Deprecate qiskit.tools #11514
Deprecate qiskit.tools #11514
Conversation
|
One or more of the the following people are requested to review this:
|
This commit deprecates the qiskit.tools module. Most of the functionality in this module will be removed in the Qiskit 1.0.0 release. The sole exception is the qiskit.tools.parallel module and it's associated public function qiskit.tools.parallel_map which has been moved to qiskit.utils.
555363b to
dc1f521
Compare
| if name in _DEPRECATED_NAMES: | ||
| module_name = _DEPRECATED_NAMES[name] | ||
| warnings.warn( | ||
| f"Accessing '{name}' from '{__name__}' is deprecated since Qiskit 0.46.0 " | ||
| f"and will be removed in Qiskit 1.0.0. Import from '{module_name}' instead ", | ||
| DeprecationWarning, | ||
| 2, | ||
| ) | ||
| return getattr(importlib.import_module(module_name), name) |
There was a problem hiding this comment.
Even with the 0.46 cheat / clever trick, this is technically breaking our deprecation policy, because there's no way to support both 0.45 and 0.46 without getting a warning - this PR is the one that moves them. Other things in 0.46 are deprecations for removals that are effected in the 1.0 release, where the alternative is "don't use this" rather than "use this from somewhere else".
Depending on how strict we want to be about that, we might need to live with qiskit.tools, but I think for the 1.0 release it's probably fine. Possibly still an edge case of the deprecation policy we ought to clarify.
Alternatively, we could remove parallel_map from the public API, but I imagine there are users of it, so it's suboptimal.
There was a problem hiding this comment.
Yeah, I was weighing the awkwardness of keeping the qiskit.tools package around with a single module vs violating the deprecation policy a bit in this case. I figured doing this all in 0.46 and violating the rules here was better than keeping qiskit.tools around in 1.0.0. But I'm not tied to that if people object I'm fine keeping this as qiskit.tools for the time being.
There was a problem hiding this comment.
I'm ok with doing it in this release - I think the set of users who might be using parallel_map will have a large overlap with the set of people who'll be broken for other reasons by 1.0.
|
Meant to say: other than that one comment, looks fine. |
There was a problem hiding this comment.
I don't mind merging as-is (unless there's a problem with a qiskit.tools import that still seems to exist?), though I think we're doing pylint a little disservice here - I think its cyclic-import warning is valid. That said, I'm sure there's loads of cases where cyclic-import is being suppressed when we could have done it better, so it doesn't need fixing now.
Summary
This commit deprecates the qiskit.tools module. Most of the functionality in this module will be removed in the Qiskit 1.0.0 release. The sole exception is the qiskit.tools.parallel module and it's associated public function qiskit.tools.parallel_map which has been moved to qiskit.utils.
Details and comments