Files
@ f734d107296e
Branch filter:
Location: kallithea/scripts/docs-headings.py - annotation
f734d107296e
2.6 KiB
text/x-python
auth: for default permissions, use existing explicit query result values instead of following dot references in ORM result objects
There has been reports of spurious crashes on resolving references like
.repository from Permissions:
File ".../kallithea/lib/auth.py", line 678, in __wrapper
if self.check_permissions(user):
File ".../kallithea/lib/auth.py", line 718, in check_permissions
return user.has_repository_permission_level(repo_name, self.required_perm)
File ".../kallithea/lib/auth.py", line 450, in has_repository_permission_level
actual_perm = self.permissions['repositories'].get(repo_name)
File ".../kallithea/lib/vcs/utils/lazy.py", line 41, in __get__
value = self._func(obj)
File ".../kallithea/lib/auth.py", line 442, in permissions
return self.__get_perms(user=self, cache=False)
File ".../kallithea/lib/auth.py", line 498, in __get_perms
return compute(user_id, user_is_admin)
File ".../kallithea/lib/auth.py", line 190, in _cached_perms_data
r_k = perm.UserRepoToPerm.repository.repo_name
File ".../sqlalchemy/orm/attributes.py", line 285, in __get__
return self.impl.get(instance_state(instance), dict_)
File ".../sqlalchemy/orm/attributes.py", line 721, in get
value = self.callable_(state, passive)
File ".../sqlalchemy/orm/strategies.py", line 710, in _load_for_state
% (orm_util.state_str(state), self.key)
sqlalchemy.orm.exc.DetachedInstanceError: Parent instance <UserRepoToPerm at ...> is not bound to a Session; lazy load operation of attribute 'repository' cannot proceed (Background on this error at: http://sqlalche.me/e/bhk3)
Permissions are cached between requests: SA result records are stored in in
beaker.cache.sql_cache_short and resued in following requests after the initial
session as been removed. References in Permission objects would usually give
lazy lookup ... but not outside the original session, where we would get an
error like this.
Permissions are indeed implemented/used incorrectly. That might explain a part
of the problem. Even if not fully explaining or fixing this problem, it is
still worth fixing:
Permissions are fetched from the database using Session().query with multiple
class/table names (joined together in way that happens to match the references
specified in the table definitions) - including Repository. The results are
thus "structs" with selected objects. If repositories always were retrieved
using this selected repository, everything would be fine. In some places, this
was what we did.
But in some places, the code happened to do what was more intuitive: just use
.repository and rely on "lazy" resolving. SA was not aware that this one
already was present in the result struct, and would try to fetch it again. Best
case, that could be inefficient. Worst case, it would fail as we see here.
Fix this by only querying from one table but use the "joinedload" option to
also fetch other referenced tables in the same select. (This might
inefficiently return the main record multiple times ... but that was already
the case with the previous approach.)
This change is thus doing multiple things with circular dependencies that can't
be split up in minor parts without taking detours:
The existing repository join like:
.join((Repository, UserGroupRepoToPerm.repository_id == Repository.repo_id))
is thus replaced by:
.options(joinedload(UserGroupRepoToPerm.repository))
Since we only are doing Session.query() on one table, the results will be of
that type instead of "structs" with multiple objects. If only querying for
UserRepoToPerm this means:
- perm.UserRepoToPerm.repository becomes perm.repository
- perm.Permission.permission_name looked at the explicitly queried Permission
in the result struct - instead it should look in the the dereferenced
repository as perm.permission.permission_name
There has been reports of spurious crashes on resolving references like
.repository from Permissions:
File ".../kallithea/lib/auth.py", line 678, in __wrapper
if self.check_permissions(user):
File ".../kallithea/lib/auth.py", line 718, in check_permissions
return user.has_repository_permission_level(repo_name, self.required_perm)
File ".../kallithea/lib/auth.py", line 450, in has_repository_permission_level
actual_perm = self.permissions['repositories'].get(repo_name)
File ".../kallithea/lib/vcs/utils/lazy.py", line 41, in __get__
value = self._func(obj)
File ".../kallithea/lib/auth.py", line 442, in permissions
return self.__get_perms(user=self, cache=False)
File ".../kallithea/lib/auth.py", line 498, in __get_perms
return compute(user_id, user_is_admin)
File ".../kallithea/lib/auth.py", line 190, in _cached_perms_data
r_k = perm.UserRepoToPerm.repository.repo_name
File ".../sqlalchemy/orm/attributes.py", line 285, in __get__
return self.impl.get(instance_state(instance), dict_)
File ".../sqlalchemy/orm/attributes.py", line 721, in get
value = self.callable_(state, passive)
File ".../sqlalchemy/orm/strategies.py", line 710, in _load_for_state
% (orm_util.state_str(state), self.key)
sqlalchemy.orm.exc.DetachedInstanceError: Parent instance <UserRepoToPerm at ...> is not bound to a Session; lazy load operation of attribute 'repository' cannot proceed (Background on this error at: http://sqlalche.me/e/bhk3)
Permissions are cached between requests: SA result records are stored in in
beaker.cache.sql_cache_short and resued in following requests after the initial
session as been removed. References in Permission objects would usually give
lazy lookup ... but not outside the original session, where we would get an
error like this.
Permissions are indeed implemented/used incorrectly. That might explain a part
of the problem. Even if not fully explaining or fixing this problem, it is
still worth fixing:
Permissions are fetched from the database using Session().query with multiple
class/table names (joined together in way that happens to match the references
specified in the table definitions) - including Repository. The results are
thus "structs" with selected objects. If repositories always were retrieved
using this selected repository, everything would be fine. In some places, this
was what we did.
But in some places, the code happened to do what was more intuitive: just use
.repository and rely on "lazy" resolving. SA was not aware that this one
already was present in the result struct, and would try to fetch it again. Best
case, that could be inefficient. Worst case, it would fail as we see here.
Fix this by only querying from one table but use the "joinedload" option to
also fetch other referenced tables in the same select. (This might
inefficiently return the main record multiple times ... but that was already
the case with the previous approach.)
This change is thus doing multiple things with circular dependencies that can't
be split up in minor parts without taking detours:
The existing repository join like:
.join((Repository, UserGroupRepoToPerm.repository_id == Repository.repo_id))
is thus replaced by:
.options(joinedload(UserGroupRepoToPerm.repository))
Since we only are doing Session.query() on one table, the results will be of
that type instead of "structs" with multiple objects. If only querying for
UserRepoToPerm this means:
- perm.UserRepoToPerm.repository becomes perm.repository
- perm.Permission.permission_name looked at the explicitly queried Permission
in the result struct - instead it should look in the the dereferenced
repository as perm.permission.permission_name
aa6f17a53b49 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 0a277465fddf f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 ed2fb6e84a02 ed2fb6e84a02 ed2fb6e84a02 ed2fb6e84a02 ed2fb6e84a02 ed2fb6e84a02 ed2fb6e84a02 ed2fb6e84a02 ed2fb6e84a02 ed2fb6e84a02 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 a188803df37e a188803df37e 01aca0a4f876 a8e6bb9ee9ea 665dfa112f2c f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 a8e6bb9ee9ea f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 a8e6bb9ee9ea f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 ed2fb6e84a02 ed2fb6e84a02 ed2fb6e84a02 ed2fb6e84a02 ed2fb6e84a02 ed2fb6e84a02 ed2fb6e84a02 ed2fb6e84a02 ed2fb6e84a02 ed2fb6e84a02 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 665dfa112f2c a188803df37e a8e6bb9ee9ea f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 | #!/usr/bin/env python3
"""
Consistent formatting of rst section titles
"""
import re
import subprocess
spaces = [
(0, 1), # we assume this is a over-and-underlined header
(2, 1),
(1, 1),
(1, 0),
(1, 0),
]
# http://sphinx-doc.org/rest.html :
# for the Python documentation, this convention is used which you may follow:
# # with overline, for parts
# * with overline, for chapters
# =, for sections
# -, for subsections
# ^, for subsubsections
# ", for paragraphs
pystyles = ['#', '*', '=', '-', '^', '"']
# match on a header line underlined with one of the valid characters
headermatch = re.compile(r'''\n*(.+)\n([][!"#$%&'()*+,./:;<=>?@\\^_`{|}~-])\2{2,}\n+''', flags=re.MULTILINE)
def main():
filenames = subprocess.check_output(['hg', 'loc', 'set:**.rst+kallithea/i18n/how_to']).splitlines()
for fn in filenames:
fn = fn.decode()
print('processing %s' % fn)
s = open(fn).read()
# find levels and their styles
lastpos = 0
styles = []
for markup in headermatch.findall(s):
style = markup[1]
if style in styles:
stylepos = styles.index(style)
if stylepos > lastpos + 1:
print('bad style %r with level %s - was at %s' % (style, stylepos, lastpos))
else:
stylepos = len(styles)
if stylepos > lastpos + 1:
print('bad new style %r - expected %r' % (style, styles[lastpos + 1]))
else:
styles.append(style)
lastpos = stylepos
# remove superfluous spacing (may however be restored by header spacing)
s = re.sub(r'''(\n\n)\n*''', r'\1', s, flags=re.MULTILINE)
if styles:
newstyles = pystyles[pystyles.index(styles[0]):]
def subf(m):
title, style = m.groups()
level = styles.index(style)
before, after = spaces[level]
newstyle = newstyles[level]
return '\n' * (before + 1) + title + '\n' + newstyle * len(title) + '\n' * (after + 1)
s = headermatch.sub(subf, s)
# remove superfluous spacing when headers are adjacent
s = re.sub(r'''(\n.+\n([][!"#$%&'()*+,./:;<=>?@\\^_`{|}~-])\2{2,}\n\n\n)\n*''', r'\1', s, flags=re.MULTILINE)
# fix trailing space and spacing before link sections
s = s.strip() + '\n'
s = re.sub(r'''\n+((?:\.\. _[^\n]*\n)+)$''', r'\n\n\n\1', s)
open(fn, 'w').write(s)
print(subprocess.check_output(['hg', 'diff'] + filenames))
if __name__ == '__main__':
main()
|