Mercurial > dive4elements > river
view flys-backend/contrib/import-kms.py @ 1243:3ebc0a7d6793
Removed all tables for different datasets at HYdrologie/Hydr.Grenzen/Linien/*
flys-backend/trunk@2716 c6561f87-3c4e-4783-a992-168aeb5c3f6f
author | Hans Plum <hans.plum@intevation.de> |
---|---|
date | Tue, 13 Sep 2011 10:34:44 +0000 |
parents | 80669241956c |
children |
line wrap: on
line source
#!/usr/bin/env python import sys import logging import re import os import sqlite3 as db import locale import codecs from optparse import OptionParser log = logging.getLogger(__name__) log.setLevel(logging.WARNING) log.addHandler(logging.StreamHandler(sys.stderr)) RANGE = re.compile("([^#]*)#(.*)") DEFAULT_DATABASE = "flys.db" SQL_NEXT_ID = "SELECT coalesce(max(id), -1) + 1 FROM %s" SQL_SELECT_ID = "SELECT id FROM %s WHERE %s = ?" SQL_INSERT_ID = "INSERT INTO %s (id, %s) VALUES (?, ?)" SQL_SELECT_RANGE_ID = """ SELECT id FROM ranges WHERE river_id = ? AND a = ? AND b = ? """ SQL_INSERT_RANGE_ID = """ INSERT INTO ranges (id, river_id, a, b) VALUES (?, ?, ?, ?) """ SQL_SELECT_ANNOTATION_ID = """ SELECT id FROM annotations WHERE range_id = ? AND attribute_id = ? AND position_id = ? """ SQL_INSERT_ANNOTATION_ID = """ INSERT INTO annotations (id, range_id, attribute_id, position_id) VALUES (?, ?, ?, ?) """ def encode(s): try: return unicode(s, "latin-1") except UnicodeDecodeError: return unicode.encode(s, locale.getpreferredencoding()) class hashabledict(dict): def __key(self): return tuple((k, self[k]) for k in sorted(self)) def __hash__(self): return hash(self.__key()) def __eq__(self, other): return self.__key() == other.__key() def cache(f): def func(*args, **kw): key = (args, hashabledict(kw)) try: return f.__cache__[key] except KeyError: value = f(*args, **kw) f.__cache__[key] = value return value f.__cache__ = {} return func NEXT_IDS = {} def next_id(cur, relation): idx = NEXT_IDS.get(relation) if idx is None: cur.execute(SQL_NEXT_ID % relation) idx = cur.fetchone()[0] NEXT_IDS[relation] = idx + 1 return idx def get_id(cur, relation, attribute, value): select_stmt = SQL_SELECT_ID % (relation, attribute) #log.debug(select_stmt) cur.execute(select_stmt, (value,)) row = cur.fetchone() if row: return row[0] idx = next_id(cur, relation) insert_stmnt = SQL_INSERT_ID % (relation, attribute) #log.debug(insert_stmnt) cur.execute(insert_stmnt, (idx, value)) cur.connection.commit() log.debug("insert %s '%s' id: '%d'" % (relation, value, idx)) return idx #@cache def get_river_id(cur, name): return get_id(cur, "rivers", "name", name) #@cache def get_attribute_id(cur, value): return get_id(cur, "attributes", "value", value) #@cache def get_position_id(cur, value): return get_id(cur, "positions", "value", value) #@cache def get_range_id(cur, river_id, a, b): cur.execute(SQL_SELECT_RANGE_ID, (river_id, a, b)) row = cur.fetchone() if row: return row[0] idx = next_id(cur, "ranges") cur.execute(SQL_INSERT_RANGE_ID, (idx, river_id, a, b)) cur.connection.commit() return idx #@cache def get_annotation_id(cur, range_id, attribute_id, position_id): cur.execute(SQL_SELECT_ANNOTATION_ID, ( range_id, attribute_id, position_id)) row = cur.fetchone() if row: return row[0] idx = next_id(cur, "annotations") cur.execute(SQL_INSERT_ANNOTATION_ID, ( idx, range_id, attribute_id, position_id)) cur.connection.commit() return idx def files(root, accept=lambda x: True): if os.path.isfile(root): if accept(root): yield root elif os.path.isdir(root): stack = [ root ] while stack: cur = stack.pop() for f in os.listdir(cur): p = os.path.join(cur, f) if os.path.isdir(p): stack.append(p) elif os.path.isfile(p) and accept(p): yield p def feed_km(cur, river_id, km_file): log.info("processing: %s" % km_file) for line in codecs.open(km_file, "r", "latin-1"): line = line.strip() if not line or line.startswith('*'): continue parts = [x.strip() for x in line.split(';')] if len(parts) < 3: log.error("cannot process: '%s'" % line) continue m = RANGE.match(parts[2]) try: if m: x = [float(x.replace(",", ".")) for x in m.groups()] a, b = min(x), max(x) if a == b: b = None else: a, b = float(parts[2].replace(",", ".")), None except ValueError: log.error("cannot process: '%s'" % line) continue attribute = parts[0] position = parts[1] attribute_id = get_attribute_id(cur, attribute) if attribute else None position_id = get_position_id(cur, position) if position else None range_id = get_range_id(cur, river_id, a, b) get_annotation_id(cur, range_id, attribute_id, position_id) def main(): usage = "usage: %prog [options] river km-file ..." parser = OptionParser(usage=usage) parser.add_option( "-v", "--verbose", action="store_true", dest="verbose", help="verbose output") parser.add_option( "-r", "--recursive", action="store_true", dest="recursive", default=False, help="recursive") parser.add_option( "-d", "--database", action="store", dest="database", help="database to connect with", default=DEFAULT_DATABASE) options, args = parser.parse_args() if options.verbose: log.setLevel(logging.INFO) if len(args) < 1: log.error("missing river argument") sys.exit(1) river = unicode(args[0], locale.getpreferredencoding()) with db.connect(options.database) as con: cur = con.cursor() river_id = get_river_id(cur, river) for arg in args[1:]: if options.recursive: for km_file in files( arg, lambda x: x.lower().endswith(".km")): feed_km(cur, river_id, km_file) else: feed_km(cur, river_id, arg) if __name__ == '__main__': main()