Version 4.2.3 breaks other packages

[tzickel]

I do not know if such a minor release has the intended consequences of breaking the API contract, but here is a project that just stopped working because of it:

https://github.com/wimglenn/johnnydep/blob/8c3f1e9eb81bc43db3b5f9d1f0b4b3a0c187eb82/johnnydep/pipper.py#L21

I assume this commit is at fault:
be507a6

@wimglenn

[wimglenn]

Oof, that is quite a backwards break for a micro bump (v4.2.2 -> v4.2.3) in what I assumed was semver

The fix for users is easy (change their import statements) but maybe this release should be yanked and compat shims / deprecation warns put in.. ?

[tkem]

cachetools uses semantic versioning, and the minor version change to 4.2.3 IMHO correctly indicates that the public API did not change.

The cachetools.ttl submodule was an undocumented implementation detail. Documentation including code examples use from cachetools import TTLCache, and have done so from the start. Why johnnydep chose to use cachetools.ttl I don't know, but this was never officially supported, nor even mentioned in the documentation.

[wimglenn]

Yeah, my bad - the autodoc always mentioned cachetools.TTLCache. Hyrum's law bites.

It might be that some IDEs which automatically write your import statements just take them from the module where the class is defined, not sure? There is sort of a convention in Python that anything without a leading underscore is fair game. Here are ~100 others making same mistake.

I'll note that I did mention those supposed "existing APIs" in comment #146 (comment)

[tkem]

Well, sorry for the inconvenience - I didn't think this would have much of an impact, but I guess you may be right about IDEs doing autocompletion. OTOH, the vast majority of users seem to do it right, and AFAICS most of the affected repos use copies/variants(?) of same code base, so who knows...

I usually do follow the underscore (or for class members, double underscore) convention, except for "private" submodules. But those were just a habit of mine, coming from languages like C/C++, and I'm slowly giving up on that, since they tend to have undesired effects.

And yes, I completely missed your comment in #146, sorry for that!

[tkem]

Temporarily reopened so others, who are affected by this, will find this more easily.

[ran-isenberg]

It broke our usage too. You should either support it backwards or change it to v5 as it's a breaking change.

[tkem]

Apparently @wimglenn was right about Hyrum's law... I guess I'll add some submodule shims for backward compatibility, but please do fix your imports ASAP.

[wimglenn]

@tkem No big deal, but there is a slightly improved way to indicate deprecation, details in #226

[tkem]

@wimglenn: Yes, of course. I know, or I knew at least... Kind of rushed this out today - thanks for pointing this out!