Código fuente para catatom2osm.csvtools
"""CSV related help functions."""
import csv
import io
import os
from catatom2osm import config
from catatom2osm.exceptions import CatIOError
[documentos]def dict2csv(csv_path, a_dict, sort=None):
"""
Write a dictionary to a csv file.
Optinally sorted by key (sort=0) or value (sort=1)
"""
with io.open(csv_path, "w", encoding=config.encoding) as csv_file:
dictitems = list(a_dict.items())
if sort in [0, 1]:
dictitems.sort(key=lambda x: x[sort])
for (k, v) in dictitems:
row = v
if isinstance(v, (list, tuple, set)):
row = config.delimiter.join(v)
csv_file.write("%s%s%s%s" % (k, config.delimiter, row, "\n"))
[documentos]def csv2dict(csv_path, a_dict=None, exists=False, single=True):
"""Read a dictionary from a csv file."""
a_dict = {} if a_dict is None else a_dict
msg = _("Failed to load CSV file '%s'") % os.path.basename(csv_path)
if os.path.exists(csv_path):
with open(csv_path) as csv_file:
csv_reader = csv.reader(csv_file, delimiter=str(config.delimiter))
for row in csv_reader:
if len(row) < 2:
raise CatIOError(msg)
a_dict[row[0]] = row[1] if single else row[1:]
elif exists:
raise CatIOError(msg)
return a_dict
[documentos]def filter(csv_path, *args, query=lambda row, args: True, stop=False):
"""
Return csv rows filtered using query.
Args:
args: aditional arguments for query function
query (func): function with args row and wargs that returns
a boolean deciding if each row will be included or not
stop (bool): Stop at first match or get all
"""
output = []
with open(csv_path) as csv_file:
csv_reader = csv.reader(csv_file, delimiter=str(config.delimiter))
for row in csv_reader:
if query(row, args):
if stop:
return row
output.append(row)
return output
[documentos]def search(csv_path, *args, query=lambda row, args: True):
"""
Return first matched row in csv.
Args:
args: aditional arguments for query function
query (func): function with args row and wargs that returns
a boolean deciding if each row will be included or not
"""
return filter(csv_path, *args, query=query, stop=True)
[documentos]def get_key(csv_path, key):
"""Get a row given first column value."""
def query(row, args):
return row[0] == args[0]
return search(csv_path, key, query=query)
[documentos]def startswith(csv_path, key):
"""Get rows which first column starts with key."""
def query(row, args):
return row[0].startswith(args[0])
return filter(csv_path, key, query=query)