`base64` roundtrip failure when `altchars` contains duplicate bytes

[T90REAL]

Bug report

Bug description:

base64.b64encode and base64.b64decode use asymmetric translation logic when the altchars argument contains duplicate bytes. This causes data to become corrupted during a roundtrip (encode -> decode).

import base64

data_in = b'\x00\x00\x00\x03\xe0'
altchars = b'\x00\x00'
encoded = base64.b64encode(data_in, altchars=altchars)
data_out = base64.b64decode(encoded, altchars=altchars)

assert data_in == data_out, "Roundtrip failed! Data was corrupted."

Traceback (most recent call last):
  File "/data/src/test.py", line 8, in <module>
    assert data_in == data_out, "Roundtrip failed! Data was corrupted."
           ^^^^^^^^^^^^^^^^^^^
AssertionError: Roundtrip failed! Data was corrupted.

The script should execute successfully and the assertion should pass.

CPython versions tested on:

3.12

Operating systems tested on:

Linux

[serhiy-storchaka]

This works as expected. altchars specifies an alternative alphabet for the '+' and '/' characters. When you use the same character for the '+' and '/' characters, you can no longer distinguish them. Your encoding is not reversible.

[terryjreedy]

Should the functions raise a value error when passed a bad altchars bytes?

[serhiy-storchaka]

This will add an overhead for each call with non-default altchars.

We would need to check also that both characters do not conflict with other Base64 characters and perhaps that they are not some control or invisible characters.

[terryjreedy]

Thinking about it, those harder to check errors seem much more likely. Duplicated bytes are more obviously wrong.