setattr a read-only property; the AttributeError should show the attribute that failed

[ztane]

BPO	27794
Nosy	@rhettinger, @ncoghlan, @bitdancer, @ztane, @zhangyangyu, @Vgr255, @tirkarthi, @uriyyo
PRs	bpo-27794: Add name attribute to property class #23967

^{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/rhettinger'
closed_at = <Date 2020-12-30.09:52:51.256>
created_at = <Date 2016-08-18.13:57:19.594>
labels = ['interpreter-core', 'type-feature', '3.10']
title = 'setattr a read-only property; the AttributeError should show the attribute that failed'
updated_at = <Date 2020-12-30.09:52:51.253>
user = 'https://github.com/ztane'

bugs.python.org fields:

activity = <Date 2020-12-30.09:52:51.253>
actor = 'rhettinger'
assignee = 'rhettinger'
closed = True
closed_date = <Date 2020-12-30.09:52:51.256>
closer = 'rhettinger'
components = ['Interpreter Core']
creation = <Date 2016-08-18.13:57:19.594>
creator = 'ztane'
dependencies = []
files = []
hgrepos = []
issue_num = 27794
keywords = ['patch']
message_count = 16.0
messages = ['273027', '273033', '273035', '273036', '273302', '273385', '273387', '273396', '301045', '383852', '383868', '383870', '383873', '383876', '384053', '384054']
nosy_count = 10.0
nosy_names = ['rhettinger', 'ncoghlan', 'r.david.murray', 'ztane', 'xiang.zhang', 'abarry', 'apatrushev', 'Muhammad Alkarouri', 'xtreak', 'uriyyo']
pr_nums = ['23967']
priority = 'normal'
resolution = 'fixed'
stage = 'resolved'
status = 'closed'
superseder = None
type = 'enhancement'
url = 'https://bugs.python.org/issue27794'
versions = ['Python 3.10']

[ztane]

Today we had an internal server error in production. I went to see the sentry logs for the error, and was dismayed: the error was AttributeError: can't set attribute, and the faulting line was setattr(obj, attr, value) that happens in a for-loop that uses computed properties coming from who knows.

Well, I quickly ruled out that it was because the code was trying to set a value to a read-only property descriptor, the only problem was to find out *which of all these read-only properties* was it trying to set:

Python 3.6.0a3+ (default, Aug 11 2016, 11:45:31) 
[GCC 5.4.0 20160609] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> class Foo:
...     @property
...     def bar(self): pass
... 
>>> setattr(Foo(), 'bar', 42)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
AttributeError: can't set attribute

Could we change this for Python 3.6 so that the message for this could include the name of the property just like AttributeError: has no attribute 'baz' does?

[bitdancer]

If someone submits a patch in time, we'd probably accept it.

[Vgr255]

The approach I'd take would be to change how {get,set,del}attr handle AttributeError; possibly by automatically filling in the information and giving a nicer error message. This would fix this as a side-effect (somewhat; some bits of code would need to change). How does that sound to core devs?

[ztane]

Unfortunately it seems that it is not that straightforward. The descriptor object doesn't know the name of the property. The error is raised in property_descr_set. However the error itself can be propagated from setting another property.

[Vgr255]

Opened bpo-27823 as a followup to this, and carried nosy list over.

[zhangyangyu]

One solution I can think of is alter the constructor of property() and add an optional name attribute to it. If users really care about the exception message, they can set the attribute to the property's name. If they don't care, everything remains the same as now. This is just like the doc parameter of property() right now and the first parameter of type() when it is used to create a class dynamically. This is just a suggestion, I am not sure it's worth to do it.

[bitdancer]

Not worth it. It would feel like boilerplate, and the situation where you want the information is almost invariably going to be one you didn't anticipate and so didn't "bother" with the name. Either that or you always have to do it, and the elegance of your python code takes a hit not commensurate with the benefit.

[ztane]

I've got one idea about how to implement this, but it would require adding a new flag field to PyExc_AttributeError type.

This flag, if set, would tell that the AttributeError in question was raised in C descriptor code or under similar circumstances, and that the attribute name was not known, and thus it is OK for setattr/delattr and attribute lookups to append ": attributename" to the end of the message, then clear the flag; then all those places that raise AttributeError in __get__, __set__, __del__ would just need to set this flag.

[MuhammadAlkarouri]

Now that the descriptor protocol has __set_name__, does this make any difference to this issue? The property can know its name, subject to a suitable patch.

[uriyyo]

I have implemented this feature using an idea provided by Xiang Zhang.
Basically, property has new optional attribute name which will be added to msg of AttributeError in a case when name is set.

[rhettinger]

Yurii, this looks nice. I question whether *name* should be a part of the constructor because it immediately gets overridden by __set_name__ method.

>>> class A:
...    def x(self):
...        return 44
...    x = property(x, name='y')
...
>>> vars(A)['x'].name
'x'

[uriyyo]

Raymond, it's a good question.

I have added name to property constructor to support cases when a property is added to a class after it was declared.

For instance:

class Foo:
    pass


Foo.foo = property(name='foo')

f = Foo()
f.foo = 10

So, in my opinion, it's expected behavior that name is overwritten by __set_name__ method.

What do you think about that? Should we change this behevior?

[rhettinger]

In my opinion, it's expected behavior that name is
overwritten by __set_name__ method.

It is almost certain that there will be others won't share that expectation. The StackOverflow questions and bug reports are inevitable.

Most examples of __set_name__() use a private attribute. I recommend that you go that route and stick to the spirit of the original bug report. The OP asked for better error messages when possible. They didn't ask for an API expansion.

I have added name to property constructor to support cases
when a property is added to a class after it was declared.

That rarely occurs in practice. I wouldn't worry about it. Also AFAICT the only time the new error message matters is in the context of a setattr() where the attribute isn't already shown in the traceback. So the case in question is really a rarity inside another rarity. Let's declare YAGNI unless an actual end-user problem arises in real-world code.

[uriyyo]

You a totally right, looks like I tried to add a feature that no one asked for:)

I have updated PR and removed name parameter from property constuctor. Thanks for your advice.

[rhettinger]

New changeset c56387f by Yurii Karabas in branch 'master':
bpo-27794: Add name attribute to property class (GH-23967)
c56387f

[rhettinger]

Thanks for the PR :-)