* src/DisplayModels.py: Fix column handling for repositories

* src/RepositoryView.py: Columns for repositories
* src/GrampsDb/_GrampsDbBase.py: default columns settings for repositories
* src/GrampsDb/_ReadGedcom.py: Import repository information
* src/RelLib/_Event.py: check Event type values


svn: r5826
This commit is contained in:
Don Allingham 2006-01-24 21:09:34 +00:00
parent e4e6e2e7de
commit c0aeb95246
6 changed files with 198 additions and 66 deletions

View File

@ -1,3 +1,10 @@
2006-01-24 Don Allingham <don@gramps-project.org>
* src/DisplayModels.py: Fix column handling for repositories
* src/RepositoryView.py: Columns for repositories
* src/GrampsDb/_GrampsDbBase.py: default columns settings for repositories
* src/GrampsDb/_ReadGedcom.py: Import repository information
* src/RelLib/_Event.py: check Event type values
2006-01-23 Don Allingham <don@gramps-project.org> 2006-01-23 Don Allingham <don@gramps-project.org>
* src/DisplayTabs.py: Add source ref model/tab * src/DisplayTabs.py: Add source ref model/tab
* src/PeopleModel.py: specify handle column * src/PeopleModel.py: specify handle column

View File

@ -646,6 +646,7 @@ class RepositoryModel(BaseModel):
self.column_id, self.column_id,
self.column_type, self.column_type,
self.column_home_url, self.column_home_url,
self.column_street,
self.column_postal_code, self.column_postal_code,
self.column_city, self.column_city,
self.column_county, self.column_county,
@ -662,6 +663,7 @@ class RepositoryModel(BaseModel):
self.column_id, self.column_id,
self.column_type, self.column_type,
self.column_home_url, self.column_home_url,
self.column_street,
self.column_postal_code, self.column_postal_code,
self.column_city, self.column_city,
self.column_county, self.column_county,
@ -700,33 +702,57 @@ class RepositoryModel(BaseModel):
except: except:
return u'' return u''
def column_street(self,data):
try:
if data[5]:
return data[5][0].get_street()
else:
return u''
except:
return u''
def column_county(self,data): def column_county(self,data):
try: try:
return data[4].get_county() if data[5]:
return data[5][0].get_county()
else:
return u''
except: except:
return u'' return u''
def column_state(self,data): def column_state(self,data):
try: try:
return data[4].get_state() if data[5]:
return data[5][0].get_state()
else:
return u''
except: except:
return u'' return u''
def column_country(self,data): def column_country(self,data):
try: try:
return data[4].get_country() if data[5]:
return data[5][0].get_country()
else:
return u''
except: except:
return u'' return u''
def column_postal_code(self,data): def column_postal_code(self,data):
try: try:
return data[4].get_postal_code() if data[5]:
return data[5][0].get_postal_code()
else:
return u''
except: except:
return u'' return u''
def column_phone(self,data): def column_phone(self,data):
try: try:
return data[4].get_phone() if data[5]:
return data[5][0].get_phone()
else:
return u''
except: except:
return u'' return u''
@ -737,11 +763,16 @@ class RepositoryModel(BaseModel):
return unicode(data[6]) return unicode(data[6])
def column_home_url(self,data): def column_home_url(self,data):
return unicode(data[7]) if data[6]:
urllist = data[6]
return unicode(urllist[0].path)
else:
return u""
def column_tooltip(self,data): def column_tooltip(self,data):
try: return ""
t = ToolTips.TipFromFunction(self.db, lambda: self.db.get_repository_from_handle(data[0])) # try:
except: # t = ToolTips.TipFromFunction(self.db, lambda: self.db.get_repository_from_handle(data[0]))
log.error("Failed to create tooltip.", exc_info=True) # except:
return t # log.error("Failed to create tooltip.", exc_info=True)
# return t

View File

@ -1666,7 +1666,7 @@ class GrampsDbBase(GrampsDBCallback):
Returns the Repository display common information stored in the Returns the Repository display common information stored in the
database's metadata. database's metadata.
""" """
default = [(1,0,200),(1,1,75),(0,5,100),(0,6,100),(1,2,100),(1,3,250),(0,4,100),(0,7,100),(0,8,100),(0,9,100),(0,10,100)] default = [(1,0,200),(1,1,75),(0,5,100),(0,6,100),(1,2,100),(1,3,250),(1,4,100),(0,7,100),(0,8,100),(0,9,100),(0,10,100)]
return self._get_columns(REPOSITORY_COL_KEY,default) return self._get_columns(REPOSITORY_COL_KEY,default)
def _delete_primary_from_reference_map(self, handle, transaction): def _delete_primary_from_reference_map(self, handle, transaction):

View File

@ -70,6 +70,10 @@ from _GedcomInfo import *
from _GedTokens import * from _GedTokens import *
from QuestionDialog import ErrorDialog, WarningDialog from QuestionDialog import ErrorDialog, WarningDialog
addr_re = re.compile('(.+)([\n\r]+)(.+)\s*,(.+)\s+(\d+)\s*(.*)')
addr2_re = re.compile('(.+)([\n\r]+)(.+)\s*,(.+)\s+(\d+)')
addr3_re = re.compile('(.+)([\n\r]+)(.+)\s*,(.+)')
#------------------------------------------------------------------------- #-------------------------------------------------------------------------
# #
# latin/utf8 conversions # latin/utf8 conversions
@ -329,6 +333,8 @@ class Reader:
self.transtable2 = self.transtable[0:128] + ('?' * 128) self.transtable2 = self.transtable[0:128] + ('?' * 128)
self.cnv = lambda s: unicode(s) self.cnv = lambda s: unicode(s)
self.broken_conc = False self.broken_conc = False
self.cnt = 0
self.index = 0
def set_charset_fn(self,cnv): def set_charset_fn(self,cnv):
self.cnv = cnv self.cnv = cnv
@ -340,14 +346,15 @@ class Reader:
if len(self.current_list) <= 1 and not self.eof: if len(self.current_list) <= 1 and not self.eof:
self.readahead() self.readahead()
try: try:
d = self.current_list.pop() self.current = self.current_list.pop()
return d return self.current
except: except:
return None return None
def readahead(self): def readahead(self):
while len(self.current_list) < 5: while len(self.current_list) < 5:
line = self.f.readline() line = self.f.readline()
self.index += 1
if line == "": if line == "":
self.f.close() self.f.close()
self.eof = True self.eof = True
@ -365,18 +372,18 @@ class Reader:
except: except:
level = 0 level = 0
data = (level,tokens.get(line[1],TOKEN_UNKNOWN),val,line[1]) data = (level,tokens.get(line[1],TOKEN_UNKNOWN),val,line[1],self.index)
if data[1] == TOKEN_CONT: if data[1] == TOKEN_CONT:
l = self.current_list[0] l = self.current_list[0]
self.current_list[0] = (l[0],l[1],l[2]+'\n'+data[2],l[3]) self.current_list[0] = (l[0],l[1],l[2]+'\n'+data[2],l[3],l[4])
elif data[1] == TOKEN_CONC: elif data[1] == TOKEN_CONC:
l = self.current_list[0] l = self.current_list[0]
if self.broken_conc: if self.broken_conc:
new_value = "%s %s" % (l[2],data[2]) new_value = "%s %s" % (l[2],data[2])
else: else:
new_value = l[2] + data[2] new_value = l[2] + data[2]
self.current_list[0] = (l[0],l[1],new_value,l[3]) self.current_list[0] = (l[0],l[1],new_value,l[3],l[4])
else: else:
self.current_list.insert(0,data) self.current_list.insert(0,data)
@ -403,7 +410,7 @@ class GedcomParser:
self.smap = {} self.smap = {}
self.note_map = note_map self.note_map = note_map
self.refn = {} self.refn = {}
self.added = {} self.added = set()
self.gedmap = GedcomInfoDB() self.gedmap = GedcomInfoDB()
self.gedsource = self.gedmap.get_from_source_tag('GEDCOM 5.5') self.gedsource = self.gedmap.get_from_source_tag('GEDCOM 5.5')
self.def_src = RelLib.Source() self.def_src = RelLib.Source()
@ -419,6 +426,12 @@ class GedcomParser:
self.sid2id = {} self.sid2id = {}
self.lid2id = {} self.lid2id = {}
self.fid2id = {} self.fid2id = {}
self.rid2id = {}
self.repo_func = {
TOKEN_NAME : self.func_repo_name,
TOKEN_ADDR : self.func_repo_addr,
}
self.name_func = { self.name_func = {
TOKEN_ALIA : self.func_name_alia, TOKEN_ALIA : self.func_name_alia,
@ -480,7 +493,6 @@ class GedcomParser:
self.lexer = Reader(filename) self.lexer = Reader(filename)
self.filename = filename self.filename = filename
self.index = 0
self.backoff = 0 self.backoff = 0
self.override = codeset self.override = codeset
@ -570,22 +582,20 @@ class GedcomParser:
# EOF ? # EOF ?
if next_line == None: if next_line == None:
self.index += 1
self.text = ""; self.text = "";
self.backoff = 0 self.backoff = 0
msg = _("Premature end of file at line %d.\n") % self.index msg = _("Premature end of file at line %d.\n") % next_line[4]
self.errmsg(msg) self.errmsg(msg)
self.error_count = self.error_count + 1 self.error_count = self.error_count + 1
self.groups = (-1, TOKEN_UNKNOWN, "","") self.groups = (-1, TOKEN_UNKNOWN, "","")
return self.groups return self.groups
self.groups = next_line self.groups = next_line
self.index += 1
self.backoff = 0 self.backoff = 0
return self.groups return self.groups
def barf(self,level): def barf(self,level):
msg = _("Line %d was not understood, so it was ignored.") % self.index msg = _("Line %d was not understood, so it was ignored.") % self.groups[4]
self.errmsg(msg) self.errmsg(msg)
self.error_count = self.error_count + 1 self.error_count = self.error_count + 1
self.ignore_sub_junk(level) self.ignore_sub_junk(level)
@ -603,9 +613,9 @@ class GedcomParser:
#self.trans.set_batch(not use_trans) #self.trans.set_batch(not use_trans)
self.db.disable_signals() self.db.disable_signals()
t = time.time() t = time.time()
self.index = 0
self.fam_count = 0 self.fam_count = 0
self.indi_count = 0 self.indi_count = 0
self.repo_count = 0
self.source_count = 0 self.source_count = 0
try: try:
self.parse_header() self.parse_header()
@ -739,7 +749,7 @@ class GedcomParser:
gid = matches[3] gid = matches[3]
gid = gid[1:-1] gid = gid[1:-1]
self.person = self.find_or_create_person(self.map_gid(gid)) self.person = self.find_or_create_person(self.map_gid(gid))
self.added[self.person.get_handle()] = 1 self.added.add(self.person.get_handle())
self.parse_individual(self.person) self.parse_individual(self.person)
if len(self.person.get_source_references()) == 0: if len(self.person.get_source_references()) == 0:
sref = RelLib.SourceRef() sref = RelLib.SourceRef()
@ -747,6 +757,14 @@ class GedcomParser:
self.person.add_source_reference(sref) self.person.add_source_reference(sref)
self.db.commit_person(self.person, self.trans) self.db.commit_person(self.person, self.trans)
del self.person del self.person
elif matches[2] == "REPO":
self.repo_count = self.repo_count + 1
self.repo = self.find_or_create_repository(matches[3][1:-1])
self.repo.set_type((RelLib.Repository.UNKNOWN,""))
self.added.add(self.repo.get_handle())
self.parse_repository(self.repo)
self.db.commit_repository(self.repo, self.trans)
del self.repo
elif matches[2] in ["SUBM","SUBN","REPO"]: elif matches[2] in ["SUBM","SUBN","REPO"]:
self.ignore_sub_junk(1) self.ignore_sub_junk(1)
elif matches[1] in (TOKEN_SUBM,TOKEN_SUBN,TOKEN_OBJE,TOKEN__EVENT_DEFN): elif matches[1] in (TOKEN_SUBM,TOKEN_SUBN,TOKEN_OBJE,TOKEN__EVENT_DEFN):
@ -813,6 +831,24 @@ class GedcomParser:
family.set_gramps_id(gramps_id) family.set_gramps_id(gramps_id)
return family return family
def find_or_create_repository(self,gramps_id):
repository = RelLib.Repository()
intid = self.rid2id.get(gramps_id)
if self.db.has_repository_handle(intid):
repository.unserialize(self.db.get_raw_repository_data(intid))
else:
intid = self.find_repository_handle(gramps_id)
repository.set_handle(intid)
repository.set_gramps_id(gramps_id)
return repository
def find_repository_handle(self,gramps_id):
intid = self.rid2id.get(gramps_id)
if not intid:
intid = create_id()
self.rid2id[gramps_id] = intid
return intid
def find_family_handle(self,gramps_id): def find_family_handle(self,gramps_id):
intid = self.fid2id.get(gramps_id) intid = self.fid2id.get(gramps_id)
if not intid: if not intid:
@ -867,7 +903,7 @@ class GedcomParser:
return place return place
def parse_cause(self,event,level): def parse_cause(self,event,level):
while 1: while True:
matches = self.get_next() matches = self.get_next()
if int(matches[0]) < level: if int(matches[0]) < level:
self.backup() self.backup()
@ -878,7 +914,7 @@ class GedcomParser:
self.barf(1) self.barf(1)
def parse_note_data(self,level): def parse_note_data(self,level):
while 1: while True:
matches = self.get_next() matches = self.get_next()
if int(matches[0]) < level: if int(matches[0]) < level:
self.backup() self.backup()
@ -894,7 +930,7 @@ class GedcomParser:
mrel = (RelLib.Person.CHILD_BIRTH,'') mrel = (RelLib.Person.CHILD_BIRTH,'')
frel = (RelLib.Person.CHILD_BIRTH,'') frel = (RelLib.Person.CHILD_BIRTH,'')
while 1: while True:
matches = self.get_next() matches = self.get_next()
if int(matches[0]) < level: if int(matches[0]) < level:
self.backup() self.backup()
@ -922,7 +958,7 @@ class GedcomParser:
def parse_family(self): def parse_family(self):
self.addr = None self.addr = None
note = "" note = ""
while 1: while True:
matches = self.get_next() matches = self.get_next()
if int(matches[0]) < 1: if int(matches[0]) < 1:
@ -972,7 +1008,7 @@ class GedcomParser:
self.db.commit_person(child, self.trans) self.db.commit_person(child, self.trans)
elif matches[1] == TOKEN_NCHI: elif matches[1] == TOKEN_NCHI:
a = RelLib.Attribute() a = RelLib.Attribute()
a.set_type("Number of Children") a.set_type((RelLib.Attribute.NUM_CHILD,''))
a.set_value(matches[2]) a.set_value(matches[2])
self.family.add_attribute(a) self.family.add_attribute(a)
elif matches[1] == TOKEN_SOUR: elif matches[1] == TOKEN_SOUR:
@ -996,11 +1032,14 @@ class GedcomParser:
try: try:
event.set_type((ged2fam[matches[3]],'')) event.set_type((ged2fam[matches[3]],''))
except: except:
val = ged2fam_custom.has_key(matches[3]) if ged2fam_custom.has_key(matches[3]):
if val: event.set_type((RelLib.Event.CUSTOM,ged2fam_custom[matches[3]]))
event.set_type((RelLib.Event.CUSTOM,val)) elif matches[3]:
event.set_type((RelLib.Event.CUSTOM,matches[3]))
else: else:
event.set_type((RelLib.Event.CUSTOM,matches[1])) event.set_type((RelLib.Event.UNKNOWN,''))
if matches[2] and not event.get_description():
event.set_description(matches[2])
self.parse_family_event(event,2) self.parse_family_event(event,2)
if event.get_type()[0] == RelLib.Event.MARRIAGE: if event.get_type()[0] == RelLib.Event.MARRIAGE:
self.family.set_relationship((RelLib.Family.MARRIED,'')) self.family.set_relationship((RelLib.Family.MARRIED,''))
@ -1029,7 +1068,7 @@ class GedcomParser:
return u"" return u""
else: else:
if old_note: if old_note:
note = "%s\n%s%s" % (old_note,matches[2]) note = "%s\n%s" % (old_note,matches[2])
else: else:
note = matches[2] note = matches[2]
task(note) task(note)
@ -1060,7 +1099,7 @@ class GedcomParser:
def parse_optional_note(self,level): def parse_optional_note(self,level):
note = "" note = ""
while 1: while True:
matches = self.get_next() matches = self.get_next()
if int(matches[0]) < level: if int(matches[0]) < level:
@ -1079,7 +1118,7 @@ class GedcomParser:
def parse_famc_type(self,level,person): def parse_famc_type(self,level,person):
ftype = (RelLib.Person.CHILD_BIRTH,'') ftype = (RelLib.Person.CHILD_BIRTH,'')
note = "" note = ""
while 1: while True:
matches = self.get_next() matches = self.get_next()
if int(matches[0]) < level: if int(matches[0]) < level:
@ -1155,7 +1194,7 @@ class GedcomParser:
filename = "" filename = ""
title = "" title = ""
note = "" note = ""
while 1: while True:
matches = self.get_next() matches = self.get_next()
if matches[1] == TOKEN_FORM: if matches[1] == TOKEN_FORM:
form = matches[2].lower() form = matches[2].lower()
@ -1193,7 +1232,7 @@ class GedcomParser:
def parse_residence(self,address,level): def parse_residence(self,address,level):
note = "" note = ""
while 1: while True:
matches = self.get_next() matches = self.get_next()
if int(matches[0]) < level: if int(matches[0]) < level:
@ -1223,7 +1262,7 @@ class GedcomParser:
def parse_address(self,address,level): def parse_address(self,address,level):
first = 0 first = 0
note = "" note = ""
while 1: while True:
matches = self.get_next() matches = self.get_next()
if int(matches[0]) < level: if int(matches[0]) < level:
if matches[1] == TOKEN_PHON: if matches[1] == TOKEN_PHON:
@ -1260,7 +1299,7 @@ class GedcomParser:
def parse_ord(self,lds_ord,level): def parse_ord(self,lds_ord,level):
note = "" note = ""
while 1: while True:
matches = self.get_next() matches = self.get_next()
if int(matches[0]) < level: if int(matches[0]) < level:
self.backup() self.backup()
@ -1294,7 +1333,7 @@ class GedcomParser:
def parse_person_event(self,event,level): def parse_person_event(self,event,level):
note = "" note = ""
while 1: while True:
matches = self.get_next() matches = self.get_next()
if int(matches[0]) < level: if int(matches[0]) < level:
if note: if note:
@ -1359,7 +1398,7 @@ class GedcomParser:
def parse_adopt_event(self,event,level): def parse_adopt_event(self,event,level):
note = "" note = ""
while 1: while True:
matches = self.get_next() matches = self.get_next()
if int(matches[0]) < level: if int(matches[0]) < level:
if note != "": if note != "":
@ -1405,7 +1444,7 @@ class GedcomParser:
def parse_adopt_famc(self,level): def parse_adopt_famc(self,level):
mrel = RelLib.Person.CHILD_ADOPTED mrel = RelLib.Person.CHILD_ADOPTED
frel = RelLib.Person.CHILD_ADOPTED frel = RelLib.Person.CHILD_ADOPTED
while 1: while True:
matches = self.get_next() matches = self.get_next()
if int(matches[0]) < level: if int(matches[0]) < level:
self.backup() self.backup()
@ -1421,7 +1460,7 @@ class GedcomParser:
def parse_person_attr(self,attr,level): def parse_person_attr(self,attr,level):
note = "" note = ""
while 1: while True:
matches = self.get_next() matches = self.get_next()
if int(matches[0]) < level: if int(matches[0]) < level:
self.backup() self.backup()
@ -1462,7 +1501,7 @@ class GedcomParser:
def parse_family_event(self,event,level): def parse_family_event(self,event,level):
note = "" note = ""
while 1: while True:
matches = self.get_next() matches = self.get_next()
if int(matches[0]) < level: if int(matches[0]) < level:
if note: if note:
@ -1472,9 +1511,6 @@ class GedcomParser:
elif matches[1] == TOKEN_TYPE: elif matches[1] == TOKEN_TYPE:
etype = event.get_type() etype = event.get_type()
if etype[0] == RelLib.Event.CUSTOM: if etype[0] == RelLib.Event.CUSTOM:
try:
event.set_type((ged2fam[matches[2]],''))
except:
event.set_type((RelLib.Event.CUSTOM,matches[2])) event.set_type((RelLib.Event.CUSTOM,matches[2]))
else: else:
note = 'Status = %s\n' % matches[2] note = 'Status = %s\n' % matches[2]
@ -1517,7 +1553,7 @@ class GedcomParser:
def parse_source_reference(self,source,level): def parse_source_reference(self,source,level):
"""Reads the data associated with a SOUR reference""" """Reads the data associated with a SOUR reference"""
note = "" note = ""
while 1: while True:
matches = self.get_next() matches = self.get_next()
if int(matches[0]) < level: if int(matches[0]) < level:
@ -1554,7 +1590,7 @@ class GedcomParser:
"""Parses the source data""" """Parses the source data"""
date = "" date = ""
note = "" note = ""
while 1: while True:
matches = self.get_next() matches = self.get_next()
if int(matches[0]) < level: if int(matches[0]) < level:
self.backup() self.backup()
@ -1586,11 +1622,10 @@ class GedcomParser:
line = self.lexer.read() line = self.lexer.read()
if line[1] != TOKEN_HEAD: if line[1] != TOKEN_HEAD:
raise Errors.GedcomError("%s is not a GEDCOM file" % self.filename) raise Errors.GedcomError("%s is not a GEDCOM file" % self.filename)
self.index += 1
def parse_header_source(self): def parse_header_source(self):
genby = "" genby = ""
while 1: while True:
matches = self.get_next() matches = self.get_next()
if int(matches[0]) < 1: if int(matches[0]) < 1:
self.backup() self.backup()
@ -1642,7 +1677,7 @@ class GedcomParser:
self.barf(2) self.barf(2)
def parse_ftw_schema(self,level): def parse_ftw_schema(self,level):
while 1: while True:
matches = self.get_next() matches = self.get_next()
if int(matches[0]) < level: if int(matches[0]) < level:
@ -1656,7 +1691,7 @@ class GedcomParser:
self.barf(2) self.barf(2)
def parse_ftw_indi_schema(self,level): def parse_ftw_indi_schema(self,level):
while 1: while True:
matches = self.get_next() matches = self.get_next()
if int(matches[0]) < level: if int(matches[0]) < level:
@ -1667,7 +1702,7 @@ class GedcomParser:
ged2gramps[matches[1]] = label ged2gramps[matches[1]] = label
def parse_label(self,level): def parse_label(self,level):
while 1: while True:
matches = self.get_next() matches = self.get_next()
if int(matches[0]) < level: if int(matches[0]) < level:
@ -1680,7 +1715,7 @@ class GedcomParser:
return None return None
def parse_ftw_fam_schema(self,level): def parse_ftw_fam_schema(self,level):
while 1: while True:
matches = self.get_next() matches = self.get_next()
if int(matches[0]) < level: if int(matches[0]) < level:
@ -1688,15 +1723,17 @@ class GedcomParser:
return return
else: else:
label = self.parse_label(level+1) label = self.parse_label(level+1)
ged2fam[matches[3]] = label ged2fam_custom[matches[3]] = label
return None return None
def ignore_sub_junk(self,level): def ignore_sub_junk(self,level):
while 1: while True:
matches = self.get_next() matches = self.get_next()
if int(matches[0]) < level: if int(matches[0]) < level:
self.backup() self.backup()
return return
# else:
# print self.lexer.current
return return
def ignore_change_data(self,level): def ignore_change_data(self,level):
@ -1707,7 +1744,7 @@ class GedcomParser:
self.backup() self.backup()
def parse_place_form(self,level): def parse_place_form(self,level):
while 1: while True:
matches = self.get_next() matches = self.get_next()
if int(matches[0]) < level: if int(matches[0]) < level:
@ -1718,7 +1755,7 @@ class GedcomParser:
def parse_date(self,level): def parse_date(self,level):
date = DateStruct() date = DateStruct()
while 1: while True:
matches = self.get_next() matches = self.get_next()
if int(matches[0]) < level: if int(matches[0]) < level:
@ -1812,7 +1849,7 @@ class GedcomParser:
prefix = self.db.iprefix prefix = self.db.iprefix
index = 0 index = 0
new_pmax = self.db.pmap_index new_pmax = self.db.pmap_index
for pid in self.added.keys(): for pid in self.added:
index = index + 1 index = index + 1
if self.refn.has_key(pid): if self.refn.has_key(pid):
val = self.refn[pid] val = self.refn[pid]
@ -1988,7 +2025,7 @@ class GedcomParser:
event_ref = RelLib.EventRef() event_ref = RelLib.EventRef()
event_ref.set_reference_handle(event.handle) event_ref.set_reference_handle(event.handle)
event_ref.set_rol((RelLib.EventRef.PRIMARY,'')) event_ref.set_role((RelLib.EventRef.PRIMARY,''))
state.person.add_event_ref(event_ref) state.person.add_event_ref(event_ref)
def func_person_deat(self,matches,state): def func_person_deat(self,matches,state):
@ -2018,7 +2055,7 @@ class GedcomParser:
(t,n) = event.get_type() (t,n) = event.get_type()
if t == RelLib.Event.CUSTOM and n in self.attrs: if t == RelLib.Event.CUSTOM and n in self.attrs:
attr = RelLib.Attribute() attr = RelLib.Attribute()
attr.set_type((self.gedattr[n],'')) attr.set_type((RelLib.Event.CUSTOM,self.gedattr[n]))
attr.set_value(event.get_description()) attr.set_value(event.get_description())
state.person.add_attribute(attr) state.person.add_attribute(attr)
else: else:
@ -2055,7 +2092,7 @@ class GedcomParser:
n = matches[3].strip() n = matches[3].strip()
if self.gedattr.has_key(n): if self.gedattr.has_key(n):
attr = RelLib.Attribute() attr = RelLib.Attribute()
attr.set_type((self.gedattr[n],'')) attr.set_type((RelLib.Event.CUSTOM,self.gedattr[n]))
attr.set_value(matches[2]) attr.set_value(matches[2])
state.person.add_attribute(attr) state.person.add_attribute(attr)
self.parse_person_attr(attr,2) self.parse_person_attr(attr,2)
@ -2166,6 +2203,60 @@ class GedcomParser:
sref = self.handle_source(matches,state.level+1) sref = self.handle_source(matches,state.level+1)
state.name.add_source_reference(sref) state.name.add_source_reference(sref)
def parse_repository(self,repo):
state = CurrentState()
state.repo = repo
while True:
matches = self.get_next()
if int(matches[0]) < 1:
self.backup()
if state.get_text():
state.repo.set_note(state.get_text())
return
else:
func = self.repo_func.get(matches[1],self.skip_record)
func(matches,state)
def func_repo_name(self,matches,state):
state.repo.set_name(matches[2])
def func_repo_addr(self,matches,state):
addr = RelLib.Address()
matched = False
match = addr_re.match(matches[2])
if match:
groups = match.groups()
addr.set_street(groups[0].strip())
addr.set_city(groups[2].strip())
addr.set_state(groups[3].strip())
addr.set_postal_code(groups[4].strip())
addr.set_country(groups[5].strip())
matched = True
match = addr2_re.match(matches[2])
if match:
groups = match.groups()
addr.set_street(groups[0].strip())
addr.set_city(groups[2].strip())
addr.set_state(groups[3].strip())
addr.set_postal_code(groups[4].strip())
matched = True
match = addr3_re.match(matches[2])
if match:
groups = match.groups()
addr.set_street(groups[0].strip())
addr.set_city(groups[2].strip())
addr.set_state(groups[3].strip())
matched = True
if not matched:
addr.set_street(matches[2])
state.repo.add_address(addr)
def skip_record(self,matches,state): def skip_record(self,matches,state):
self.ignore_sub_junk(2) self.ignore_sub_junk(2)

View File

@ -305,6 +305,8 @@ class Event(PrimaryObject,SourceNote,MediaBase,DateBase,PlaceBase):
the_type = (the_type,'') the_type = (the_type,'')
else: else:
the_type = (Event.CUSTOM,the_type) the_type = (Event.CUSTOM,the_type)
assert(type(the_type[0]) == int)
assert(type(the_type[1]) == unicode or type(the_type[1]) == str)
self.type = the_type self.type = the_type
def get_type(self): def get_type(self):

View File

@ -53,6 +53,7 @@ column_names = [
_('ID'), _('ID'),
_('Type'), _('Type'),
_('Home URL'), _('Home URL'),
_('Street'),
_('ZIP/Postal Code'), _('ZIP/Postal Code'),
_('City'), _('City'),
_('County'), _('County'),