$ python3 -c "'{0:s}'.format(b'qwe')"
Traceback (most recent call last):
  File "<string>", line 1, in <module>
TypeError: non-empty format string passed to object.__format__

Spent many hours to detect bug in my code.

Bytes don't support formatting codes, so they are passed through to object, which doesn't support any formatting codes, and produces the error message you see. Since all other built in types reject invalid codes with a message that mentions their type, I think bytes should too. This would change the message to:

ValueError: Unknown format code 's' for object of type 'bytes'

Would that have lessened your confusion sufficiently?

Yes, I think we should implement this for object.__format__. Marking as easy.

Yes, that message will be sufficient. Perfectly.

Here's a patch. Changes in test_builtin might be redundant, but I added them anyway.

I left a comment on Rietveld, but repeating it here because for me those messages often go to spam:

I think you want to make this more general, by modifying the message for the TypeError that follows. Any type that doesn't provide it's own __format__ should produce an error when using a non-empty format string.

For example:

>>> format({}, 'a')
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: non-empty format string passed to object.__format__

Would be better as:
TypeError: non-empty format string passed to __format__ for object of type "class <'dict'>"

(Or some prettier way to print out the type).

Sorry, but it looks to me that bpo-28385.diff solves wrong issue. The issue is not in bytes, but in default __format__. object.__format__ should raise an error message that contains the name of the actual type. There are other issues with object.__format__, following patch tries to solve them.

1. Error message now contains the name of the actual type.

2. Format spec is checked before converting the value to str. Converting bytes and bytearray to str can raise a BytesWarning, and it is better if the type of this error doesn't depend on the -b option.

3. Since format spec is always empty, avoid calling PyObject_Format(). In theory PyObject_Str() can raise a subclass of str with non-trivial __format__, but I think we can ignore this subtle behavior difference.

LGTM, although I'm not so sure about your #3. Maybe it should be a separate issue and raised on python-dev? But I don't feel strongly enough about it to argue the point.

Thanks!

Removed item 3 from updated patch.

From msg214162:

This is basically the definition of object.__format__:

def __format__(self, specifier):
if len(specifier) == 0:
return str(self)
raise TypeError('non-empty format string passed to object.__format__')

This simple algorithm is implemented in object-format.patch.

But current implementation uses format(str(self), '') instead of just str(self). object-format-2.patch keeps this.

New changeset 92cae79fa5d9 by Serhiy Storchaka in branch '3.5':
Issue bpo-28385: An error message when non-empty format spec is passed to
https://hg.python.org/cpython/rev/92cae79fa5d9

New changeset 0a985f7c6731 by Serhiy Storchaka in branch '3.6':
Issue bpo-28385: An error message when non-empty format spec is passed to
https://hg.python.org/cpython/rev/0a985f7c6731

New changeset 6e8183abcc35 by Serhiy Storchaka in branch 'default':
Issue bpo-28385: An error message when non-empty format spec is passed to
https://hg.python.org/cpython/rev/6e8183abcc35

Buildbots are failing:

======================================================================
FAIL: test_errors (test.test_fstring.TestCase) (str="f'{(lambda: 0):x}'")
----------------------------------------------------------------------
TypeError: unsupported format string passed to function.__format__

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/root/buildarea/3.6.angelico-debian-amd64/build/Lib/test/test_fstring.py", line 20, in assertAllRaise
    eval(str)
AssertionError: "non-empty" does not match "unsupported format string passed to function.__format__"

======================================================================
FAIL: test_errors (test.test_fstring.TestCase) (str="f'{(0,):x}'")
----------------------------------------------------------------------
TypeError: unsupported format string passed to tuple.__format__

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/root/buildarea/3.6.angelico-debian-amd64/build/Lib/test/test_fstring.py", line 20, in assertAllRaise
    eval(str)
AssertionError: "non-empty" does not match "unsupported format string passed to tuple.__format__"

Ran 48 tests in 1.389s

What is better?

1. Replace "unsupported format string passed to" back to "non-empty format string passed to " in generated error message? Note that full error message is changed in any case, this is the purpose of this issue.

2. Change the pattern "non-empty" to "unsupported" in f-string tests. Or to other part of new error message. It may be changed again in future.

I would change the f-string tests. I realize it's a fragile test, but it's the only way to test the strings in the exception.

New changeset f43078ec598c by Serhiy Storchaka in branch '3.6':
Update the f-string test broken in issue bpo-28385.
https://hg.python.org/cpython/rev/f43078ec598c

New changeset 931410a04240 by Serhiy Storchaka in branch 'default':
Update the f-string test broken in issue bpo-28385.
https://hg.python.org/cpython/rev/931410a04240

Shouldn't the type of error be changed from TypeError to ValueError?

TypeError is documented as "Raised when an operation or function is applied to an object of inappropriate type". I think that fits this case: you're applying some format code to a type that doesn't support it.

But on other hand, the error depends on the value of format specifier:

>>> format('abc', '')
'abc'
>>> format('abc', 'j')
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
ValueError: Unknown format code 'j' for object of type 'str'
>>> format([1, 2, 3], '')
'[1, 2, 3]'
>>> format([1, 2, 3], 'j')
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: unsupported format string passed to list.__format__

If keep TypeError I think it would be better to restore former error message "non-empty format string".

I don't have a strong feeling one way or the other. I'd be surprised if anyone is catching these errors.

The proposition of making default __format__ equivalent to str() will be addressed in separate issue. Python-Dev thread: https://mail.python.org/pipermail/python-dev/2016-October/146765.html.