releases
releases copied to clipboard
bitprophet
Fix compatibility with >=python-semanticversion-2.7.0
Context
The recent versions of python-semanticversion made changes to private
APIs, removing the interaction between Version(x, partial=True) and
Spec() (partial=True was designed for implementing the Spec class
only).
This breaks releases, has mentioned in #84 and #85; it has also been reported as affecting twine's doc building (https://github.com/pypa/twine/issues/492).
The attached patch switches to the recommended usage of python-semanticversion (whose docs are still lacking in that regard, sorry :disappointed:), while keeping all tests green - both with updated and old python-semanticversion releases.
Implementation notes
The code used Version(..., partial=True) to exclude ranges of version whose major
component didn't match a bugfix/issue range; the code went akin to:
Version('1', partial=True) in Spec('>=1.0')
This no longer works; this patch changes that behaviour to exclude families where no actual release matches the bugfix/issue range - this should be more accurate.
The patch also uses Version.coerce, the intended API to manage
non semver-compliant version strings.
Compatibility
The patch has been tested with both python-semanticversion==2.6.0 and
python-semanticversion==2.8.1; it can be included to an upgraded version
of releases even if users haven't yet upgraded python-semanticversion.
The two failing tests seem to me to be a travis-side issue: I haven't been able to reproduce them locally with the same Sphinx and Python version.
Travis reports an issue with a sphinx build exiting with an unexpected error code, but the trace mentions that the build has succeeded and no error message...
I've confirmed this change fixes releases where it currently fails. Please find a way to release this change.
Any news on this PR? It would be nice to be able to unpin python-semanticversion.
I'm actually leaning towards just pinning our requirements at semantic-version<2.7 (re #84), since they've now proven a tendency to break APIs on minor versions.
Do you have needs for features added in semantic-version in 2.7 or 2.8? To my knowledge Releases itself does not, so I'd rather punt until we do.
@bitprophet As the maintainer of python-semanticversion, I feel a bit hurt by the wording of your decision — derivating a "tendency" from a single change to a private, non-documented API seems a bit harsh.
Since releases is a release management tool, pinning the version on your side forces that decision onto users of releases too; they'll have to choose between:
- Not using
releases - Using an outdated version of
python-semanticversion - Using two different virtual environments, one for building/testing and another for releasing (and all calls to
releases.
How can we avoid this blocking situation? Would you be interested by a further pull request that would either vendor python-semanticversion, or simply include a minimal subset of its functionalities?
@rbarrois That's fair, "tendency" was poor word choice on my part, I should have put something less implying a string of events - apologies.
I'm confused by the assertion that the API call is private though - all our subclass did was call Version with partial=True and that seems to be part of the public API. Am I overlooking something? I definitely didn't have a mental note that we were using its undocumented innards (like I do have for all of our Sphinx interactions...).
EDIT after reading the PR more and discovering the commit message especially: got it. partial I guess was undocumented previously?
Re: the rest: I'm well aware of the tradeoffs involved in pinning, sadly. It's never ideal but especially with limited dev time it's often a very tempting band-aid.
That said, since I last visited this ticket, I've decided to pop out a 2.0 of Releases for Sphinx reasons (dropping older Sphinxes; no API change on our end); so I'm tempted to merge this after all, and if necessary raise our semantic_version requirement (but hopefully not?).
Unfortunately as-is the PR breaks the test suite. If I can't figure out exactly why, I may try creating a variant that is less of a delta from master while still being compliant with modern versions of semantic_version.
Still poking but it always seems to come back to the fact that a lot of the logic in Releases /really/ wants that partial-version feature.
Many such spots feel like they logically should kinda be able to be Spec objects instead (and in fact this is suggested in semantic_version's docs, eg SimpleSpec('1.x.x')), but Specs are not Versions and so cannot be treated quite like a partial Version could (eg you cannot ask if one of these Specs matches another, as they lack truncate).
Trying to see if I can work around this one way or another (e.g. trying to use 'floor' versions, 1.0.0 instead of '1.*' etc).
Ran out of working tendons but have an interim branch (autolinked above). Some changes the same as this PR, some different. Only one test fails vs the many failing in this PR, so I think I am on the right track and missing one spot somewhere.
The changes here imply I want to keep a Version subclass around that "acts partial" but in a way that does not rely on the existing/deprecated, internal-to-semantic_version partial kwarg. (It would instead just isolate the dumb "tack on zeroes" stuff seen in my branch.)
As seen in TODOs, also, a lot of the Releases-level code is probably kinda dumb and wants reorganizing anyways. Whether I'll do that as part of this or punt, not sure.
@bitprophet awesome, thanks for looking into this! I'll take a look at your branch later today, hopefully I'll be able to help you finding a good solution, or to see what kind of APIs would be needed over in python-semantic_version to support your use case ;)