function attribute segfault

[gregball]

BPO	403305
Nosy	@gvanrossum, @warsaw
Files	None: None

^{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 = 'https://github.com/warsaw'
closed_at = <Date 2001-01-19.19:46:14.000>
created_at = <Date 2001-01-17.22:01:46.000>
labels = ['interpreter-core']
title = 'function attribute segfault'
updated_at = <Date 2001-01-19.19:46:14.000>
user = 'https://bugs.python.org/gregball'

bugs.python.org fields:

activity = <Date 2001-01-19.19:46:14.000>
actor = 'barry'
assignee = 'barry'
closed = True
closed_date = None
closer = None
components = ['Interpreter Core']
creation = <Date 2001-01-17.22:01:46.000>
creator = 'greg_ball'
dependencies = []
files = ['3043']
hgrepos = []
issue_num = 403305
keywords = ['patch']
message_count = 8.0
messages = ['35402', '35403', '35404', '35405', '35406', '35407', '35408', '35409']
nosy_count = 3.0
nosy_names = ['gvanrossum', 'barry', 'greg_ball']
pr_nums = []
priority = 'normal'
resolution = None
stage = None
status = 'closed'
superseder = None
type = None
url = 'https://bugs.python.org/issue403305'
versions = []

[gregball]

In current CVS, I get a segfault for executing either of

del f.__dict__
del f.arbitrary_attr

when f is a function, and arbitrary_attr can be set or
not.

[gregball]

At two points, the code did not check for NULL,
leading to segmentation fault soon after.

[gregball]

I have ready a new version of the patch
(I won't upload it yet in case it's not welcome)
which does the following:

reference to f.__dict__ or f.func_dict returns a new dictionary if f did not previously have one. (Similar to what happens on first assignent to arbitrary attribute.)

Then the following code works:

>>> def f(): pass
... 
>>> d=f.__dict__
>>> f.a=1
>>> d['a']
1

instead of d being an unsubscriptable object.
To me it seems nasty if somethings __dict__ attribute
might not be a dictionary.

Since f.__dict__ never now _appears_ to be None,
I also removed the feature of assigning it to None to get rid of it.
del f.__dict__ can now be used for this purpose.
(previously only for creating sig11 =) )

One other thought: as it stands, function attributes are strangely behaving in restricted mode. You can't set any attributes directly. You can't directly get attributes that don't start with an underscore. However, you can get __dict__ and then do anything you like to it...

[warsaw]

Assigning this to me to look at later.

[gvanrossum]

This is a high priority patch -- needs to be deal with before 2.1a1 can go out.

[warsaw]

Attached is a new version of this patch, which I'm about to check in. greg_ball's patch and discovery of the bug was excellent, but there were a few other things to clean up in the file too (crashing on del f.func_defaults, raising the wrong exception when del'ing a non-existant attribute).

Applying and closing.

[warsaw]

Oops, I forgot to write out the new patch. Here 'tis for the historical record.

[warsaw]

Sorry, one more try.