search

oscbyspro / Ultimathnum

Binary arithmetic reimagined in Swift
Apache License 2.0
9 stars 1 forks source link

Miscellaneous 0.11.0 stuff #92

Closed oscbyspro closed 14 hours ago

oscbyspro commented 1 month ago

I'll use this thread to document various minor changes.

oscbyspro commented 1 month ago

It looks like Swift 6.0 (#91) lets us mark BinaryInteger.exactly(_:) as borrowing. So the borrowing get problem must be solved. It also looks like Bool && @autoclosure () -> Bool consumes the righ-hand-side, which is a bit unfortunate in the borrowing case, but I can perhaps work around it using T.isSigned ? x : false instead of T.isSigned && x.

oscbyspro commented 1 month ago

It looks like I missed one of the Nonzero/ilog2() methods in (#90). I'll remove the version that doesn't reduce the number of failure modes. I.e. Nonzero<some UnsignedInteger> guarantees non-optional results, but the other one doesn't.

oscbyspro commented 1 month ago

I'll also fix a bug in one of my TestKit2 randomness utilities that may or may not cause spurious expectation failures.

oscbyspro commented 2 weeks ago

LiteralInt will conform Interoperable with StaticBigInt as its Stdlib type.

oscbyspro commented 2 weeks ago

I suppose I vaguely remember why LiteralInt doesn't yet conform to Interoperable. The StaticBigInt model is slow, but it might be possible to fix the most common use cases by turning LiteralInt into a small-integer-or-big-integer enum. In that case, it could no longer be conform to Interoperable. I should explore that idea before I change it.

oscbyspro commented 1 week ago

It's about 2x faster to generate random U256 by filling MutableDataInt.Body with an index loop instead of filling UnsafeRawBufferPointer with Randomness/fill(_:). This approach does not change the performance of generating random UXL instances, however.

oscbyspro commented 1 day ago

Apparently, adding the Fibonacci/components() method degrades the performance of an unrelated benchmark by about 2%. It is consistent, so I thought I'd comment on it. My benchmark setup is not particularly serious at this moment, so this observation should probably be taken with a grain of salt, however.