`maskingShiftLeft(by:)`, `maskingShiftRight(by:)` and
`resizingShiftLeft(by:)` handle a negative amount by negating it and
shifting the other way, so `Int.min` trapped with an arithmetic
overflow. All three have a defined result for it: an amount of at least
`count` clears or removes every value, which is what `Int.min + 1`
already did.
Clamp the magnitude of a negative amount to `Int.max` in these three.
`resizingShiftRight(by:)` is left alone: a negative amount inserts
values there, and no array can hold as many as `Int.min` asks for.
The masking shifts had no tests. Add tests for both of them over
word-crossing counts and amounts on either side of `count`, for their
doc examples, and for the extreme amounts.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
`BitSet[members:]` resolved range expressions against
`Int.min ..< Int.max`. A closed range ending at `Int.max`
(`bits[members: ...Int.max]`, `bits[members: 0 ... Int.max]`) needs the
index after `Int.max`, so it trapped with "Index out of bounds", while
`..<Int.max` and `0...` worked.
Resolve against a small internal collection of all `Int` values whose
`index(after:)` saturates at `Int.max`. `Int.max` can never be a bit set
member, so `lower ... Int.max` resolving to `lower ..< Int.max` doesn't
change which members are selected. `BitSet.Counted[members:]` forwards
to `BitSet`, so it's fixed too.
converting a negative bit array to `Int128` lost the sign above the first
64 bits , so `Int128(truncatingIfNeeded: "1" as BitArray)` gave
18446744073709551615 , not -1 , `Int128(exactly:)` gave nil , and
`Int128(_:)` trapped . a negative value now starts out as all ones , so the
words shifted in keep the sign .
the trap message also named `BitArray` as the type it could not convert
to , it names the integer type now .
The examples for `intersection(_ other: Range<Int>)` and
`formIntersection(_ other: Range<Int>)` (on both `BitSet` and
`BitSet.Counted`) showed the result of the corresponding `subtract`
example instead: intersecting `[1, 2, 3, 4]` with `-10 ..< 3` yields
`[1, 2]`, not `[3, 4]`.
_resizingShiftRight(by:) copied the surviving suffix down to index 0 and
then removed `count &- amount` trailing bits instead of `amount`, so it
truncated the result to `amount` stale bits rather than keeping
`count - amount` valid ones. resizingShiftLeft(by: -n) was affected via
the same path, and the API had no test coverage at all.
Also fix the doc examples, which demonstrated the resizing shifts with
the masking shift function names (and, in one case, a result whose bit
count contradicted the documented growth by `amount`).
Add regression tests asserting parity with the documented semantics
(dropFirst / prepending false values) across word-boundary counts,
negative amounts, boundary `amount` values, and the doc examples
verbatim.
The generic `isEqualSet(to: some Sequence<Int>)` overload returned false
whenever `other.underestimatedCount` exceeded `count`. A sequence with
duplicate items can be longer than the set it equals, so
`([1, 2, 3, 4] as BitSet.Counted).isEqualSet(to: [4, 1, 1, 2, 3, 3, 4] as [Int])`
returned false, contradicting the documented behavior.
Forward straight to `BitSet.isEqualSet(to:)`, which already handles
duplicates and keeps the count shortcut for unique collections.
BitSet.Counted caches its size in `_count` so that `count` is O(1).
Every mutating operation is responsible for keeping that cache in sync
with the underlying bit set.
`update(with:)` forwarded straight to `BitSet.update(with:)` without
adjusting `_count`. That method inserts the member when it isn't already
present (it is defined as `insert(newMember).inserted ? nil : newMember`),
so inserting a new member this way left the cached count one too low.
The stale count is observable in a few ways:
- `count` under-reports the number of members.
- `isEmpty` is `_count == 0`, so a set that still contains members can
report itself as empty.
- The `Collection` conformance promises `count` elements but iteration
yields more, which breaks the standard library. `Array(set)` traps with
"Fatal error: invalid Collection: more than 'count' elements in
collection".
In builds with COLLECTIONS_INTERNAL_CHECKS this also violates
`_checkInvariants()`, which asserts `_count == _bits.count`.
Mirror what `insert(_:)` already does and bump `_count` when a new member
was inserted. `update(with:)` returns nil exactly in that case.
`BitSet` has had a `test_update` for this all along; `BitSet.Counted` did
not, which is how this went unnoticed. Add the matching test, following
the same combinatorial style and using `Set<Int>` as the oracle.
Six `- Parameter` entries in `Deque` and `BitSet` write the parameter
name inside backticks (``- Parameter `i`:``). DocC binds parameter
documentation by bare identifier, so these never attach: no Parameters
section is rendered at all, and the entry instead falls into the
Discussion as a literal "Parameter `i`: ..." bullet, which also swallows
the `- Complexity:` callout that follows it.
Dropping the backticks around the name restores the Parameters section on
all six pages. Backticks inside the description text are left alone.
`_toUInt()` returns nil whenever a bound is negative, and the guard maps that
straight to `false`. That is right for a nonempty range -- a bit set can never
hold a negative value -- but an empty range equals the empty set regardless of
where its bounds sit, so `BitSet().isEqualSet(to: -5 ..< -5)` answered false.
Five of the six `Range<Int>` predicates already handle this: isSubset and
isDisjoint clamp, isStrictSubset composes those two, and isSuperset and
isStrictSuperset test `other.isEmpty` before converting. isEqualSet was the
one that did not, which let a bit set be both a subset and a superset of a
range while reporting that it was not equal to it.
As of 2026-09-01, conformance checks are now diagnosed as errors:
```
70 | // Fast paths for when we know elements are all unique
71 | if elements is _UniqueCollection {
| `- error: cannot perform a dynamic cast to a type involving protocol '_UniqueCollection' in Embedded Swift
```
Twenty-five doc comments in the shipping modules name an argument that the
declaration below them does not have, so DocC has nothing to bind the
description to and drops it. Nothing warns about this.
Most of them are renames that the comment did not follow, and the file itself
usually holds the proof: RigidDeque.prepend says - maxCount while its own
Complexity line already says newItemCount, and append(copying:) has three
overloads where the oldest still takes newElements and the two newer ones take
items with the older one's wording.
Three entries name an argument that no longer exists at all and are removed
rather than renamed.
This enables faster iteration over subranges of a BitSet than the `Slice` type’s default indexing iterator. (Depending on the fill factor, the speedup can be as high as 4x — not insignificant.)
DocC keeps changing its link syntax. The latest iteration appears to have trouble dealing with `Self`-taking or `Self`-returning methods, so change our API to replace `Self` with a direct mention of the concrete type. ¯\\\_(ツ)\_/¯
That said, the new compiler also found a large number of mismatching parameter names, and that was really helpful.
The indexes passed to span(after:) and friends on UniqueArray
and RigidArray are modified before return. This adds notes
about the updated values to the method documentation, along
with a few other minor formatting changes (dropping smart
quotes, single-sentence abstracts).
- Stop using unstable import attributes with undefined/flaky semantics in package contexts.
- Freeze the contents of the Collections module. Do not make the new RigidDeque/UniqueDeque types from there, at least not yet. Replace reexported types with typealiases, and reexported protocol extensions with redefinitions forwarding to a common implementation in the original module.
Recent development compiler versions have started emitting errors for previously undiagnosed cases where a conformance in another module is referenced by a source file that does not itself explicitly import that module. Add the missing imports.
We are forced to do this due to an Xcode issue (#364) with underscored module names.
This module remains unstable, despite the removal of the underscore. Directly importing it from outside the package remains unsupported. This module and/or its entire contents may go away or change in incompatible ways in any future swift-collections release.