math: Make Interval and IntervalSet covariant#180
Draft
llucax wants to merge 3 commits into
Draft
Conversation
`Interval`'s type parameter no longer includes `None`. The exported type variable `LessThanComparableOrNoneT` (bound to `LessThanComparable | None`) was deprecated and replaced with `LessThanComparableT` (bound to `LessThanComparable`). `None` is still accepted as a value for `start` / `end` to indicate an unbounded side, but it is treated purely as bound metadata rather than a value in the interval's comparable space. Note that explicitly using `Interval[int | None]` still works, as `(int | None) | None` is equivalent to `int | None`, even when it is no longer needed nor recommended. Old code will mostly work, but there is one soft breaking change: `None in some_interval` and `None in some_interval_set` is now a type-check error at call sites, matching the design intent that `None` is a bound marker and never a member value. Signed-off-by: Leandro Lucarella <luca-frequenz@llucax.com>
Add IntervalSet[T] to frequenz.core.math: a normalized, immutable set of Interval[T] values that supports O(log n) membership testing via the `in` operator. On construction, IntervalSet sorts the passed intervals by start and merges any that overlap or share an inclusive endpoint. Only the normalized form is stored, so equality, hashing, and iteration are well-defined regardless of input order; callers that need the original inputs can keep them separately. Unbounded intervals are handled via the existing Interval[T] convention: a None start is treated as -infinity and a None end as +infinity during sort/merge, so for example IntervalSet((Interval(None, 5), Interval(3, None))) collapses to a single Interval(None, None). The generic container-of-containers variant discussed during design is deferred: there is no immediate use case, and adding IntervalSet alone is a fully additive step that does not foreclose a future generic type in a separate module. Signed-off-by: Leandro Lucarella <luca-frequenz@llucax.com>
`Interval` and `IntervalSet` are immutable (frozen) containers whose value type only ever appears in output positions: the `start` / `end` fields, the stored `intervals`, and iteration. Their type parameter can therefore be covariant, so an `Interval[T]` / `IntervalSet[T]` is accepted where an `Interval[S]` / `IntervalSet[S]` is expected when `T` is a subtype of `S` (for example, passing an `Interval[Power]` where an `Interval[Quantity]` is wanted). The previous invariant typing rejected these safe assignments for no benefit; contravariance is unsound, as it would allow a stored `S` to be read back as a `T`. A covariant type variable must not appear in any input position, so `__contains__` now takes `object` instead of the value type, following the convention used by the standard library containers (`Container`, `Sequence`, ...). The item is cast back to the value type internally, so runtime behaviour is unchanged: comparing a value that is not comparable to the bounds still raises `TypeError`. This reverts the soft breaking change from "Remove `None` from `Interval` type parameter": with `__contains__` widened to `object`, `None in some_interval` (and other incompatible membership checks) are no longer flagged by type checkers at call sites. Losing that call-site check is the accepted trade-off for covariance, and misuse still fails loudly at runtime. The release notes are updated accordingly. Add tests for both aspects. The covariance tests assign an `Interval[bool]` / `IntervalSet[bool]` to the `int` variant; these only type check under covariance and are enforced by mypy, since the test suite is type checked (they are no-ops at runtime). The runtime tests assert that a membership check with a value that can't be compared to the bounds raises `TypeError`. Signed-off-by: Leandro Lucarella <luca-frequenz@llucax.com>
llucax
requested review from
florian-wagner-frequenz
and removed request for
a team
github-actions
Bot
added
part:docs
llucax
changed the title
Interval and IntervalSet covariant
Contributor
Author
|
Draft because it is based on #179. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
IntervalandIntervalSetare immutable (frozen) containers whose value type only ever appears in output positions: thestart/endfields, the storedintervals, and iteration. Their type parameter can therefore be covariant, so anInterval[T]/IntervalSet[T]is accepted where anInterval[S]/IntervalSet[S]is expected whenTis a subtype ofS(for example, passing anInterval[Power]where anInterval[Quantity]is wanted). The previous invariant typing rejected these safe assignments for no benefit; contravariance is unsound, as it would allow a storedSto be read back as aT.A covariant type variable must not appear in any input position, so
__contains__now takesobjectinstead of the value type, following the convention used by the standard library containers (Container,Sequence, ...). The item is cast back to the value type internally, so runtime behaviour is unchanged: comparing a value that is not comparable to the bounds still raisesTypeError.This reverts the soft breaking change from "Remove
NonefromIntervaltype parameter": with__contains__widened toobject,None in some_interval(and other incompatible membership checks) are no longer flagged by type checkers at call sites. Losing that call-site check is the accepted trade-off for covariance, and misuse still fails loudly at runtime. The release notes are updated accordingly.Add tests for both aspects. The covariance tests assign an
Interval[bool]/IntervalSet[bool]to theintvariant; these only type check under covariance and are enforced by mypy, since the test suite is type checked (they are no-ops at runtime). The runtime tests assert that a membership check with a value that can't be compared to the bounds raisesTypeError.