[SR-4634] Error in exact floating point initializers

[martinr448]

Previous ID	SR-4634
Radar	rdar://problem/31836766
Original Reporter	@martinr448
Type	Bug
Status	Resolved
Resolution	Done

Additional Detail from JIRA

| | | |------------------|-----------------| |Votes | 1 | |Component/s | Standard Library | |Labels | Bug | |Assignee | @natecook1000 | |Priority | Medium | md5: e3cafae41d33e915edd76d83f2a2b2b5

is duplicated by:

• SR-4800 Float.init?(exactly: UInt32) does not return nil for UInt32.max or other un-representable values
• SR-7056 Float(exactly: Int32) and Double(exactly: Int64) do not work properly

Issue Description:

The exact conversion from integers to floating point types succeeds even if precision is lost. Example:

if let f = Float(exactly: Int64(9000000000000000001)) {
    print(f, Int64(f))
    // 9e+18 9000000202358128640
}

The reason is a wrong preprocessor condition at FloatingPointTypes.swift.gyb:L840:

%   if srcBits < SignificandBitCount:
    if ${That}(self) != value {
      return nil
    }
%   end

which means that the test it not performed if the integer has more bits than the mantissa of the floating point type.

It should be if srcBits >= SignificandBitCount: or if srcBits > SignificandBitCount: instead.

[airspeedswift]

@swift-ci create

[swift-ci]

Comment by Ian Partridge (JIRA)

https://github.com/apple/swift/pull/11311 is open to fix this.

Looking forward to getting this fixed as it blocks improving the test coverage for NSNumber bridging in swift-corelibs-foundation.

[xwu]

Looks like this was resolved in #12739 (by Nate).