`bytes()` and `bytearray()` should have the default values for `errors` for string sources

[hyperkai]

bytes() and bytearray() don't show the default values for source, encoding and errors as shown below:

class bytearray(source=b'')
class bytearray(source, encoding)
class bytearray(source, encoding, errors)

class bytes(source=b'')
class bytes(source, encoding)
class bytes(source, encoding, errors)

So, they should have the default values as shown below:

class bytearray(source=b'')
class bytearray(source, encoding='utf-8')
class bytearray(source, encoding='utf-8', errors='strict')

class bytes(source=b'')
class bytes(source, encoding='utf-8')
class bytes(source, encoding='utf-8', errors='strict')

[ZeroIntensity]

I don't think we need the three separate lines anymore. It should suffice to just have one with all the defaults. PR welcome!

[skirpichev]

class bytes(source=b'') # (1)
class bytes(source=b'', encoding='utf-8') # (2)
class bytes(source=b'', encoding='utf-8', errors='strict')

No. That will be just wrong. Same for bytearray.

You can omit encoding for integer, iterable of integers or bytes. For strings - you must provide encoding, there is no default.

[ZeroIntensity]

I meant by just documenting encoding=None or something like that.

[skirpichev]

I meant by just documenting encoding=None or something like that.

But there is no such default.

This is an example of a callable with multiple signatures. This function accept up to 3 arguments:

1. 1-arg form, in source: bytes, integer or iterable of integers.
2. 2-arg form: str in source + encoding
3. 3-arg form: str in source + encoding + errors

You can't specify string + errors - you must also set encoding in this case.

Unless I miss something, we can't simplify this.

[terryjreedy]

Kai, I fixed the duplication of bytearray that omitted bytes.

One must omit both encoding and errors when source is not a string, or one gets "TypeError: encoding(errors) without a string argument".

There are 2, but I would say, only 2 signatures, one for non-string sources (defaulting to b'') and one for string sources, which requires encoding and allows errors Since errors has a de facto default of strict for string sources, whence str.encode is called, I think we can simplify and make the situation clearer by combining the 2 and 3 arg lines into bytes(source, encoding, errors=strict). It is irrelevant to the user whether bytes calls str.encode with 'strict' or not and I don't think this implementation detail should be part of the definition of bytes/bytearray. (Having separate 2 and 3 arg lines implies that bytes makes separate 2 or 3 arg calls to encode.)

Bytes has no inspect.signature, which is to say, inspect.signature(bytes) raises ValueError. Its signature in the docstring, as formatted by 'help', is:

Help on class bytes in module builtins:

class bytes(object)
 |  bytes(iterable_of_ints) -> bytes
 |  bytes(string, encoding[, errors]) -> bytes
 |  bytes(bytes_or_buffer) -> immutable copy of bytes_or_buffer
 |  bytes(int) -> bytes object of size given by the parameter initialized with null bytes
 |  bytes() -> empty bytes object

The entry for bytes(string, ...) should be either first, which matched the doc entry order, or put last and bytes() put first. In either case, separate it from the others.

I think this separation should be increased in the entry text. Something like

If source is a string (the second signature above),
... .
If source is not a string,
*
*
*

• If omitted, it defaults to b'' and an empty bytes of length 0 is created.

Currently, the call details are in the bytearray entry and the bytes entry refers to the former. I personally would put the details in the bytes entry and reference that in the bytearray entry. But that might be a separate issue.