ndbm can't iterate through values on OS X

[ForestGregg]

BPO	30388
Nosy	@terryjreedy, @ronaldoussoren, @ned-deily, @emilyemorehouse
Files	ndbm.zip: The ndbm files created by shelve. ndbm-repro.c

^{Note: these values reflect the state of the issue at the time it was migrated and might not reflect the current state.}

Show more details

GitHub fields:

assignee = None
closed_at = None
created_at = <Date 2017-05-17.16:03:55.042>
labels = ['OS-mac', 'type-bug', 'library']
title = "ndbm can't iterate through values on OS X"
updated_at = <Date 2020-11-27.11:32:30.539>
user = 'https://bugs.python.org/ForestGregg'

bugs.python.org fields:

activity = <Date 2020-11-27.11:32:30.539>
actor = 'ronaldoussoren'
assignee = 'none'
closed = False
closed_date = None
closer = None
components = ['Library (Lib)', 'macOS']
creation = <Date 2017-05-17.16:03:55.042>
creator = 'Forest Gregg'
dependencies = []
files = ['46872', '49627']
hgrepos = []
issue_num = 30388
keywords = []
message_count = 11.0
messages = ['293858', '293934', '293974', '293975', '294040', '295189', '295215', '295282', '295512', '380963', '381944']
nosy_count = 5.0
nosy_names = ['terry.reedy', 'ronaldoussoren', 'ned.deily', 'emilyemorehouse', 'Forest Gregg']
pr_nums = []
priority = 'normal'
resolution = None
stage = None
status = 'open'
superseder = None
type = 'behavior'
url = 'https://bugs.python.org/issue30388'
versions = ['Python 3.5']

Linked PRs

• gh-74573: document that ndbm can silently corrupt databases on macOS #113354
• [3.12] gh-74573: document that ndbm can silently corrupt databases on macOS (GH-113354) #113431
• [3.11] gh-74573: document that ndbm can silently corrupt databases on macOS (GH-113354) #113432

[ForestGregg]

On Mac OS 10.12.4, a large shelve, backed by ndbm, can be created. But when I attempt to iterate through the values of the shelve it raises this exception:

File "/Library/Frameworks/Python.framework/Versions/3.5/lib/python3.5/site-packages/dedupe/api.py", line 281, in _blockData
for block in viewvalues(blocks):
File "/Library/Frameworks/Python.framework/Versions/3.5/lib/python3.5/collections_abc.py", line 693, in __iter_
for key in self._mapping:
File "/Library/Frameworks/Python.framework/Versions/3.5/lib/python3.5/shelve.py", line 95, in __iter__
for k in self.dict.keys():
SystemError: Negative size passed to PyBytes_FromStringAndSize

I've confirmed that this works with Python 3.6.1. All the Python versions were installed from homebrew.

I cannot reproduce on linux on windows machines. I've attached a zip file of the ndbm files that python created.

[ForestGregg]

The ndbm db's two files (in the attachment) have the following size

tmp___otctx 0 bytes
tmp___otctx.db 12857344 bytes

[terryjreedy]

Do you mean that on OSX, you code works with 3.6.1 and fails with 3.5.3?

[ForestGregg]

Very sorry for the ambiguity.

The bug appears for both python 3.5 and python 3.6.1 on OS X. I have not tried other versions of python on OS X.

[ned-deily]

Can you provide a complete example of how to reproduce the behavior you are seeing? Ideally, it would include producing the shelve object. Also, please provide the output of this command:

otool -L $(python3.6 -c 'import _dbm;print(dbm.__file_)')

[ForestGregg]

I have been trying to make a small reproducible example, but haven't been able to isolate it.

Running this script [1] csv_example on mac os x under either py27 or py3 does seem *often* cause this problem [2], [3]:

[1] https://github.com/dedupeio/dedupe-examples/blob/master/csv_example/csv_example.py
[2] dedupeio/dedupe#571
[3] dedupeio/dedupe-examples#54

[ned-deily]

It would be helpful if you could provide the output of this command:

otool -L $(python3.6 -c 'import _dbm;print(dbm.__file_)')

if necessary, substituting whatever command name(s) you are using to invoke python when failing.

[ForestGregg]

From one user who had problems under both 3.5 and 3.6

otool -L $(python3.5 -c 'import _dbm;print(dbm.__file_)')

/Library/Frameworks/Python.framework/Versions/3.5/lib/python3.5/lib-dynload/_dbm.cpython-35m-darwin.so:
/usr/lib/libSystem.B.dylib (compatibility version 1.0.0,
current version 125.2.0)

otool -L $(python3.6 -c 'import _dbm;print(dbm.__file_)')

/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/lib-dynload/_
dbm.cpython-36m-darwin.so:

/usr/lib/libSystem.B.dylib (compatibility version 1.0.0, current version
125.2.0)

[ForestGregg]

A different user:

otool -L $(python3.6 -c 'import _dbm;print(_dbm.file)')
/usr/local/var/pyenv/versions/3.6.1/lib/python3.6/lib-dynload/_dbm.cpython-36m-darwin.so:
/usr/lib/libSystem.B.dylib (compatibility version 1.0.0, current version 1238.50.2)

[ronaldoussoren]

See also bpo-33074

[ronaldoussoren]

I've attached a C program that demonstrates the problem, and have filed an issue with Apple about this: FB8919203

It is not clear to me what we can do about this, other than not building the ndbm binding on macOS. The manpages for the ndbm APIs do not mention limitations on item sizes, and the dbopen manpage mention that there a no limitations on item sizes).

[ronaldoussoren]

The best we can do is document the issue.

[ronaldoussoren]

The just merged PRs document the problem in main, 3.12 and 3.11. As we cannot fix Apple's code that's the best we can do.