the
* current input location.
* @throw IO_ERROR with the location within the input file of the problem.
*/
void Unexpected( const wxString& aErrorMsg ) throw( IO_ERROR );
void Unexpected( const char* aToken ) throw( IO_ERROR );
/**
* Function NeedLEFT
......
......@@ -171,6 +171,7 @@ set(PCBNEW_SRCS
specctra.cpp
specctra_export.cpp
specctra_import.cpp
specctra_keywords.cpp
surbrill.cpp
swap_layers.cpp
tool_modedit.cpp
......@@ -245,6 +246,16 @@ set_source_files_properties( dialogs/dialog_freeroute_exchange.cpp
OBJECT_DEPENDS ${CMAKE_CURRENT_SOURCE_DIR}/dialogs/dialog_freeroute_exchange_help_html.h
)
# auto-generate specctra_lexer.h and specctra_keywords.cpp
make_lexer(
${CMAKE_CURRENT_SOURCE_DIR}/specctra.keywords
${CMAKE_CURRENT_SOURCE_DIR}/specctra_lexer.h
${CMAKE_CURRENT_SOURCE_DIR}/specctra_keywords.cpp
DSN
)
###
# Create the pcbnew executable
###
......
This source diff could not be displayed because it is too large. You can view the blob instead.
......@@ -34,13 +34,15 @@
#include "fctsys.h"
#include "dsnlexer.h"
#include "specctra_lexer.h"
#include "pcbnew.h"
class TYPE_COLLECTOR; // outside the DSN namespace
typedef DSN::T DSN_T;
using namespace DSN;
/**
......@@ -68,419 +70,6 @@ class TYPE_COLLECTOR; // outside the DSN namespace
namespace DSN {
enum DSN_T {
// these first few are negative special ones for syntax, and are
// inherited from DSNLEXER.
T_NONE = DSN_NONE,
T_COMMENT = DSN_COMMENT,
T_STRING_QUOTE = DSN_STRING_QUOTE,
T_QUOTE_DEF = DSN_QUOTE_DEF,
T_DASH = DSN_DASH,
T_SYMBOL = DSN_SYMBOL,
T_NUMBER = DSN_NUMBER,
T_RIGHT = DSN_RIGHT, // right bracket, ')'
T_LEFT = DSN_LEFT, // left bracket, '('
T_STRING = DSN_STRING, // a quoted string, stripped of the quotes
T_EOF = DSN_EOF, // special case for end of file
// This should be coordinated with the
// const static KEYWORD tokens[] array, and both must be sorted
// identically and alphabetically. Remember that '_' is less than any
// alpha character according to ASCII.
T_absolute = 0, // this one should be == zero
T_added,
T_add_group,
T_add_pins,
T_allow_antenna,
T_allow_redundant_wiring,
T_amp,
T_ancestor,
T_antipad,
T_aperture_type,
T_array,
T_attach,
T_attr,
T_average_pair_length,
T_back,
T_base_design,
T_bbv_ctr2ctr,
T_bend_keepout,
T_bond,
T_both,
T_bottom,
T_bottom_layer_sel,
T_boundary,
T_brickpat,
T_bundle,
T_bus,
T_bypass,
T_capacitance_resolution,
T_capacitor,
T_case_sensitive,
T_cct1,
T_cct1a,
T_center_center,
T_checking_trim_by_pin,
T_circ,
T_circle,
T_circuit,
T_class,
T_class_class,
T_classes,
T_clear,
T_clearance,
T_cluster,
T_cm,
T_color,
T_colors,
T_comment,
T_comp,
T_comp_edge_center,
T_comp_order,
T_component,
T_composite,
T_conductance_resolution,
T_conductor,
T_conflict,
T_connect,
T_constant,
T_contact,
T_control,
T_corner,
T_corners,
T_cost,
T_created_time,
T_cross,
T_crosstalk_model,
T_current_resolution,
T_delete_pins,
T_deleted,
T_deleted_keepout,
T_delta,
T_diagonal,
T_direction,
T_directory,
T_discrete,
T_effective_via_length,
T_elongate_keepout,
T_exclude,
T_expose,
T_extra_image_directory,
T_family,
T_family_family,
T_family_family_spacing,
T_fanout,
T_farad,
T_file,
T_fit,
T_fix,
T_flip_style,
T_floor_plan,
T_footprint,
T_forbidden,
T_force_to_terminal_point,
T_free,
T_forgotten,
T_fromto,
T_front,
T_front_only,
T_gap,
T_gate,
T_gates,
T_generated_by_freeroute,
T_global,
T_grid,
T_group,
T_group_set,
T_guide,
T_hard,
T_height,
T_high,
T_history,
T_horizontal,
T_host_cad,
T_host_version,
T_image,
T_image_conductor,
T_image_image,
T_image_image_spacing,
T_image_outline_clearance,
T_image_set,
T_image_type,
T_inch,
T_include,
T_include_pins_in_crosstalk,
T_inductance_resolution,
T_insert,
T_instcnfg,
T_inter_layer_clearance,
T_jumper,
T_junction_type,
T_keepout,
T_kg,
T_kohm,
T_large,
T_large_large,
T_layer,
T_layer_depth,
T_layer_noise_weight,
T_layer_pair,
T_layer_rule,
T_length,
T_length_amplitude,
T_length_factor,
T_length_gap,
T_library,
T_library_out,
T_limit,
T_limit_bends,
T_limit_crossing,
T_limit_vias,
T_limit_way,
T_linear,
T_linear_interpolation,
T_load,
T_lock_type,
T_logical_part,
T_logical_part_mapping,
T_low,
T_match_fromto_delay,
T_match_fromto_length,
T_match_group_delay,
T_match_group_length,
T_match_net_delay,
T_match_net_length,
T_max_delay,
T_max_len,
T_max_length,
T_max_noise,
T_max_restricted_layer_length,
T_max_stagger,
T_max_stub,
T_max_total_delay,
T_max_total_length,
T_max_total_vias,
T_medium,
T_mhenry,
T_mho,
T_microvia,
T_mid_driven,
T_mil,
T_min_gap,
T_mirror,
T_mirror_first,
T_mixed,
T_mm,
T_negative_diagonal,
T_net,
T_net_number,
T_net_out,
T_net_pin_changes,
T_nets,
T_network,
T_network_out,
T_no,
T_noexpose,
T_noise_accumulation,
T_noise_calculation,
T_normal,
T_object_type,
T_off,
T_off_grid,
T_offset,
T_on,
T_open,
T_opposite_side,
T_order,
T_orthogonal,
T_outline,
T_overlap,
T_pad,
T_pad_pad,
T_padstack,
T_pair,
T_parallel,
T_parallel_noise,
T_parallel_segment,
T_parser,
T_part_library,
T_path,
T_pcb,
T_permit_orient,
T_permit_side,
T_physical,
T_physical_part_mapping,
T_piggyback,
T_pin,
T_pin_allow,
T_pin_cap_via,
T_pin_via_cap,
T_pin_width_taper,
T_pins,
T_pintype,
T_place,
T_place_boundary,
T_place_control,
T_place_keepout,
T_place_rule,
T_placement,
T_plan,
T_plane,
T_pn,
T_point,
T_polyline_path,
T_polygon,
T_position,
T_positive_diagonal,
T_power,
T_power_dissipation,
T_power_fanout,
T_prefix,
T_primary,
T_priority,
T_property,
T_protect,
T_qarc,
T_quarter,
T_radius,
T_ratio,
T_ratio_tolerance,
T_rect,
T_reduced,
T_region,
T_region_class,
T_region_class_class,
T_region_net,
T_relative_delay,
T_relative_group_delay,
T_relative_group_length,
T_relative_length,
T_reorder,
T_reroute_order_viols,
T_resistance_resolution,
T_resistor,
T_resolution,
T_restricted_layer_length_factor,
T_room,
T_rotate,
T_rotate_first,
T_round,
T_roundoff_rotation,
T_route,
T_route_to_fanout_only,
T_routes,
T_routes_include,
T_rule,
T_same_net_checking,
T_sample_window,
T_saturation_length,
T_sec,
T_secondary,
T_self,
T_sequence_number,
T_session,
T_set_color,
T_set_pattern,
T_shape,
T_shield,
T_shield_gap,
T_shield_loop,
T_shield_tie_down_interval,
T_shield_width,
T_side,
T_signal,
T_site,
T_small,
T_smd,
T_snap,
T_snap_angle,
T_soft,
T_source,
T_space_in_quoted_tokens,
T_spacing,
T_spare,
T_spiral_via,
T_square,
T_stack_via,
T_stack_via_depth,
T_standard,
T_starburst,
T_status,
T_structure,
T_structure_out,
T_subgate,
T_subgates,
T_substituted,
T_such,
T_suffix,
T_super_placement,
T_supply,
T_supply_pin,
T_swapping,
T_switch_window,
T_system,
T_tandem_noise,
T_tandem_segment,
T_tandem_shield_overhang,
T_terminal,
T_terminator,
T_term_only,
T_test,
T_test_points,
T_testpoint,
T_threshold,
T_time_length_factor,
T_time_resolution,
T_tjunction,
T_tolerance,
T_top,
T_topology,
T_total,
T_track_id,
T_turret,
T_type,
T_um,
T_unassigned,
T_unconnects,
T_unit,
T_up,
T_use_array,
T_use_layer,
T_use_net,
T_use_via,
T_value,
T_vertical,
T_via,
T_via_array_template,
T_via_at_smd,
T_via_keepout,
T_via_number,
T_via_rotate_first,
T_via_site,
T_via_size,
T_virtual_pin,
T_volt,
T_voltage_resolution,
T_was_is,
T_way,
T_weight,
T_width,
T_window,
T_wire,
T_wire_keepout,
T_wires,
T_wires_include,
T_wiring,
T_write_resolution,
T_x,
T_xy,
T_y
};
class SPECCTRA_DB;
......@@ -490,7 +79,7 @@ class SPECCTRA_DB;
* SPECCTRA_DB::keyword. We needed a non-instanance function to get at
* the SPECCTRA_DB::keyword[] and class SPECCTRA_DB is not defined yet.
*/
const char* GetTokenText( int aTok );
const char* GetTokenText( T aTok );
/**
......@@ -820,7 +409,6 @@ public:
*/
static UNIT_RES Default;
UNIT_RES( ELEM* aParent, DSN_T aType ) :
ELEM( aType, aParent )
{
......@@ -1814,14 +1402,10 @@ class GRID : public ELEM
friend class SPECCTRA_DB;
DSN_T grid_type; ///< T_via | T_wire | T_via_keepout | T_place | T_snap
double dimension;
int direction; ///< T_x | T_y | -1 for both
DSN_T direction; ///< T_x | T_y | -1 for both
double offset;
int image_type; // DSN_T
DSN_T image_type;
public:
......@@ -2060,18 +1644,18 @@ public:
PLACE( ELEM* aParent ) :
ELEM( T_place, aParent )
{
side = DSN_T( T_front );
side = T_front;
rotation = 0.0;
hasVertex = false;
mirror = DSN_T( T_NONE );
status = DSN_T( T_NONE );
mirror = T_NONE;
status = T_NONE;
place_rules = 0;
lock_type = DSN_T( T_NONE );
lock_type = T_NONE;
rules = 0;
region = 0;
}
......@@ -3969,16 +3553,15 @@ typedef boost::ptr_set<PADSTACK> PADSTACKSET;
/**
* Class SPECCTRA_DB
* holds a DSN data tree, usually coming from a DSN file.
* holds a DSN data tree, usually coming from a DSN file. Is essentially a
* SPECCTRA_PARSER class.
*/
class SPECCTRA_DB
class SPECCTRA_DB : public SPECCTRA_LEXER
{
/// specctra DSN keywords
static const KEYWORD keywords[];
static const unsigned keywordCount;
DSNLEXER* lexer;
PCB* pcb;
SESSION* session;
wxString filename;
......@@ -4024,76 +3607,6 @@ class SPECCTRA_DB
*/
int findLayerName( const std::string& aLayerName ) const;
/**
* Function nextTok
* returns the next token from the lexer as a DSN_T. Note to anybody
* who wants to use SPECCTRA_DB as a model for usage of DSNLEXER, you
* want to have this function return an enum, not an int, and to use
* that enum type whereever you can, because this allows the debugger
* to show you symbolic values for your tokens.
*/
DSN_T nextTok();
/**
* Function isSymbol
* tests a token to see if it is a symbol. This means it cannot be a
* special delimiter character such as T_LEFT, T_RIGHT, T_QUOTE, etc. It may
* however, coincidentally match a keyword and still be a symbol.
*/
static bool isSymbol( DSN_T aTok )
{
return DSNLEXER::IsSymbol( aTok );
}
/**
* Function needLEFT
* calls nextTok() and then verifies that the token read in is a T_LEFT.
* If it is not, an IO_ERROR is thrown.
* @throw IO_ERROR, if the next token is not a T_LEFT
*/
void needLEFT() throw( IO_ERROR )
{
lexer->NeedLEFT();
}
/**
* Function needRIGHT
* calls nextTok() and then verifies that the token read in is a T_RIGHT.
* If it is not, an IO_ERROR is thrown.
* @throw IO_ERROR, if the next token is not a T_RIGHT
*/
void needRIGHT() throw( IO_ERROR )
{
lexer->NeedRIGHT();
}
/**
* Function needSYMBOL
* calls nextTok() and then verifies that the token read in
* satisfies bool isSymbol().
* If not, an IO_ERROR is thrown.
* @return DSN_T - the actual token read in.
* @throw IO_ERROR, if the next token does not satisfy isSymbol()
*/
DSN_T needSYMBOL() throw( IO_ERROR )
{
return (DSN_T) lexer->NeedSYMBOL();
}
/**
* Function needSYMBOLorNUMBER
* calls nextTok() and then verifies that the token read in
* satisfies bool isSymbol() or tok==T_NUMBER.
* If not, an IO_ERROR is thrown.
* @return DSN_T - the actual token read in.
* @throw IO_ERROR, if the next token does not satisfy the above test
*/
DSN_T needSYMBOLorNUMBER() throw( IO_ERROR )
{
return (DSN_T) lexer->NeedSYMBOLorNUMBER();
}
/**
* Function readCOMPnPIN
* reads a &lt;pin_reference&gt; and splits it into the two parts which are
......@@ -4103,7 +3616,7 @@ class SPECCTRA_DB
* single T_SYMBOL, so in that case we have to split it into two here.
* <p>
* The caller should have already read in the first token comprizing the
* pin_reference and it will be tested through lexer->CurTok().
* pin_reference and it will be tested through CurTok().
*
* @param component_id Where to put the text preceeding the '-' hyphen.
* @param pin_d Where to put the text which trails the '-'.
......@@ -4112,7 +3625,6 @@ class SPECCTRA_DB
*/
void readCOMPnPIN( std::string* component_id, std::string* pid_id ) throw( IO_ERROR );
/**
* Function readTIME
* reads a &lt;time_stamp&gt; which consists of 8 lexer tokens:
......@@ -4129,24 +3641,6 @@ class SPECCTRA_DB
*/
void readTIME( time_t* time_stamp ) throw( IO_ERROR );
/**
* Function expecting
* throws an IO_ERROR exception with an input file specific error message.
* @param int is the token type which was expected at the current input location.
* @throw IO_ERROR with the location within the input file of the problem.
*/
void expecting( DSN_T aTok ) throw( IO_ERROR )
{
lexer->Expecting( aTok );
}
void unexpected( DSN_T aTok ) throw( IO_ERROR )
{
lexer->Unexpected( aTok );
}
void expecting( const char* text ) throw( IO_ERROR );
void unexpected( const char* text ) throw( IO_ERROR );
void doPCB( PCB* growth ) throw( IO_ERROR );
void doPARSER( PARSER* growth ) throw( IO_ERROR );
void doRESOLUTION( UNIT_RES* growth ) throw( IO_ERROR );
......@@ -4295,9 +3789,11 @@ class SPECCTRA_DB
public:
SPECCTRA_DB()
SPECCTRA_DB() :
SPECCTRA_LEXER( 0 ) // LINE_READER* == NULL, no DSNLEXER::PushReader()
{
lexer = 0;
iOwnReaders = true; // if an exception is thrown, close file.
pcb = 0;
session = 0;
quote_char += '"';
......@@ -4306,22 +3802,12 @@ public:
virtual ~SPECCTRA_DB()
{
delete lexer;
delete pcb;
delete session;
deleteNETs();
}
static const char* TokenName( int aToken );
/**
* Function GetTokenString
* returns the wxString representation of aToken.
*/
static wxString GetTokenString( int aToken );
/**
* Function MakePCB
* makes a PCB with all the default ELEMs and parts on the heap.
......
# Copyright (C) 2007-2011 SoftPLC Corporation, Dick Hollenbeck <dick@softplc.com>
# These are keywords for the Specctra DSN (s-expression) lexer and parser.
absolute
added
add_group
add_pins
allow_antenna
allow_redundant_wiring
amp
ancestor
antipad
aperture_type
array
attach
attr
average_pair_length
back
base_design
bbv_ctr2ctr
bend_keepout
bond
both
bottom
bottom_layer_sel
boundary
brickpat
bundle
bus
bypass
capacitance_resolution
capacitor
case_sensitive
cct1
cct1a
center_center
checking_trim_by_pin
circ
circle
circuit
class
class_class
classes
clear
clearance
cluster
cm
color
colors
comment
comp
comp_edge_center
comp_order
component
composite
conductance_resolution
conductor
conflict
connect
constant
contact
control
corner
corners
cost
created_time
cross
crosstalk_model
current_resolution
delete_pins
deleted
deleted_keepout
delta
diagonal
direction
directory
discrete
effective_via_length
elongate_keepout
exclude
expose
extra_image_directory
family
family_family
family_family_spacing
fanout
farad
file
fit
fix
flip_style
floor_plan
footprint
forbidden
force_to_terminal_point
forgotten
free
fromto
front
front_only
gap
gate
gates
generated_by_freeroute
global
grid
group
group_set
guide
hard
height
high
history
horizontal
host_cad
host_version
image
image_conductor
image_image
image_image_spacing
image_outline_clearance
image_set
image_type
inch
include
include_pins_in_crosstalk
inductance_resolution
insert
instcnfg
inter_layer_clearance
jumper
junction_type
keepout
kg
kohm
large
large_large
layer
layer_depth
layer_noise_weight
layer_pair
layer_rule
length
length_amplitude
length_factor
length_gap
library
library_out
limit
limit_bends
limit_crossing
limit_vias
limit_way
linear
linear_interpolation
load
lock_type
logical_part
logical_part_mapping
low
match_fromto_delay
match_fromto_length
match_group_delay
match_group_length
match_net_delay
match_net_length
max_delay
max_len
max_length
max_noise
max_restricted_layer_length
max_stagger
max_stub
max_total_delay
max_total_length
max_total_vias
medium
mhenry
mho
microvia
mid_driven
mil
min_gap
mirror
mirror_first
mixed
mm
negative_diagonal
net
net_number
net_out
net_pin_changes
nets
network
network_out
no
noexpose
noise_accumulation
noise_calculation
normal
object_type
off
off_grid
offset
on
open
opposite_side
order
orthogonal
outline
overlap
pad
pad_pad
padstack
pair
parallel
parallel_noise
parallel_segment
parser
part_library
path
pcb
permit_orient
permit_side
physical
physical_part_mapping
piggyback
pin
pin_allow
pin_cap_via
pin_via_cap
pin_width_taper
pins
pintype
place
place_boundary
place_control
place_keepout
place_rule
placement
plan
plane
pn
point
polyline_path
polygon
position
positive_diagonal
power
power_dissipation
power_fanout
prefix
primary
priority
property
protect
qarc
quarter
radius
ratio
ratio_tolerance
rect
reduced
region
region_class
region_class_class
region_net
relative_delay
relative_group_delay
relative_group_length
relative_length
reorder
reroute_order_viols
resistance_resolution
resistor
resolution
restricted_layer_length_factor
room
rotate
rotate_first
round
roundoff_rotation
route
route_to_fanout_only
routes
routes_include
rule
same_net_checking
sample_window
saturation_length
sec
secondary
self
sequence_number
session
set_color
set_pattern
shape
shield
shield_gap
shield_loop
shield_tie_down_interval
shield_width
side
signal
site
small
smd
snap
snap_angle
soft
source
space_in_quoted_tokens
spacing
spare
spiral_via
square
stack_via
stack_via_depth
standard
starburst
status
structure
structure_out
subgate
subgates
substituted
such
suffix
super_placement
supply
supply_pin
swapping
switch_window
system
tandem_noise
tandem_segment
tandem_shield_overhang
terminal
terminator
term_only
test
test_points
testpoint
threshold
time_length_factor
time_resolution
tjunction
tolerance
top
topology
total
track_id
turret
type
um
unassigned
unconnects
unit
up
use_array
use_layer
use_net
use_via
value
vertical
via
via_array_template
via_at_smd
via_keepout
via_number
via_rotate_first
via_site
via_size
virtual_pin
volt
voltage_resolution
was_is
way
weight
width
window
wire
wire_keepout
wires
wires_include
wiring
write_resolution
x # test cmake script with indent and comment
xy
y
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment