Skip to content

Commit

Permalink
Fix hash caching race condition
Browse files Browse the repository at this point in the history
- Clean up more unused code
- Fixes #3289

Signed-off-by: Dan Ryan <dan@danryan.co>
  • Loading branch information
techalchemy committed Nov 23, 2018
1 parent b03983e commit 9b1b901
Show file tree
Hide file tree
Showing 2 changed files with 42 additions and 39 deletions.
1 change: 1 addition & 0 deletions news/3289.bugfix.rst
Original file line number Diff line number Diff line change
@@ -0,0 +1 @@
Fixed a race condition in hash resolution for dependencies for certain dependencies with missing cache entries or fresh Pipenv installs.
80 changes: 41 additions & 39 deletions pipenv/utils.py
Original file line number Diff line number Diff line change
Expand Up @@ -25,20 +25,6 @@

from vistir.compat import ResourceWarning

try:
from weakref import finalize
except ImportError:
try:
from .vendor.backports.weakref import finalize
except ImportError:

class finalize(object):
def __init__(self, *args, **kwargs):
logging.warn("weakref.finalize unavailable, not cleaning...")

def detach(self):
return False


logging.basicConfig(level=logging.ERROR)

Expand Down Expand Up @@ -411,40 +397,56 @@ def resolve(self):
self.resolved_tree.update(results)
return self.resolved_tree

def resolve_hashes(self):
def _should_include_hash(ireq):
from pipenv.vendor.vistir.compat import Path, to_native_string
from pipenv.vendor.vistir.path import url_to_path
@staticmethod
def _should_include_hash(ireq):
from pipenv.vendor.vistir.compat import Path, to_native_string
from pipenv.vendor.vistir.path import url_to_path

# We can only hash artifacts.
try:
if not ireq.link.is_artifact:
return False
except AttributeError:
# We can only hash artifacts.
try:
if not ireq.link.is_artifact:
return False
except AttributeError:
return False

# But we don't want normal pypi artifcats since the normal resolver
# handles those
if is_pypi_url(ireq.link.url):
return False
# But we don't want normal pypi artifcats since the normal resolver
# handles those
if is_pypi_url(ireq.link.url):
return False

# We also don't want to try to hash directories as this will fail
# as these are editable deps and are not hashable.
if (ireq.link.scheme == "file" and
Path(to_native_string(url_to_path(ireq.link.url))).is_dir()):
return False
return True
# We also don't want to try to hash directories as this will fail
# as these are editable deps and are not hashable.
if (ireq.link.scheme == "file" and
Path(to_native_string(url_to_path(ireq.link.url))).is_dir()):
return False
return True

def resolve_hashes(self):
if self.results is not None:
resolved_hashes = self.resolver.resolve_hashes(self.results)
for ireq, ireq_hashes in resolved_hashes.items():
# We _ALWAYS MUST PRIORITIZE_ the inclusion of hashes from local sources
# PLEASE *DO NOT MODIFY THIS* TO CHECK WHETHER AN IREQ ALREADY HAS A HASH
# RESOLVED. The resolver will pull hashes from PyPI and only from PyPI.
# The entire purpose of this approach is to include missing hashes.
# This fixes a race condition in resolution for missing dependency caches
# see pypa/pipenv#3289
if self._should_include_hash(ireq) and (
not ireq_hashes or ireq.link.scheme == "file"
):
if not ireq_hashes:
ireq_hashes = set()
ireq_hashes |= set(
self.resolver.repository._hash_cache.get_hash(ireq.link)
)
else:
ireq_hashes = set(ireq_hashes)
# The _ONLY CASE_ where we flat out set the value is if it isn't present
# It's a set, so otherwise we *always* need to do a union update
if ireq not in self.hashes:
if _should_include_hash(ireq):
self.hashes[ireq] = [
self.resolver.repository._hash_cache.get_hash(ireq.link)
]
else:
self.hashes[ireq] = ireq_hashes
self.hashes[ireq] = ireq_hashes
else:
self.hashes[ireq] |= ireq_hashes
return self.hashes


Expand Down

0 comments on commit 9b1b901

Please sign in to comment.