Python 3.6.0 (default, Mar  3 2017, 00:15:36) 
[GCC 4.9.2] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> b'a\x00%i' % 1
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: not all arguments converted during bytes formatting
>>> b'a%i' % 1
b'a1'
>>> b'a%i\x00' % 1
b'a1\x00'


On python3.5, this works in all the scenarios

Python 3.5.1 (default, Jan 14 2017, 03:58:20) 
[GCC 4.9.2] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> b'a\x00%i' % 1
b'a\x001'
>>> b'a%i' % 1
b'a1'
>>> b'a%i\x00' % 1
b'a1\x00'

Could this be due to using _PyBytesWriter?

The problem is now _PyBytes_FormatEx uses strchr to sniff %. It should use memchr instead.

Thanks for the report, Nick!

I confirm that I introduced the regression with the commit fa7762ec066aa3632a25b6a52bb7597b8f17c2f3: "Issue #25349: Optimize bytes % args using the new private _PyBytesWriter API". Sorry about that, I always forget that str*() functions should not be used in Python since we support embedded null bytes :-/ It seems like only Python 3.6.0 release is affected.

"git tag --contains fa7762ec066aa3632a25b6a52bb7597b8f17c2f3" also returns v3.5.2, but I don't understand since I only pushed my change to the 3.6 branch...

Oh, and thank you for adding new unit tests ;-) Do we need similar unit tests for str%args and str.format()?

I think no. String is not affected now and its code uses related macros so I don't think it could suffer possible regression.

Xiang Zhang added the comment:
> I think no. String is not affected now and its code uses related macros so I don't think it could suffer possible regression.

Regressions are not something "expected", shit happens, all the time.
Unexpected corner cases are common :-)

New changeset df6d7b406f3d1b2e4e2014751bfa25574c4df222 by Xiang Zhang in branch '3.6':
[3.6] bpo-29714: Fix a regression that bytes format may fail when containing zero bytes inside. (GH-504)
https://github.com/python/cpython/commit/df6d7b406f3d1b2e4e2014751bfa25574c4df222

New changeset b76ad5121e2cfa89d6476d700cbcb65b7ffc39ac by Xiang Zhang in branch 'master':
bpo-29714:  Fix a regression that bytes format may fail when containing zero bytes inside. (GH-499)
https://github.com/python/cpython/commit/b76ad5121e2cfa89d6476d700cbcb65b7ffc39ac