Issue in `binascii.{a2b,b2a}_qp` functions with newlines characters

[picnixz]

Bug report

Bug description:

This was detected here: https://github.com/python/cpython/actions/runs/12200589926/job/34037087571#step:22:214

>>> import binascii
>>> x = b'\n\r\n'
>>> y = binascii.b2a_qp(x, quotetabs=False, istext=True, header=False)
>>> z = binascii.a2b_qp(y, header=False)
>>> x
b'\n\r\n'
>>> y
'\n\n'
>>> z
b'\n\n'

I think the issue should only concern new lines characters like that but I honestly don't know whether this is the expected result or not. Note that the docs say:

If the optional argument header is present and false, newline characters will be encoded as well; otherwise linefeed conversion might corrupt the binary data stream.

So it appears that '\r\n' is not correctly encoded as such. I'll investigate.

CPython versions tested on:

CPython main branch

Operating systems tested on:

No response

Related

• quopri_codec newline handling #64320

[picnixz]

... This is a known issue in the code:

    /* XXX: this function has the side effect of converting all of
     * the end of lines to be the same depending on this detection
     * here */

@serhiy-storchaka what can we do :( should we consider \r\n lines as being a single newline character or...?

[picnixz]

I'll consider this one different from #64320 since the binascii.{a2b,b2a}_qp functions were made to reflect the quopri module. But on the other hand, I don't know whether it's better to keep everything under the same issue (at least, here we would only focus on binascii and possibly change the tests and acknowledge the issue as a feature).

[picnixz]

Ok, so the pure Python implementation of quopri is correct:

import quopri
# C implementation
assert quopri.encodestring(b'\n\r\n') == b'\n\n'
assert quopri.decodestring(b'\n\n') == b'\n\n'
quopri.a2b_qp = quopri.b2a_qp = None
# Python implementation
assert quopri.encodestring(b'\n\r\n') == b'\n=0D\n'
assert quopri.decodestring(b'\n=0D\n') == b'\n\r\n'

[picnixz]

I think we should force the {encode,decode} functions to be compose as the identity. We explicitly state in binascii.{a2b,b2a}_qp docs that we don't want stream corruption. As a user, I expect dec(enc(x)) == x. @serhiy-storchaka Do you think we would break things if we were to do this? (namely reflect the Python implementation)

[serhiy-storchaka]

Isn't this a duplicate of #62222 (which cover more cases)?

[picnixz]

Oh... I didn't think there was yet another one. I'll close this one in favor of the other (it has a nicer issue number). Sorry for the duplicate (I'll also post the above).