ref: 8eba39dd52a2a731225067fa4c3ae42cfb5b232f
parent: 1d122c86cfeebcb14abde2ceec340ae2baa73cd2
author: Rangi <remy.oukaour+rangi42@gmail.com>
date: Wed Mar 11 11:06:34 EDT 2020
Add tools/used_space.py to visualize the ROM based on the .map file
--- /dev/null
+++ b/tools/mapreader.py
@@ -1,0 +1,172 @@
+# -*- coding: utf-8 -*-
+
+# A library for parsing the pokecrystal.map file output by rgbds.
+
+import re
+
+class MapReader:
+
+ # {'ROM Bank': { 0: { 'sections': [ { 'beg': 1234,
+ # 'end': 5678,
+ # 'name': 'Section001',
+ # 'symbols': [ { 'symbol': 'Function1234',
+ # 'address: 1234,
+ # },
+ # ]
+ # },
+ # ],
+ # 'used': 1234,
+ # 'slack': 4567,
+ # },
+ # },
+ # 'OAM': { 'sections': [ { 'beg': 1234,
+ # 'end': 5678,
+ # 'name': 'Section002',
+ # 'symbols': [ { 'symbol': 'Data1234',
+ # 'address: 1234,
+ # },
+ # ]
+ # },
+ # ],
+ # 'used': 1234,
+ # 'slack': 4567,
+ # },
+ # }
+ #
+ bank_data = {}
+
+ bank_types = {
+ 'HRAM' : { 'size': 0x80, 'banked': False, },
+ 'OAM' : { 'size': 0xA0, 'banked': False, },
+ 'ROM Bank' : { 'size': 0x4000, 'banked': True, },
+ 'SRAM Bank': { 'size': 0x2000, 'banked': True, },
+ 'VRAM Bank': { 'size': 0x1000, 'banked': True, },
+ 'WRAM Bank': { 'size': 0x2000, 'banked': True, },
+ }
+
+ # FSM states
+ INIT, BANK, SECTION = range(3)
+
+ # $506D-$519A ($012E bytes) ["Type Matchups"]
+ section_header_regex = re.compile('\$([0-9A-Fa-f]{4})-\$([0-9A-Fa-f]{4}) \(.*\) \["(.*)"\]')
+ # $506D = TypeMatchups
+ section_data_regex = re.compile('\$([0-9A-Fa-f]{4}) = (.*)')
+ # $3ED2 bytes
+ slack_regex = re.compile('\$([0-9A-Fa-f]{4}) bytes')
+
+ def __init__(self, *args, **kwargs):
+ self.__dict__.update(kwargs)
+
+ def _parse_init(self, line):
+
+ line = line.split(':', 1)[0]
+ parts = line.split(' #', 1)
+
+ if (parts[0] in self.bank_types):
+ self._cur_bank_name = parts[0]
+ self._cur_bank_type = self.bank_types[self._cur_bank_name]
+ if (self._cur_bank_type['banked'] and len(parts) > 1):
+ parts[1] = parts[1].split(':', 1)[0]
+ parts[1] = parts[1].split(' ', 1)[0]
+ self._cur_bank = int(parts[1], 10)
+ if self._cur_bank_name not in self.bank_data:
+ self.bank_data[self._cur_bank_name] = {}
+ if self._cur_bank_type['banked']:
+ if self._cur_bank not in self.bank_data[self._cur_bank_name]:
+ self.bank_data[self._cur_bank_name][self._cur_bank] = {}
+ self._cur_data = self.bank_data[self._cur_bank_name][self._cur_bank]
+ else:
+ self._cur_data = self.bank_data[self._cur_bank_name]
+
+ if ({} == self._cur_data):
+ self._cur_data['sections'] = []
+ self._cur_data['used'] = 0
+ self._cur_data['slack'] = self._cur_bank_type['size']
+ return True
+
+ return False
+
+ def _parse_section_header(self, header):
+
+ section_data = self.section_header_regex.match(header)
+ if section_data is not None:
+ beg = int(section_data.group(1), 16)
+ end = int(section_data.group(2), 16)
+ name = section_data.group(3)
+ self._cur_section = {'beg': beg, 'end': end, 'name': name, 'symbols': []}
+ self._cur_data['sections'].append(self._cur_section)
+ return True
+ return False
+
+ def _parse_slack(self, data):
+
+ slack_data = self.slack_regex.match(data)
+ slack_bytes = int(slack_data.group(1), 16)
+ self._cur_data['slack'] = slack_bytes
+
+ used_bytes = 0
+
+ for s in self._cur_data['sections']:
+ used_bytes += s['end'] - s['beg'] + 1
+
+ self._cur_data['used'] = used_bytes
+
+ def read_map_data(self, map):
+
+ if type(map) is str:
+ map = map.split('\n')
+
+ self._state = MapReader.INIT
+ self._cur_bank_name = ''
+ self._cur_bank_type = {}
+ self._cur_bank = 0
+ self._cur_data = {}
+
+ for line in map:
+
+ line = line.rstrip()
+ if (MapReader.INIT == self._state):
+
+ if (self._parse_init(line)):
+ self._state = MapReader.BANK
+
+ elif (MapReader.BANK == self._state or MapReader.SECTION == self._state):
+
+ if ('' == line):
+ self._state = MapReader.INIT
+ else:
+
+ line = line.lstrip()
+ parts = line.split(': ', 1)
+
+ if (MapReader.SECTION == self._state):
+ section_data = self.section_data_regex.match(parts[0])
+ if section_data is not None:
+ address = int(section_data.group(1), 16)
+ name = section_data.group(2)
+ self._cur_section['symbols'].append({'name': name, 'address': address})
+ continue
+
+ if ('SECTION' == parts[0]):
+ if (self._parse_section_header(parts[1])):
+ self._state = MapReader.SECTION
+ elif ('SLACK' == parts[0]):
+ self._parse_slack(parts[1])
+ self._state = MapReader.INIT
+ elif ('EMPTY' == parts[0]):
+ self._cur_data = {'sections': [], 'used': 0, 'slack': self._cur_bank_type['size']}
+ self._state = MapReader.INIT
+
+ else:
+ pass
+
+ for k, v in self.bank_data.items():
+ if (self.bank_types[k]['banked']):
+ for _, vv in v.items():
+ vv['sections'].sort(key=lambda x: x['beg'])
+ for vvv in vv['sections']:
+ vvv['symbols'].sort(key=lambda x: x['address'])
+ else:
+ v['sections'].sort(key=lambda x: x['beg'])
+ for vv in v['sections']:
+ vv['symbols'].sort(key=lambda x: x['address'])