2
0
mirror of https://expo.survex.com/repositories/troggle/.git synced 2024-11-30 05:41:52 +00:00
troggle/parsers/survex.py
martin speleo c132477f80 [svn] Added cavedescription and new subcave.
Changed parsers/survex to read *title into subcave
2009-07-04 00:28:28 +01:00

157 lines
6.5 KiB
Python

import troggle.settings as settings
import troggle.core.models as models
from troggle.parsers.people import GetPersonExpeditionNameLookup
import re
import os
roles = {"Insts": "Insts",
"insts": "Insts",
"Instruments": "Insts",
"instruments": "Insts",
"Inst": "Insts",
"inst": "Insts",
"dog": "Other",
"Dog": "Other",
"other": "Other",
"Other": "Other",
"Notes": "Notes",
"notes": "notes",
"pics": "Pics",
"Pics": "Pics",
"Tape": "Tape",
"tape": "Tape"}
re_include_extension = re.compile(r"^\s*\*include\s+([^\s]*).svx\s*$", re.IGNORECASE)
re_include_no_extension = re.compile(r"^\s*\*include\s+([^\s]*)\s*$", re.IGNORECASE)
flags = {"begin": re.compile(r"^\s*\*begin\s+(.*?)\s*$", re.IGNORECASE),
"end": re.compile(r"^\s*\*end\s+(.*?)\s*$", re.IGNORECASE),
"date": re.compile(r"^\s*\*date\s+(.*?)\s*$", re.IGNORECASE),
"team": re.compile(r"^\s*\*team\s+(.*?)\s*$", re.IGNORECASE),
"title": re.compile(r"^\s*\*title\s+(.*?)\s*$", re.IGNORECASE)}
def fileIterator(directory, filename):
survex_file = os.path.join(directory, filename + ".svx")
try:
f = open(os.path.join(settings.SURVEX_DATA, survex_file), "rb")
except:
f = open(os.path.join(settings.SURVEX_DATA, survex_file).lower(), "rb")
char = 0
for line in f.readlines():
line = unicode(line, "latin1")
include_extension = re_include_extension.match(line)
include_no_extension = re_include_no_extension.match(line)
def a(include):
link = re.split(r"/|\\", include)
return fileIterator(os.path.join(directory, *link[:-1]), link[-1])
if include_extension:
for sf, c, l in a(include_extension.groups()[0]):
yield sf, c, l
elif include_no_extension:
for sf, c, l in a(include_no_extension.groups()[0]):
yield sf, c, l
else:
yield survex_file, char, line
char = char + len(line)
def make_model(name, parent, iter_lines, sf, c, l):
m = models.SurvexBlock(name = name, begin_file = sf, begin_char = c, text = l)
m.survexpath = m.name
if parent:
m.parent = parent
m.survexpath = m.parent.survexpath + "." + m.name
m.save()
# horrible local function
def saveEnd(survex_file, count):
if m.start_year and team:
try:
exp = models.Expedition.objects.get(year = str(m.start_year))
for file_, (role, names) in team:
if names.strip("\t").strip(" ") == "both" or names.strip("\t").strip(" ") == "Both":
names = reduce(lambda x, y: x + u" & " + y,
[names for file_, (role, names) in team
if names.strip("\t").strip(" ") != "both"
and names.strip("\t").strip(" ") != "Both"])
for name in re.split("&|/|\+|,|;", names):
sname = name.strip(". ").lower()
try:
personexpedition = GetPersonExpeditionNameLookup(exp).get(sname)
if personexpedition:
models.PersonRole(personexpedition = personexpedition,
person = personexpedition.person,
survex_block = m,
role = models.Role.objects.get(name = roles[role])).save()
else:
print ("no person", exp, sname, role)
except AttributeError:
print ("Person not found: " + name + " in " + file_ + " " + role).encode('ascii', 'xmlcharrefreplace')
except AssertionError, inst:
print (unicode(inst) + ": " + unicode(file_year[0])).encode('ascii', 'xmlcharrefreplace')
#except models.Expedition.DoesNotExist:
# print "Expo"+str(file_year[1]).encode('ascii', 'xmlcharrefreplace')
m.end_file = survex_file
m.end_char = count
if m.start_day:
m.date = "%04d-%02d-%02d" % (int(m.start_year), int(m.start_month), int(m.start_day))
m.save()
team = []
file_year = None
for survex_file, count, line in iter_lines:
#Dictionary compreshension
res = dict([(key, regex.match(line.split(";")[0])) for key, regex in flags.iteritems()])
if res["begin"]:
make_model(res["begin"].groups()[0], m, iter_lines, survex_file, count, line)
else:
m.text = m.text + line
if res["end"]:
saveEnd(survex_file, count)
assert (res["end"].groups()[0]).lower() == (name).lower()
return None
elif res["date"]:
datere = re.match("(\d+)(?:\.(\d+))?(?:\.(\d+))?(?:-(\d+))?(?:\.(\d+))?(?:\.(\d+))?",
res["date"].groups()[0])
if datere is not None:
startYear, startMonth, startDay, endYear, endMonth, endDay = datere.groups()
m.start_year = startYear
m.start_month = startMonth
m.start_day = startDay
m.end_year = endYear
m.end_month = endMonth
m.end_day = endDay
file_year = survex_file, startYear
elif res["team"]:
h = re.match("((?:[Ii]nst(?:s|ruments)?)|(?:[Pp]ics)|(?:[Tt]ape)|(?:[Nn]otes)|(?:[Oo]ther))\s*(.*)",
res["team"].groups()[0])
if h:
team.append((survex_file, h.groups()))
else:
print ("Role not found: " + line + " in: " + sf).encode('ascii', 'xmlcharrefreplace')
elif res["title"]:
nsb, success = models.NewSubCave.objects.get_or_create(name = res["title"].groups()[0])
m.text = m.text + line
saveEnd(survex_file, count)
#def LoadSurvexBlocks():
# survex_file = os.path.join(directory, filename + ".svx")
# f = open(os.path.join(settings.SURVEX_DATA, survex_file), "rb")
def LoadAllSurvexBlocks():
models.Role.objects.all().delete()
models.SurvexBlock.objects.all().delete()
for role in ["Insts", "Notes", "Pics", "Tape", "Other"]:
models.Role(name = role).save()
filename = "all"
make_model("all", None, fileIterator("", filename), filename, 0, "")