This isn't in a release, but it will affect CI jobs until the fix lands.

Due to a subtlety in the pbr codebase the test coverage in the
SemanticVersion object patch that recently landed wasn't what everyone
thought it was : there were untested code paths.

We've put together a fix [1] which includes test coverage :) - but if
you're impacted in the mean time - sorry!

Some details for the interested: one of the Bad Things pbr does today
is generate a version number like 1234567 - thats right, a number that
is indistinguishable from a very large major release :(. It does this,
and still bad but less problematic versions like abd12 when the git
tree you run sdist, or develop in, has never been

What we *thought* it did was create versions like 0.0.g$shaprefix but
due to the test suite setting PBR_VERSION to 0 that was because only
the pre-versioned code path was being tested at all.

The reason this is problematic is that if one of these all-alpha
versions got released - aiee. So the fix we've chosen is this:
 - stop generating versions like that (we were going to anyway, as the
semver stack progressed)
 - *not* handle them if someone has an sdisted version - its a bad
version that shouldn't ever have escaped
 - cross fingers that noone had already uploaded such a beast to PyPI.

The same patch also fixes some other variants of local versions that
*are* valid, but weren't tested (due to to the aforementioned
oversight in coverage).



Robert Collins <>
Distinguished Technologist
HP Converged Cloud

OpenStack-dev mailing list

Reply via email to