md5 missmatch.

[gregtap]

On the current develop branch using Docker compose my local python lib gets uploaded with:

python setup.py sdist upload -r local

The source file is in my local directory (dist) and it ends up on the web container root folder. Something like <repo_name>/source/o/<package_name>.

So far so good.

However during the installation I get a md5 missmatch error.
After some digging my source md5 is indeed different than the file in the container. The file in the container is corrupted and I can not introspect it. Yet the file size seems to be close to the original.

Am I the only one experiencing such issues?

I am going to add some debug on the upload method but its hard to tell where the issue could be.

[gregtap]

I have redone an installation from scratch on a Vagrant vm with ansible and still experience the same issue.

[mvantellingen]

Ok, will try to reproduce. I assume the filesize doesn't match either?

[gregtap]

Right they do not. I digged a bit deeper and at one point I thought it was the distutils upload method from Python 2.7.8 on Mac going wrong during the multipart upload.

I finally gave up and setup pypiserver for now. The upload step worked well and did not corrupt the tar gz.

[eisensheng]

I'm also affected by this.

I put a bit of research into it and it appears that there is an unnecessary UTF-8 encoding happening somewhere. An ZIP archive starting with

000000 50 4b 03 04 14 00 00 00 08 00 c5 6e 65 47 cb af  >PK.........neG..<

becomes when uploaded

000000 50 4b 03 04 14 00 00 00 08 00 c3 85 6e 65 47 c3  >PK..........neG.<

Notice that the byte sequence 00 c5 6e 65 changed to the sequence 00 c3 85 6e 65. The byte sequence c3 85 is the UTF-8 encoded value for the character Å which has the code point 197 or to express it in hexadecimal c5, the byte of the original input.

I can't tell who's here the one to be blamed though.

Edit:
It appears to be working with twine though. So it's highly likely that distutils/setuptools is the one to be blamed here. But then again why is it working with other implementations of the PYPI?