65816.tmbundle/mac-gen.py

452 lines
12 KiB
Python
Raw Normal View History

2018-10-13 22:02:15 +00:00
import sublime
import sublime_plugin
import os
import re
import sys
import time
def each_line(file):
f = open(file, "r", encoding="ascii")
try:
while True:
x = f.readline()
if not x: break
x = x.rstrip("\r\n")
yield x
except Exception as e:
raise e
finally:
f.close()
def macro_files(dirs, filter):
rv = []
for root in dirs:
try:
tmp = [x for x in os.listdir(root) if filter(x)]
tmp.sort(key=str.upper)
tmp = [os.path.join(root, x) for x in tmp]
tmp = [x for x in tmp if os.path.isfile(x)]
rv += tmp
except Exception as e:
pass
return rv
2024-10-30 20:55:21 +00:00
2018-10-13 22:02:15 +00:00
re_pmc = re.compile('^[^.,/\\- ]*')
re_dot_s = re.compile('\\.S$', re.IGNORECASE)
re_ws = re.compile('[\t ]+')
re_comment = re.compile('^[\t ]*[;*]')
2024-10-30 20:55:21 +00:00
class CommonMacGen(sublime_plugin.TextCommand):
2018-10-13 22:02:15 +00:00
2024-10-30 20:55:21 +00:00
def run(self, edit):
view = self.view
missing = set()
known = set()
self.build_missing(known, missing)
if len(missing) == 0: return
buffer = ""
files = self.build_file_list()
for f in files:
buffer += self.one_macro_file(f, known, missing)
if len(missing) == 0: break;
buffer = self.front_matter() + self.missing_matter(missing) + buffer
self.create_macro_file(buffer)
2018-10-13 22:02:15 +00:00
def front_matter(self):
buffer = ""
name = self.view.file_name()
if name:
buffer += "* Generated from " + os.path.basename(name) + " on " + time.asctime() + "\n"
else:
buffer += "* Generated on " + time.asctime() + "\n"
buffer += "\n"
return buffer
def missing_matter(self, missing):
buffer = ""
if len(missing) == 0:
return ""
buffer = "* Missing Macros\n"
missing = list(missing)
missing.sort
for macro in missing:
buffer += "* " + macro + "\n"
buffer += "\n"
return buffer
2024-10-30 20:55:21 +00:00
def create_macro_file(self, buffer):
2018-10-13 22:02:15 +00:00
view = self.view
2024-10-30 20:55:21 +00:00
mv = view.window().new_file()
mv.settings().set("auto_indent", False)
mv.run_command('insert', {'characters': buffer})
2018-10-13 22:02:15 +00:00
2024-10-30 20:55:21 +00:00
# override these:
def build_missing(self, known, missing):
pass
2018-10-13 22:02:15 +00:00
2024-10-30 20:55:21 +00:00
def build_file_list(self):
return []
2018-10-13 22:02:15 +00:00
2024-10-30 20:55:21 +00:00
def one_macro_file(self, file, known, missing):
return ""
2018-10-13 22:02:15 +00:00
2024-10-30 20:55:21 +00:00
class MerlinMacGen(CommonMacGen):
2018-10-13 22:02:15 +00:00
2024-10-30 20:55:21 +00:00
opcodes = {
'BRL', 'COP', 'JML', 'JMPL', 'JSL', 'MVN', 'MVP', 'MX', 'PEA',
'PEI', 'PER', 'REP', 'SEP', 'WDM', 'XCE', 'ADR', 'ADRL', 'ASC',
'AST', 'BRK', 'CHK', 'CYC', 'DA', 'DAT', 'DB', 'DCI', 'DDB', 'DEND',
'DFB', 'DL', 'DS', 'DSK', 'DUM', 'DW', 'END', 'ENT', 'EQU', 'ERR', 'EXP',
'EXT', 'FLS', 'HEX', 'INV', 'KBD', 'LST', 'LSTL', 'LUP', 'OBJ',
'ORG', 'PAG', 'PAU', 'PMC', 'PUT', 'REL', 'REV', 'SAV', 'SKP',
'STR', 'TR', 'TTL', 'TYP', 'USE', 'USR', 'VAR', 'XC', 'XREF','=',
'>>>', 'DO', 'ELS', 'EOM', 'FIN', 'IF', 'MAC', '--^', '<<<',
'CLC', 'CLD', 'CLI', 'CLV', 'DEX', 'DEY', 'INX', 'INY', 'NOP',
'PHA', 'PHB', 'PHD', 'PHK', 'PHX', 'PHY', 'PLA', 'PLB', 'PLD',
'PLP', 'PLX', 'PLY', 'RTI', 'RTL', 'RTS', 'SEC', 'SED', 'SEI',
'STP', 'SWA', 'TAD', 'TAS', 'TAX', 'TAY', 'TCD', 'TCS', 'TDA',
'TDC', 'TSA', 'TSC', 'TSX', 'TXA', 'TXS', 'TXY', 'TYA', 'TYX',
'WAI', 'XBA', 'BCC', 'BCS', 'BEQ', 'BGE', 'BLT', 'BMI', 'BNE',
'BPL', 'BRA', 'BVC', 'BVS', 'ADCL', 'ANDL', 'CMPL', 'EORL',
'LDAL', 'ORAL', 'SBCL', 'STAL', 'ADC', 'AND', 'ASL', 'BIT',
'CMP', 'CPX', 'CPY', 'DEC', 'EOR', 'INC', 'JMP', 'JSR', 'LDA',
'LDX', 'LDY', 'LSR', 'ORA', 'ROL', 'ROR', 'SBC', 'STA', 'STX',
'STY', 'STZ', 'TRB', 'TSB', 'PHP', 'STRL', 'FLO', 'EXD', 'CAS',
''
}
2018-10-13 22:02:15 +00:00
2024-10-30 20:55:21 +00:00
def build_file_list(self):
dirlist = self.view.settings().get('merlin_macro_directories', [])
files = macro_files(dirlist, lambda x: x[-2:].upper() == ".S")
return files
2018-10-13 22:02:15 +00:00
def build_missing(self, known, missing):
view = self.view
all = sublime.Region(0, view.size())
for r in view.lines(all):
text = view.substr(r).rstrip()
if text == "" or re_comment.match(text):
continue
tokens = re.split(re_ws, text, maxsplit=2)
if len(tokens) < 2:
continue
# _macro args
2024-10-30 20:55:21 +00:00
# PMC _macro,args << . / , - ( Space are separators.
2018-10-13 22:02:15 +00:00
# >>> _macro
label = tokens[0].upper()
opcode = tokens[1].upper()
if opcode == "" or opcode[0] == ";":
continue
if opcode == 'MAC':
# label MAC
if label != "":
known.add(label)
continue
if opcode in {'PMC', '>>>'} and len(tokens) > 2:
tmp = tokens[2].upper()
opcode = re_pmc.match(tmp).group(0)
if opcode == "":
continue
if opcode in self.opcodes:
continue
if opcode in known:
continue
missing.add(opcode)
def one_macro_file(self, file, known, missing):
inmac = False
buffer = ""
for text in each_line(file):
text = text.rstrip()
if inmac:
buffer += text + "\n"
if text == "": continue
if re_comment.match(text): continue
tokens = re.split(re_ws, text, maxsplit=2)
if len(tokens) < 2: continue
label = tokens[0].upper()
opcode = tokens[1].upper()
if opcode == "" or opcode[0] == ";": continue
if opcode == "MAC" and label != "":
# print(label,file=sys.stderr)
# print(label in missing, file=sys.stderr)
if inmac:
known.add(label)
missing.discard(label)
elif label in missing:
inmac = True
known.add(label)
missing.discard(label)
buffer += text + "\n"
continue
if opcode in { '<<<', 'EOM' }:
inmac = False
continue
if not inmac: continue
if opcode in {'PMC', '>>>'} and len(tokens) > 2:
tmp = tokens[2].upper()
opcode = re_pmc.match(tmp).group(0)
if opcode == "":
continue
if opcode in self.opcodes: continue
if opcode in known: continue
missing.add(opcode)
return buffer
2024-10-30 20:55:21 +00:00
class OrcaMacGen(CommonMacGen):
2018-10-13 22:02:15 +00:00
opcodes = {
'ADC', 'AND', 'CMP', 'EOR', 'LDA', 'ORA', 'SBC', 'ASL', 'LSR',
'ROR', 'ROL', 'BIT', 'CPX', 'CPY', 'DEC', 'INC', 'LDX', 'LDY',
'STA', 'STX', 'STY', 'JMP', 'JSR', 'STZ', 'TRB', 'TSB', 'JSL',
'JML', 'COP', 'MVN', 'MVP', 'PEA', 'PEI', 'REP', 'SEP', 'PER',
'BRL', 'BRA', 'BEQ', 'BMI', 'BNE', 'BPL', 'BVC', 'BVS', 'BCC',
'BCS', 'CLI', 'CLV', 'DEX', 'DEY', 'INX', 'INY', 'NOP', 'PHA',
'PLA', 'PHP', 'PLP', 'RTI', 'RTS', 'SEC', 'SED', 'SEI', 'TAX',
'TAY', 'TSX', 'TXA', 'TXS', 'TYA', 'BRK', 'CLC', 'CLD', 'PHX',
'PHY', 'PLX', 'PLY', 'DEA', 'INA', 'PHB', 'PHD', 'PHK', 'PLB',
'PLD', 'WDM', 'RTL', 'STP', 'TCD', 'TCS', 'TDC', 'TSC', 'TXY',
'TYX', 'WAI', 'XBA', 'XCE', 'CPA', 'BLT', 'BGE', 'GBLA', 'GBLB',
'GBLC', 'LCLA', 'LCLB', 'LCLC', 'SETA', 'SETB', 'SETC', 'AMID',
'ASEARCH', 'AINPUT', 'AIF', 'AGO', 'ACTR', 'MNOTE', 'ANOP', 'DS',
'ORG', 'OBJ', 'EQU', 'GEQU', 'MERR', 'DIRECT', 'KIND', 'SETCOM',
'EJECT', 'ERR', 'GEN', 'MSB', 'LIST', 'SYMBOL', 'PRINTER',
'65C02', '65816', 'LONGA', 'LONGI', 'DATACHK', 'CODECHK',
'DYNCHK', 'IEEE', 'NUMSEX', 'CASE', 'OBJCASE', 'ABSADDR',
'INSTIME', 'TRACE', 'EXPAND', 'DC', 'USING', 'ENTRY', 'OBJEND',
'DATA', 'PRIVDATA', 'END', 'ALIGN', 'START', 'PRIVATE', 'MEM',
'TITLE', 'RENAME', 'KEEP', 'COPY', 'APPEND', 'MCOPY', 'MDROP',
'MLOAD', 'MACRO', 'MEXIT', 'MEND'
}
2024-10-30 20:55:21 +00:00
def build_file_list(self):
dirlist = self.view.settings().get('orca_macro_directories', [])
2018-10-13 22:02:15 +00:00
files = macro_files(dirlist, lambda x: x[:4].upper() == "M16.")
2024-10-30 20:55:21 +00:00
return files
2018-10-13 22:02:15 +00:00
def build_missing(self, known, missing):
view = self.view
all = sublime.Region(0, view.size())
for r in view.lines(all):
text = view.substr(r).rstrip()
if text == "": continue
if text[0] in "!*": continue
2018-10-13 22:02:15 +00:00
if re_comment.match(text): continue
tokens = re.split(re_ws, text, maxsplit=2)
if len(tokens) < 2: continue
label = tokens[0].upper()
opcode = tokens[1].upper()
if opcode in self.opcodes: continue
missing.add(opcode)
def one_macro_file(self, file, known, missing):
buffer = ""
state = 0
for text in each_line(file):
if state == 2:
buffer += text + "\n"
if text == "": continue
if re_comment.match(text): continue
tokens = re.split(re_ws, text, maxsplit=2)
if len(tokens) < 2: continue
label = tokens[0].upper()
opcode = tokens[1].upper()
if state == 0:
if opcode == 'MACRO': state = 1
elif state == 1:
if opcode in missing:
known.add(opcode)
missing.discard(opcode)
buffer += "\tMACRO\n"
buffer += text + "\n"
state = 2
else:
state = 0
elif state == 2:
if opcode == 'MEND': state = 0
if opcode not in self.opcodes and opcode not in known:
missing.add(opcode)
return buffer
2024-10-30 20:55:21 +00:00
class MpwMacGen(CommonMacGen):
2018-10-13 22:02:15 +00:00
opcodes = {
'ACTR', 'ADC', 'AERROR', 'ALIGN', 'AND', 'ANOP', 'ASL',
'BBC0', 'BBC1', 'BBC2', 'BBC3', 'BBC4', 'BBC5', 'BBC6', 'BBC7',
'BBR0', 'BBR1', 'BBR2', 'BBR3', 'BBR4', 'BBR5', 'BBR6', 'BBR7',
'BBS0', 'BBS1', 'BBS2', 'BBS3', 'BBS4', 'BBS5', 'BBS6', 'BBS7',
'BCC', 'BCS', 'BEQ', 'BGE', 'BIT', 'BLANKS', 'BLT', 'BMI', 'BNE',
'BPL', 'BRA', 'BRK', 'BRL', 'BVC', 'BVS', 'CASE',
'CLB0', 'CLB1', 'CLB2', 'CLB3', 'CLB4', 'CLB5', 'CLB6', 'CLB7',
'CLC', 'CLD', 'CLI', 'CLT', 'CLV', 'CMP', 'CODECHK', 'COM', 'COP', 'CPA',
'CPX', 'CPY', 'CYCLE', 'DATACHK', 'DC', 'DCB', 'DS', 'DEA', 'DEC', 'DEX', 'DEY',
2018-10-13 22:02:15 +00:00
'DIRECT', 'DUMP', 'EJECT', 'ELSE', 'ELSEIF', 'END', 'ENDF', 'ENDFUNC',
'ENDI', 'ENDIF', 'ENDINIT', 'ENDM', 'ENDMACRO', 'ENDP', 'ENDPROC',
'ENDR', 'ENDS', 'ENDSTACK', 'ENDW', 'ENDWHILE', 'ENDWITH', 'ENTRY',
'EOR', 'EQU', 'EXITM', 'EXPORT', 'FUNC', 'GBLA', 'GBLC', 'GOTO',
'IF', 'IMPORT', 'INA', 'INC', 'INCLUDE', 'INIT', 'INX', 'INY', 'JML', 'JMP',
2018-10-13 22:02:15 +00:00
'JSL', 'JSR', 'LCLA', 'LCLC', 'LDA', 'LDM', 'LDX', 'LDY', 'LEAVE',
'LOAD', 'LONGA', 'LONGI', 'LSR', 'MACHINE', 'MACRO', 'MEND', 'MEXIT',
'MSB', 'MVN', 'MVP', 'NOP', 'ORA', 'ORG', 'PAGESIZE', 'PEA', 'PEI',
'PER', 'PHA', 'PHB', 'PHD', 'PHK', 'PHP', 'PHX', 'PHY', 'PLA', 'PLB',
'PLD', 'PLP', 'PLX', 'PLY', 'PRINT', 'PROC', 'RECORD', 'REP', 'RMB0',
'RMB1', 'RMB2', 'RMB3', 'RMB4', 'RMB5', 'RMB6', 'RMB7', 'ROL', 'ROR',
'RRF', 'RTI', 'RTL', 'RTS', 'SBC', 'SEB0', 'SEB1', 'SEB2', 'SEB3',
'SEB4', 'SEB5', 'SEB6', 'SEB7', 'SEC', 'SED', 'SEG', 'SEGATTR', 'SEI',
'SEP', 'SET', 'SETA', 'SETC', 'SETT', 'SMB0', 'SMB1', 'SMB2', 'SMB3',
'SMB4', 'SMB5', 'SMB6', 'SMB7', 'SPACE', 'STA', 'STACKDP', 'STP',
'STRING', 'STX', 'STY', 'STZ', 'SWA', 'TAD', 'TAS', 'TAX', 'TAY', 'TCD',
'TCS', 'TDA', 'TDC', 'TITLE', 'TRB', 'TSA', 'TSB', 'TSC', 'TST', 'TSX',
'TXA', 'TXS', 'TXY', 'TYA', 'TYX', 'WAI', 'WHILE', 'WITH', 'WRITE',
'WRITELN', 'XBA', 'XCE'
}
2024-10-30 20:55:21 +00:00
def build_file_list(self):
dirlist = self.view.settings().get('mpw_macro_directories', [])
2018-10-13 22:02:15 +00:00
files = macro_files(dirlist, lambda x: x[:4].upper() == "M16.")
2024-10-30 20:55:21 +00:00
return files
2018-10-13 22:02:15 +00:00
def build_missing(self, known, missing):
state = 0
view = self.view
all = sublime.Region(0, view.size())
for r in view.lines(all):
text = view.substr(r).rstrip()
if text == "": continue
if re_comment.match(text): continue
tokens = re.split(re_ws, text, maxsplit=2)
if len(tokens) < 2: continue
label = tokens[0].upper()
opcode = tokens[1].upper()
ix = opcode.find('.')
if ix >= 0: opcode = opcode[:ix]
2018-10-13 22:02:15 +00:00
if opcode == "MACRO":
state = 1
continue
if opcode == "MEND":
state = 0
continue
if state == 1:
known.add(opcode)
missing.discard(opcode)
state = 0
continue
if opcode in self.opcodes: continue
if opcode in known: continue
2018-10-13 22:02:15 +00:00
missing.add(opcode)
def one_macro_file(self, file, known, missing):
buffer = ""
state = 0
for text in each_line(file):
if state == 2:
buffer += text + "\n"
if text == "": continue
if re_comment.match(text): continue
tokens = re.split(re_ws, text, maxsplit=2)
if len(tokens) < 2: continue
label = tokens[0].upper()
opcode = tokens[1].upper()
if state == 0:
if opcode == 'MACRO': state = 1
elif state == 1:
if opcode in missing:
known.add(opcode)
missing.discard(opcode)
buffer += "\tMACRO\n"
buffer += text + "\n"
state = 2
else:
state = 0
elif state == 2:
if opcode == 'MEND': state = 0
if opcode not in self.opcodes and opcode not in known:
missing.add(opcode)
return buffer