unnecessary list() added during py3 port
instead of recursive bytes2unicode use a proper JSONEncoder to conver bytes->unicode for json.dumps() removing excessive isinstance(data, bytes) checks py3: / -> // and list() around .items() that gets modified in loop moved lbrynet.undecorated to where its actually used and hopefully we can delete it eventually removed build/upload_assets.py, travis can do all this now
This commit is contained in:
parent
f061ca2b15
commit
10b34d6b33
18 changed files with 85 additions and 254 deletions
|
@ -124,7 +124,8 @@ disable=
|
||||||
keyword-arg-before-vararg,
|
keyword-arg-before-vararg,
|
||||||
assignment-from-no-return,
|
assignment-from-no-return,
|
||||||
useless-return,
|
useless-return,
|
||||||
assignment-from-none
|
assignment-from-none,
|
||||||
|
stop-iteration-return
|
||||||
|
|
||||||
|
|
||||||
[REPORTS]
|
[REPORTS]
|
||||||
|
@ -389,7 +390,7 @@ int-import-graph=
|
||||||
[DESIGN]
|
[DESIGN]
|
||||||
|
|
||||||
# Maximum number of arguments for function / method
|
# Maximum number of arguments for function / method
|
||||||
max-args=5
|
max-args=10
|
||||||
|
|
||||||
# Argument names that match this expression will be ignored. Default to name
|
# Argument names that match this expression will be ignored. Default to name
|
||||||
# with leading underscore
|
# with leading underscore
|
||||||
|
|
|
@ -149,7 +149,7 @@ class BlobFile:
|
||||||
def writer_finished(self, writer, err=None):
|
def writer_finished(self, writer, err=None):
|
||||||
def fire_finished_deferred():
|
def fire_finished_deferred():
|
||||||
self._verified = True
|
self._verified = True
|
||||||
for p, (w, finished_deferred) in self.writers.items():
|
for p, (w, finished_deferred) in list(self.writers.items()):
|
||||||
if w == writer:
|
if w == writer:
|
||||||
del self.writers[p]
|
del self.writers[p]
|
||||||
finished_deferred.callback(self)
|
finished_deferred.callback(self)
|
||||||
|
|
|
@ -46,8 +46,6 @@ class BlobFileCreator:
|
||||||
def write(self, data):
|
def write(self, data):
|
||||||
if not self._is_open:
|
if not self._is_open:
|
||||||
raise IOError
|
raise IOError
|
||||||
if not isinstance(data, bytes):
|
|
||||||
data = data.encode()
|
|
||||||
self._hashsum.update(data)
|
self._hashsum.update(data)
|
||||||
self.len_so_far += len(data)
|
self.len_so_far += len(data)
|
||||||
self.buffer.write(data)
|
self.buffer.write(data)
|
||||||
|
|
|
@ -12,6 +12,13 @@ from lbrynet.core.HTTPBlobDownloader import HTTPBlobDownloader
|
||||||
log = logging.getLogger(__name__)
|
log = logging.getLogger(__name__)
|
||||||
|
|
||||||
|
|
||||||
|
class JSONBytesEncoder(json.JSONEncoder):
|
||||||
|
def default(self, obj): # pylint: disable=E0202
|
||||||
|
if isinstance(obj, bytes):
|
||||||
|
return obj.decode()
|
||||||
|
return super().default(obj)
|
||||||
|
|
||||||
|
|
||||||
class StreamDescriptorReader:
|
class StreamDescriptorReader:
|
||||||
"""Classes which derive from this class read a stream descriptor file return
|
"""Classes which derive from this class read a stream descriptor file return
|
||||||
a dictionary containing the fields in the file"""
|
a dictionary containing the fields in the file"""
|
||||||
|
@ -66,16 +73,6 @@ class BlobStreamDescriptorReader(StreamDescriptorReader):
|
||||||
return threads.deferToThread(get_data)
|
return threads.deferToThread(get_data)
|
||||||
|
|
||||||
|
|
||||||
def bytes2unicode(value):
|
|
||||||
if isinstance(value, bytes):
|
|
||||||
return value.decode()
|
|
||||||
elif isinstance(value, (list, tuple)):
|
|
||||||
return [bytes2unicode(v) for v in value]
|
|
||||||
elif isinstance(value, dict):
|
|
||||||
return {key: bytes2unicode(v) for key, v in value.items()}
|
|
||||||
return value
|
|
||||||
|
|
||||||
|
|
||||||
class StreamDescriptorWriter:
|
class StreamDescriptorWriter:
|
||||||
"""Classes which derive from this class write fields from a dictionary
|
"""Classes which derive from this class write fields from a dictionary
|
||||||
of fields to a stream descriptor"""
|
of fields to a stream descriptor"""
|
||||||
|
@ -83,7 +80,9 @@ class StreamDescriptorWriter:
|
||||||
pass
|
pass
|
||||||
|
|
||||||
def create_descriptor(self, sd_info):
|
def create_descriptor(self, sd_info):
|
||||||
return self._write_stream_descriptor(json.dumps(bytes2unicode(sd_info), sort_keys=True))
|
return self._write_stream_descriptor(
|
||||||
|
json.dumps(sd_info, sort_keys=True, cls=JSONBytesEncoder).encode()
|
||||||
|
)
|
||||||
|
|
||||||
def _write_stream_descriptor(self, raw_data):
|
def _write_stream_descriptor(self, raw_data):
|
||||||
"""This method must be overridden by subclasses to write raw data to
|
"""This method must be overridden by subclasses to write raw data to
|
||||||
|
|
|
@ -68,14 +68,14 @@ class StreamBlobDecryptor:
|
||||||
|
|
||||||
def write_bytes():
|
def write_bytes():
|
||||||
if self.len_read < self.length:
|
if self.len_read < self.length:
|
||||||
num_bytes_to_decrypt = greatest_multiple(len(self.buff), (AES.block_size / 8))
|
num_bytes_to_decrypt = greatest_multiple(len(self.buff), (AES.block_size // 8))
|
||||||
data_to_decrypt, self.buff = split(self.buff, num_bytes_to_decrypt)
|
data_to_decrypt, self.buff = split(self.buff, num_bytes_to_decrypt)
|
||||||
write_func(self.cipher.update(data_to_decrypt))
|
write_func(self.cipher.update(data_to_decrypt))
|
||||||
|
|
||||||
def finish_decrypt():
|
def finish_decrypt():
|
||||||
bytes_left = len(self.buff) % (AES.block_size / 8)
|
bytes_left = len(self.buff) % (AES.block_size // 8)
|
||||||
if bytes_left != 0:
|
if bytes_left != 0:
|
||||||
log.warning(self.buff[-1 * (AES.block_size / 8):].encode('hex'))
|
log.warning(self.buff[-1 * (AES.block_size // 8):].encode('hex'))
|
||||||
raise Exception("blob %s has incorrect padding: %i bytes left" %
|
raise Exception("blob %s has incorrect padding: %i bytes left" %
|
||||||
(self.blob.blob_hash, bytes_left))
|
(self.blob.blob_hash, bytes_left))
|
||||||
data_to_decrypt, self.buff = self.buff, b''
|
data_to_decrypt, self.buff = self.buff, b''
|
||||||
|
@ -128,8 +128,6 @@ class CryptStreamBlobMaker:
|
||||||
max bytes are written. num_bytes_to_write is the number
|
max bytes are written. num_bytes_to_write is the number
|
||||||
of bytes that will be written from data in this call
|
of bytes that will be written from data in this call
|
||||||
"""
|
"""
|
||||||
if not isinstance(data, bytes):
|
|
||||||
data = data.encode()
|
|
||||||
max_bytes_to_write = MAX_BLOB_SIZE - self.length - 1
|
max_bytes_to_write = MAX_BLOB_SIZE - self.length - 1
|
||||||
done = False
|
done = False
|
||||||
if max_bytes_to_write <= len(data):
|
if max_bytes_to_write <= len(data):
|
||||||
|
|
|
@ -102,12 +102,6 @@ class CryptStreamCreator:
|
||||||
while 1:
|
while 1:
|
||||||
yield os.urandom(AES.block_size // 8)
|
yield os.urandom(AES.block_size // 8)
|
||||||
|
|
||||||
def get_next_iv(self):
|
|
||||||
iv = next(self.iv_generator)
|
|
||||||
if not isinstance(iv, bytes):
|
|
||||||
return iv.encode()
|
|
||||||
return iv
|
|
||||||
|
|
||||||
def setup(self):
|
def setup(self):
|
||||||
"""Create the symmetric key if it wasn't provided"""
|
"""Create the symmetric key if it wasn't provided"""
|
||||||
|
|
||||||
|
@ -127,7 +121,7 @@ class CryptStreamCreator:
|
||||||
|
|
||||||
yield defer.DeferredList(self.finished_deferreds)
|
yield defer.DeferredList(self.finished_deferreds)
|
||||||
self.blob_count += 1
|
self.blob_count += 1
|
||||||
iv = self.get_next_iv()
|
iv = next(self.iv_generator)
|
||||||
final_blob = self._get_blob_maker(iv, self.blob_manager.get_blob_creator())
|
final_blob = self._get_blob_maker(iv, self.blob_manager.get_blob_creator())
|
||||||
stream_terminator = yield final_blob.close()
|
stream_terminator = yield final_blob.close()
|
||||||
terminator_info = yield self._blob_finished(stream_terminator)
|
terminator_info = yield self._blob_finished(stream_terminator)
|
||||||
|
@ -138,7 +132,7 @@ class CryptStreamCreator:
|
||||||
if self.current_blob is None:
|
if self.current_blob is None:
|
||||||
self.next_blob_creator = self.blob_manager.get_blob_creator()
|
self.next_blob_creator = self.blob_manager.get_blob_creator()
|
||||||
self.blob_count += 1
|
self.blob_count += 1
|
||||||
iv = self.get_next_iv()
|
iv = next(self.iv_generator)
|
||||||
self.current_blob = self._get_blob_maker(iv, self.next_blob_creator)
|
self.current_blob = self._get_blob_maker(iv, self.next_blob_creator)
|
||||||
done, num_bytes_written = self.current_blob.write(data)
|
done, num_bytes_written = self.current_blob.write(data)
|
||||||
data = data[num_bytes_written:]
|
data = data[num_bytes_written:]
|
||||||
|
|
|
@ -19,8 +19,8 @@ from lbrynet.core import utils
|
||||||
from lbrynet.core.Error import ComponentsNotStarted, ComponentStartConditionNotMet
|
from lbrynet.core.Error import ComponentsNotStarted, ComponentStartConditionNotMet
|
||||||
from lbrynet.core.looping_call_manager import LoopingCallManager
|
from lbrynet.core.looping_call_manager import LoopingCallManager
|
||||||
from lbrynet.daemon.ComponentManager import ComponentManager
|
from lbrynet.daemon.ComponentManager import ComponentManager
|
||||||
from lbrynet.daemon.auth.util import APIKey, get_auth_message, LBRY_SECRET
|
from .util import APIKey, get_auth_message, LBRY_SECRET
|
||||||
from lbrynet.undecorated import undecorated
|
from .undecorated import undecorated
|
||||||
from .factory import AuthJSONRPCResource
|
from .factory import AuthJSONRPCResource
|
||||||
log = logging.getLogger(__name__)
|
log = logging.getLogger(__name__)
|
||||||
|
|
||||||
|
|
|
@ -684,11 +684,8 @@ class SQLiteStorage(WalletDatabase):
|
||||||
).fetchone()
|
).fetchone()
|
||||||
if not known_sd_hash:
|
if not known_sd_hash:
|
||||||
raise Exception("stream not found")
|
raise Exception("stream not found")
|
||||||
known_sd_hash = known_sd_hash[0]
|
|
||||||
if not isinstance(known_sd_hash, bytes):
|
|
||||||
known_sd_hash = known_sd_hash.encode()
|
|
||||||
# check the claim contains the same sd hash
|
# check the claim contains the same sd hash
|
||||||
if known_sd_hash != claim.source_hash:
|
if known_sd_hash[0].encode() != claim.source_hash:
|
||||||
raise Exception("stream mismatch")
|
raise Exception("stream mismatch")
|
||||||
|
|
||||||
# if there is a current claim associated to the file, check that the new claim is an update to it
|
# if there is a current claim associated to the file, check that the new claim is an update to it
|
||||||
|
|
|
@ -7,7 +7,7 @@ from lbrynet.dht import constants
|
||||||
|
|
||||||
def is_valid_ipv4(address):
|
def is_valid_ipv4(address):
|
||||||
try:
|
try:
|
||||||
ip = ipaddress.ip_address(address.encode().decode()) # this needs to be unicode, thus re-encode-able
|
ip = ipaddress.ip_address(address)
|
||||||
return ip.version == 4
|
return ip.version == 4
|
||||||
except ipaddress.AddressValueError:
|
except ipaddress.AddressValueError:
|
||||||
return False
|
return False
|
||||||
|
@ -22,8 +22,8 @@ class _Contact:
|
||||||
|
|
||||||
def __init__(self, contactManager, id, ipAddress, udpPort, networkProtocol, firstComm):
|
def __init__(self, contactManager, id, ipAddress, udpPort, networkProtocol, firstComm):
|
||||||
if id is not None:
|
if id is not None:
|
||||||
if not len(id) == constants.key_bits / 8:
|
if not len(id) == constants.key_bits // 8:
|
||||||
raise ValueError("invalid node id: %s" % hexlify(id.encode()))
|
raise ValueError("invalid node id: {}".format(hexlify(id).decode()))
|
||||||
if not 0 <= udpPort <= 65536:
|
if not 0 <= udpPort <= 65536:
|
||||||
raise ValueError("invalid port")
|
raise ValueError("invalid port")
|
||||||
if not is_valid_ipv4(ipAddress):
|
if not is_valid_ipv4(ipAddress):
|
||||||
|
|
|
@ -1,26 +1,23 @@
|
||||||
from binascii import hexlify
|
from binascii import hexlify
|
||||||
|
|
||||||
from lbrynet.dht import constants
|
from lbrynet.dht import constants
|
||||||
import sys
|
|
||||||
if sys.version_info > (3,):
|
|
||||||
long = int
|
|
||||||
|
|
||||||
|
|
||||||
class Distance:
|
class Distance:
|
||||||
"""Calculate the XOR result between two string variables.
|
"""Calculate the XOR result between two string variables.
|
||||||
|
|
||||||
Frequently we re-use one of the points so as an optimization
|
Frequently we re-use one of the points so as an optimization
|
||||||
we pre-calculate the long value of that point.
|
we pre-calculate the value of that point.
|
||||||
"""
|
"""
|
||||||
|
|
||||||
def __init__(self, key):
|
def __init__(self, key):
|
||||||
if len(key) != constants.key_bits / 8:
|
if len(key) != constants.key_bits // 8:
|
||||||
raise ValueError("invalid key length: %i" % len(key))
|
raise ValueError("invalid key length: %i" % len(key))
|
||||||
self.key = key
|
self.key = key
|
||||||
self.val_key_one = long(hexlify(key), 16)
|
self.val_key_one = int(hexlify(key), 16)
|
||||||
|
|
||||||
def __call__(self, key_two):
|
def __call__(self, key_two):
|
||||||
val_key_two = long(hexlify(key_two), 16)
|
val_key_two = int(hexlify(key_two), 16)
|
||||||
return self.val_key_one ^ val_key_two
|
return self.val_key_one ^ val_key_two
|
||||||
|
|
||||||
def is_closer(self, a, b):
|
def is_closer(self, a, b):
|
||||||
|
|
|
@ -4,9 +4,6 @@ from binascii import hexlify
|
||||||
from . import constants
|
from . import constants
|
||||||
from .distance import Distance
|
from .distance import Distance
|
||||||
from .error import BucketFull
|
from .error import BucketFull
|
||||||
import sys
|
|
||||||
if sys.version_info > (3,):
|
|
||||||
long = int
|
|
||||||
|
|
||||||
log = logging.getLogger(__name__)
|
log = logging.getLogger(__name__)
|
||||||
|
|
||||||
|
@ -141,7 +138,7 @@ class KBucket:
|
||||||
@rtype: bool
|
@rtype: bool
|
||||||
"""
|
"""
|
||||||
if isinstance(key, bytes):
|
if isinstance(key, bytes):
|
||||||
key = long(hexlify(key), 16)
|
key = int(hexlify(key), 16)
|
||||||
return self.rangeMin <= key < self.rangeMax
|
return self.rangeMin <= key < self.rangeMax
|
||||||
|
|
||||||
def __len__(self):
|
def __len__(self):
|
||||||
|
|
|
@ -17,7 +17,7 @@ class Message:
|
||||||
def __init__(self, rpcID, nodeID):
|
def __init__(self, rpcID, nodeID):
|
||||||
if len(rpcID) != constants.rpc_id_length:
|
if len(rpcID) != constants.rpc_id_length:
|
||||||
raise ValueError("invalid rpc id: %i bytes (expected 20)" % len(rpcID))
|
raise ValueError("invalid rpc id: %i bytes (expected 20)" % len(rpcID))
|
||||||
if len(nodeID) != constants.key_bits / 8:
|
if len(nodeID) != constants.key_bits // 8:
|
||||||
raise ValueError("invalid node id: %i bytes (expected 48)" % len(nodeID))
|
raise ValueError("invalid node id: %i bytes (expected 48)" % len(nodeID))
|
||||||
self.id = rpcID
|
self.id = rpcID
|
||||||
self.nodeID = nodeID
|
self.nodeID = nodeID
|
||||||
|
|
|
@ -1,11 +1,3 @@
|
||||||
#!/usr/bin/env python
|
|
||||||
#
|
|
||||||
# This library is free software, distributed under the terms of
|
|
||||||
# the GNU Lesser General Public License Version 3, or any later version.
|
|
||||||
# See the COPYING file included in this archive
|
|
||||||
#
|
|
||||||
# The docstrings in this module contain epytext markup; API documentation
|
|
||||||
# may be created by processing this file with epydoc: http://epydoc.sf.net
|
|
||||||
import binascii
|
import binascii
|
||||||
import hashlib
|
import hashlib
|
||||||
import struct
|
import struct
|
||||||
|
@ -34,7 +26,7 @@ def expand_peer(compact_peer_info):
|
||||||
host = ".".join([str(ord(d)) for d in compact_peer_info[:4]])
|
host = ".".join([str(ord(d)) for d in compact_peer_info[:4]])
|
||||||
port, = struct.unpack('>H', compact_peer_info[4:6])
|
port, = struct.unpack('>H', compact_peer_info[4:6])
|
||||||
peer_node_id = compact_peer_info[6:]
|
peer_node_id = compact_peer_info[6:]
|
||||||
return (peer_node_id, host, port)
|
return peer_node_id, host, port
|
||||||
|
|
||||||
|
|
||||||
def rpcmethod(func):
|
def rpcmethod(func):
|
||||||
|
@ -348,7 +340,7 @@ class Node(MockKademliaHelper):
|
||||||
stored_to = yield DeferredDict({contact: self.storeToContact(blob_hash, contact) for contact in contacts})
|
stored_to = yield DeferredDict({contact: self.storeToContact(blob_hash, contact) for contact in contacts})
|
||||||
contacted_node_ids = [binascii.hexlify(contact.id) for contact in stored_to.keys() if stored_to[contact]]
|
contacted_node_ids = [binascii.hexlify(contact.id) for contact in stored_to.keys() if stored_to[contact]]
|
||||||
log.debug("Stored %s to %i of %i attempted peers", binascii.hexlify(blob_hash),
|
log.debug("Stored %s to %i of %i attempted peers", binascii.hexlify(blob_hash),
|
||||||
len(list(contacted_node_ids)), len(contacts))
|
len(contacted_node_ids), len(contacts))
|
||||||
defer.returnValue(contacted_node_ids)
|
defer.returnValue(contacted_node_ids)
|
||||||
|
|
||||||
def change_token(self):
|
def change_token(self):
|
||||||
|
|
|
@ -2,10 +2,9 @@
|
||||||
Utilities for turning plain files into LBRY Files.
|
Utilities for turning plain files into LBRY Files.
|
||||||
"""
|
"""
|
||||||
|
|
||||||
import six
|
|
||||||
import binascii
|
|
||||||
import logging
|
import logging
|
||||||
import os
|
import os
|
||||||
|
from binascii import hexlify
|
||||||
|
|
||||||
from twisted.internet import defer
|
from twisted.internet import defer
|
||||||
from twisted.protocols.basic import FileSender
|
from twisted.protocols.basic import FileSender
|
||||||
|
@ -38,14 +37,14 @@ class EncryptedFileStreamCreator(CryptStreamCreator):
|
||||||
def _finished(self):
|
def _finished(self):
|
||||||
# calculate the stream hash
|
# calculate the stream hash
|
||||||
self.stream_hash = get_stream_hash(
|
self.stream_hash = get_stream_hash(
|
||||||
hexlify(self.name), hexlify(self.key), hexlify(self.name),
|
hexlify(self.name.encode()), hexlify(self.key), hexlify(self.name.encode()),
|
||||||
self.blob_infos
|
self.blob_infos
|
||||||
)
|
)
|
||||||
|
|
||||||
# generate the sd info
|
# generate the sd info
|
||||||
self.sd_info = format_sd_info(
|
self.sd_info = format_sd_info(
|
||||||
EncryptedFileStreamType, hexlify(self.name), hexlify(self.key),
|
EncryptedFileStreamType, hexlify(self.name.encode()), hexlify(self.key),
|
||||||
hexlify(self.name), self.stream_hash.encode(), self.blob_infos
|
hexlify(self.name.encode()), self.stream_hash.encode(), self.blob_infos
|
||||||
)
|
)
|
||||||
|
|
||||||
# sanity check
|
# sanity check
|
||||||
|
@ -126,15 +125,7 @@ def create_lbry_file(blob_manager, storage, payment_rate_manager, lbry_file_mana
|
||||||
)
|
)
|
||||||
log.debug("adding to the file manager")
|
log.debug("adding to the file manager")
|
||||||
lbry_file = yield lbry_file_manager.add_published_file(
|
lbry_file = yield lbry_file_manager.add_published_file(
|
||||||
sd_info['stream_hash'], sd_hash, binascii.hexlify(file_directory.encode()), payment_rate_manager,
|
sd_info['stream_hash'], sd_hash, hexlify(file_directory.encode()), payment_rate_manager,
|
||||||
payment_rate_manager.min_blob_data_payment_rate
|
payment_rate_manager.min_blob_data_payment_rate
|
||||||
)
|
)
|
||||||
defer.returnValue(lbry_file)
|
defer.returnValue(lbry_file)
|
||||||
|
|
||||||
|
|
||||||
def hexlify(str_or_unicode):
|
|
||||||
if isinstance(str_or_unicode, six.text_type):
|
|
||||||
strng = str_or_unicode.encode('utf-8')
|
|
||||||
else:
|
|
||||||
strng = str_or_unicode
|
|
||||||
return binascii.hexlify(strng)
|
|
||||||
|
|
|
@ -1,149 +0,0 @@
|
||||||
import glob
|
|
||||||
import json
|
|
||||||
import os
|
|
||||||
import subprocess
|
|
||||||
import sys
|
|
||||||
|
|
||||||
import github
|
|
||||||
import uritemplate
|
|
||||||
import boto3
|
|
||||||
|
|
||||||
|
|
||||||
def main():
|
|
||||||
#upload_to_github_if_tagged('lbryio/lbry')
|
|
||||||
upload_to_s3('daemon')
|
|
||||||
|
|
||||||
|
|
||||||
def get_asset_filename():
|
|
||||||
root_dir = os.path.dirname(os.path.dirname(os.path.realpath(__file__)))
|
|
||||||
return glob.glob(os.path.join(root_dir, 'dist/*'))[0]
|
|
||||||
|
|
||||||
|
|
||||||
def get_cli_output(command):
|
|
||||||
return subprocess.check_output(command.split()).decode().strip()
|
|
||||||
|
|
||||||
|
|
||||||
def upload_to_s3(folder):
|
|
||||||
asset_path = get_asset_filename()
|
|
||||||
branch = get_cli_output('git rev-parse --abbrev-ref HEAD')
|
|
||||||
if branch = 'master':
|
|
||||||
tag = get_cli_output('git describe --always --abbrev=8 HEAD')
|
|
||||||
commit = get_cli_output('git show -s --format=%cd --date=format:%Y%m%d-%H%I%S HEAD')
|
|
||||||
bucket = 'releases.lbry.io'
|
|
||||||
key = '{}/{}-{}/{}'.format(folder, commit, tag, os.path.basename(asset_path))
|
|
||||||
else:
|
|
||||||
key = '{}/{}-{}/{}'.format(folder, commit_date, tag, os.path.basename(asset_path))
|
|
||||||
|
|
||||||
print("Uploading {} to s3://{}/{}".format(asset_path, bucket, key))
|
|
||||||
|
|
||||||
if 'AWS_ACCESS_KEY_ID' not in os.environ or 'AWS_SECRET_ACCESS_KEY' not in os.environ:
|
|
||||||
print('Must set AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY to publish assets to s3')
|
|
||||||
return 1
|
|
||||||
|
|
||||||
s3 = boto3.resource(
|
|
||||||
's3',
|
|
||||||
aws_access_key_id=os.environ['AWS_ACCESS_KEY_ID'],
|
|
||||||
aws_secret_access_key=os.environ['AWS_SECRET_ACCESS_KEY'],
|
|
||||||
config=boto3.session.Config(signature_version='s3v4')
|
|
||||||
)
|
|
||||||
s3.meta.client.upload_file(asset_path, bucket, key)
|
|
||||||
|
|
||||||
|
|
||||||
def upload_to_github_if_tagged(repo_name):
|
|
||||||
try:
|
|
||||||
current_tag = subprocess.check_output(
|
|
||||||
['git', 'describe', '--exact-match', 'HEAD']).strip()
|
|
||||||
except subprocess.CalledProcessError:
|
|
||||||
print('Not uploading to GitHub as we are not currently on a tag')
|
|
||||||
return 1
|
|
||||||
|
|
||||||
print("Current tag: " + current_tag)
|
|
||||||
|
|
||||||
if 'GH_TOKEN' not in os.environ:
|
|
||||||
print('Must set GH_TOKEN in order to publish assets to a release')
|
|
||||||
return 1
|
|
||||||
|
|
||||||
gh_token = os.environ['GH_TOKEN']
|
|
||||||
auth = github.Github(gh_token)
|
|
||||||
repo = auth.get_repo(repo_name)
|
|
||||||
|
|
||||||
if not check_repo_has_tag(repo, current_tag):
|
|
||||||
print('Tag {} is not in repo {}'.format(current_tag, repo))
|
|
||||||
# TODO: maybe this should be an error
|
|
||||||
return 1
|
|
||||||
|
|
||||||
asset_path = get_asset_filename()
|
|
||||||
print("Uploading " + asset_path + " to Github tag " + current_tag)
|
|
||||||
release = get_github_release(repo, current_tag)
|
|
||||||
upload_asset_to_github(release, asset_path, gh_token)
|
|
||||||
|
|
||||||
|
|
||||||
def check_repo_has_tag(repo, target_tag):
|
|
||||||
tags = repo.get_tags().get_page(0)
|
|
||||||
for tag in tags:
|
|
||||||
if tag.name == target_tag:
|
|
||||||
return True
|
|
||||||
return False
|
|
||||||
|
|
||||||
|
|
||||||
def get_github_release(repo, current_tag):
|
|
||||||
for release in repo.get_releases():
|
|
||||||
if release.tag_name == current_tag:
|
|
||||||
return release
|
|
||||||
raise Exception('No release for {} was found'.format(current_tag))
|
|
||||||
|
|
||||||
|
|
||||||
def upload_asset_to_github(release, asset_to_upload, token):
|
|
||||||
basename = os.path.basename(asset_to_upload)
|
|
||||||
for asset in release.raw_data['assets']:
|
|
||||||
if asset['name'] == basename:
|
|
||||||
print('File {} has already been uploaded to {}'.format(basename, release.tag_name))
|
|
||||||
return
|
|
||||||
|
|
||||||
upload_uri = uritemplate.expand(release.upload_url, {'name': basename})
|
|
||||||
count = 0
|
|
||||||
while count < 10:
|
|
||||||
try:
|
|
||||||
output = _curl_uploader(upload_uri, asset_to_upload, token)
|
|
||||||
if 'errors' in output:
|
|
||||||
raise Exception(output)
|
|
||||||
else:
|
|
||||||
print('Successfully uploaded to {}'.format(output['browser_download_url']))
|
|
||||||
except Exception:
|
|
||||||
print('Failed uploading on attempt {}'.format(count + 1))
|
|
||||||
count += 1
|
|
||||||
|
|
||||||
|
|
||||||
def _curl_uploader(upload_uri, asset_to_upload, token):
|
|
||||||
# using requests.post fails miserably with SSL EPIPE errors. I spent
|
|
||||||
# half a day trying to debug before deciding to switch to curl.
|
|
||||||
#
|
|
||||||
# TODO: actually set the content type
|
|
||||||
print('Using curl to upload {} to {}'.format(asset_to_upload, upload_uri))
|
|
||||||
cmd = [
|
|
||||||
'curl',
|
|
||||||
'-sS',
|
|
||||||
'-X', 'POST',
|
|
||||||
'-u', ':{}'.format(os.environ['GH_TOKEN']),
|
|
||||||
'--header', 'Content-Type: application/octet-stream',
|
|
||||||
'--data-binary', '@-',
|
|
||||||
upload_uri
|
|
||||||
]
|
|
||||||
# '-d', '{"some_key": "some_value"}',
|
|
||||||
print('Calling curl:')
|
|
||||||
print(cmd)
|
|
||||||
print('')
|
|
||||||
with open(asset_to_upload, 'rb') as fp:
|
|
||||||
p = subprocess.Popen(cmd, stdin=fp, stderr=subprocess.PIPE, stdout=subprocess.PIPE)
|
|
||||||
stdout, stderr = p.communicate()
|
|
||||||
print('curl return code: {}'.format(p.returncode))
|
|
||||||
if stderr:
|
|
||||||
print('stderr output from curl:')
|
|
||||||
print(stderr)
|
|
||||||
print('stdout from curl:')
|
|
||||||
print(stdout)
|
|
||||||
return json.loads(stdout)
|
|
||||||
|
|
||||||
|
|
||||||
if __name__ == '__main__':
|
|
||||||
sys.exit(main())
|
|
|
@ -1,3 +1,4 @@
|
||||||
|
from binascii import hexlify
|
||||||
from twisted.internet import task
|
from twisted.internet import task
|
||||||
from twisted.trial import unittest
|
from twisted.trial import unittest
|
||||||
from lbrynet.core.utils import generate_id
|
from lbrynet.core.utils import generate_id
|
||||||
|
@ -10,50 +11,62 @@ class ContactOperatorsTest(unittest.TestCase):
|
||||||
def setUp(self):
|
def setUp(self):
|
||||||
self.contact_manager = ContactManager()
|
self.contact_manager = ContactManager()
|
||||||
self.node_ids = [generate_id(), generate_id(), generate_id()]
|
self.node_ids = [generate_id(), generate_id(), generate_id()]
|
||||||
self.firstContact = self.contact_manager.make_contact(self.node_ids[1], '127.0.0.1', 1000, None, 1)
|
make_contact = self.contact_manager.make_contact
|
||||||
self.secondContact = self.contact_manager.make_contact(self.node_ids[0], '192.168.0.1', 1000, None, 32)
|
self.first_contact = make_contact(self.node_ids[1], '127.0.0.1', 1000, None, 1)
|
||||||
self.secondContactCopy = self.contact_manager.make_contact(self.node_ids[0], '192.168.0.1', 1000, None, 32)
|
self.second_contact = make_contact(self.node_ids[0], '192.168.0.1', 1000, None, 32)
|
||||||
self.firstContactDifferentValues = self.contact_manager.make_contact(self.node_ids[1], '192.168.1.20',
|
self.second_contact_copy = make_contact(self.node_ids[0], '192.168.0.1', 1000, None, 32)
|
||||||
1000, None, 50)
|
self.first_contact_different_values = make_contact(self.node_ids[1], '192.168.1.20', 1000, None, 50)
|
||||||
self.assertRaises(ValueError, self.contact_manager.make_contact, self.node_ids[1], '192.168.1.20',
|
|
||||||
100000, None)
|
|
||||||
self.assertRaises(ValueError, self.contact_manager.make_contact, self.node_ids[1], '192.168.1.20.1',
|
|
||||||
1000, None)
|
|
||||||
self.assertRaises(ValueError, self.contact_manager.make_contact, self.node_ids[1], 'this is not an ip',
|
|
||||||
1000, None)
|
|
||||||
self.assertRaises(ValueError, self.contact_manager.make_contact, "this is not a node id", '192.168.1.20.1',
|
|
||||||
1000, None)
|
|
||||||
|
|
||||||
def testNoDuplicateContactObjects(self):
|
def test_make_contact_error_cases(self):
|
||||||
self.assertTrue(self.secondContact is self.secondContactCopy)
|
self.assertRaises(
|
||||||
self.assertTrue(self.firstContact is not self.firstContactDifferentValues)
|
ValueError, self.contact_manager.make_contact, self.node_ids[1], '192.168.1.20', 100000, None)
|
||||||
|
self.assertRaises(
|
||||||
|
ValueError, self.contact_manager.make_contact, self.node_ids[1], '192.168.1.20.1', 1000, None)
|
||||||
|
self.assertRaises(
|
||||||
|
ValueError, self.contact_manager.make_contact, self.node_ids[1], 'this is not an ip', 1000, None)
|
||||||
|
self.assertRaises(
|
||||||
|
ValueError, self.contact_manager.make_contact, b'not valid node id', '192.168.1.20.1', 1000, None)
|
||||||
|
|
||||||
def testBoolean(self):
|
def test_no_duplicate_contact_objects(self):
|
||||||
|
self.assertTrue(self.second_contact is self.second_contact_copy)
|
||||||
|
self.assertTrue(self.first_contact is not self.first_contact_different_values)
|
||||||
|
|
||||||
|
def test_boolean(self):
|
||||||
""" Test "equals" and "not equals" comparisons """
|
""" Test "equals" and "not equals" comparisons """
|
||||||
self.assertNotEqual(
|
self.assertNotEqual(
|
||||||
self.firstContact, self.secondContact,
|
self.first_contact, self.second_contact,
|
||||||
'Contacts with different IDs should not be equal.')
|
'Contacts with different IDs should not be equal.')
|
||||||
self.assertEqual(
|
self.assertEqual(
|
||||||
self.firstContact, self.firstContactDifferentValues,
|
self.first_contact, self.first_contact_different_values,
|
||||||
'Contacts with same IDs should be equal, even if their other values differ.')
|
'Contacts with same IDs should be equal, even if their other values differ.')
|
||||||
self.assertEqual(
|
self.assertEqual(
|
||||||
self.secondContact, self.secondContactCopy,
|
self.second_contact, self.second_contact_copy,
|
||||||
'Different copies of the same Contact instance should be equal')
|
'Different copies of the same Contact instance should be equal')
|
||||||
|
|
||||||
def testIllogicalComparisons(self):
|
def test_illogical_comparisons(self):
|
||||||
""" Test comparisons with non-Contact and non-str types """
|
""" Test comparisons with non-Contact and non-str types """
|
||||||
msg = '"{}" operator: Contact object should not be equal to {} type'
|
msg = '"{}" operator: Contact object should not be equal to {} type'
|
||||||
for item in (123, [1, 2, 3], {'key': 'value'}):
|
for item in (123, [1, 2, 3], {'key': 'value'}):
|
||||||
self.assertNotEqual(
|
self.assertNotEqual(
|
||||||
self.firstContact, item,
|
self.first_contact, item,
|
||||||
msg.format('eq', type(item).__name__))
|
msg.format('eq', type(item).__name__))
|
||||||
self.assertTrue(
|
self.assertTrue(
|
||||||
self.firstContact != item,
|
self.first_contact != item,
|
||||||
msg.format('ne', type(item).__name__))
|
msg.format('ne', type(item).__name__))
|
||||||
|
|
||||||
def testCompactIP(self):
|
def test_compact_ip(self):
|
||||||
self.assertEqual(self.firstContact.compact_ip(), b'\x7f\x00\x00\x01')
|
self.assertEqual(self.first_contact.compact_ip(), b'\x7f\x00\x00\x01')
|
||||||
self.assertEqual(self.secondContact.compact_ip(), b'\xc0\xa8\x00\x01')
|
self.assertEqual(self.second_contact.compact_ip(), b'\xc0\xa8\x00\x01')
|
||||||
|
|
||||||
|
def test_id_log(self):
|
||||||
|
self.assertEqual(self.first_contact.log_id(False), hexlify(self.node_ids[1]))
|
||||||
|
self.assertEqual(self.first_contact.log_id(True), hexlify(self.node_ids[1])[:8])
|
||||||
|
|
||||||
|
def test_hash(self):
|
||||||
|
# fails with "TypeError: unhashable type: '_Contact'" if __hash__ is not implemented
|
||||||
|
self.assertEqual(
|
||||||
|
len({self.first_contact, self.second_contact, self.second_contact_copy}), 2
|
||||||
|
)
|
||||||
|
|
||||||
|
|
||||||
class TestContactLastReplied(unittest.TestCase):
|
class TestContactLastReplied(unittest.TestCase):
|
||||||
|
|
|
@ -1,8 +1,10 @@
|
||||||
# -*- coding: utf-8 -*-
|
import json
|
||||||
from cryptography.hazmat.primitives.ciphers.algorithms import AES
|
import mock
|
||||||
from twisted.trial import unittest
|
from twisted.trial import unittest
|
||||||
from twisted.internet import defer
|
from twisted.internet import defer
|
||||||
|
|
||||||
|
from cryptography.hazmat.primitives.ciphers.algorithms import AES
|
||||||
|
from lbrynet.database.storage import SQLiteStorage
|
||||||
from lbrynet.core.StreamDescriptor import get_sd_info, BlobStreamDescriptorReader
|
from lbrynet.core.StreamDescriptor import get_sd_info, BlobStreamDescriptorReader
|
||||||
from lbrynet.core.StreamDescriptor import StreamDescriptorIdentifier
|
from lbrynet.core.StreamDescriptor import StreamDescriptorIdentifier
|
||||||
from lbrynet.core.BlobManager import DiskBlobManager
|
from lbrynet.core.BlobManager import DiskBlobManager
|
||||||
|
@ -12,7 +14,7 @@ from lbrynet.core.PaymentRateManager import OnlyFreePaymentsManager
|
||||||
from lbrynet.database.storage import SQLiteStorage
|
from lbrynet.database.storage import SQLiteStorage
|
||||||
from lbrynet.file_manager import EncryptedFileCreator
|
from lbrynet.file_manager import EncryptedFileCreator
|
||||||
from lbrynet.file_manager.EncryptedFileManager import EncryptedFileManager
|
from lbrynet.file_manager.EncryptedFileManager import EncryptedFileManager
|
||||||
from lbrynet.core.StreamDescriptor import bytes2unicode
|
from lbrynet.core.StreamDescriptor import JSONBytesEncoder
|
||||||
from tests import mocks
|
from tests import mocks
|
||||||
from tests.util import mk_db_and_blob_dir, rm_db_and_blob_dir
|
from tests.util import mk_db_and_blob_dir, rm_db_and_blob_dir
|
||||||
|
|
||||||
|
@ -30,7 +32,7 @@ MB = 2**20
|
||||||
|
|
||||||
def iv_generator():
|
def iv_generator():
|
||||||
while True:
|
while True:
|
||||||
yield '3' * (AES.block_size // 8)
|
yield b'3' * (AES.block_size // 8)
|
||||||
|
|
||||||
|
|
||||||
class CreateEncryptedFileTest(unittest.TestCase):
|
class CreateEncryptedFileTest(unittest.TestCase):
|
||||||
|
@ -87,7 +89,8 @@ class CreateEncryptedFileTest(unittest.TestCase):
|
||||||
# this comes from the database, the blobs returned are sorted
|
# this comes from the database, the blobs returned are sorted
|
||||||
sd_info = yield get_sd_info(self.storage, lbry_file.stream_hash, include_blobs=True)
|
sd_info = yield get_sd_info(self.storage, lbry_file.stream_hash, include_blobs=True)
|
||||||
self.maxDiff = None
|
self.maxDiff = None
|
||||||
self.assertDictEqual(bytes2unicode(sd_info), sd_file_info)
|
unicode_sd_info = json.loads(json.dumps(sd_info, sort_keys=True, cls=JSONBytesEncoder))
|
||||||
|
self.assertDictEqual(unicode_sd_info, sd_file_info)
|
||||||
self.assertEqual(sd_info['stream_hash'], expected_stream_hash)
|
self.assertEqual(sd_info['stream_hash'], expected_stream_hash)
|
||||||
self.assertEqual(len(sd_info['blobs']), 3)
|
self.assertEqual(len(sd_info['blobs']), 3)
|
||||||
self.assertNotEqual(sd_info['blobs'][0]['length'], 0)
|
self.assertNotEqual(sd_info['blobs'][0]['length'], 0)
|
||||||
|
|
Loading…
Reference in a new issue