- update to 110.99.4:
* MLRISC: Modified the way that zero-extension (ZX) is handled
when the source bit width is smaller than the target.
Previously, the code assumed that the high bits would be zero,
but as demonstrated by issue #272, this is not guaranteed, so
we now mask out the high bits.
* Added the NativeInt and NativeWord structure aliases to
provide a portable way to refer to the native numeric types in signatures.
* Reimplementation of the Random structure to use the Mersenne
Twister algorithm. There are both 32-bit and 64-bit versions
of the generator (which one is included depends on the target platform).
* Reworked the code for determining the maximum hash-table size
(used in the HashSetFn and HashTableRep modules) so that it
does not depend on Int.int and Word.word being the same size (an
issue for MLton). Also split that code out into the internal
MaxHashTableSize module.
* Added support for the end-of-line assertion ("$") and full
support for intervals to the Thompson engine in the RegExp library.
* fix issue 87: ml-makedepend trashes dependency file on error
* fix issue 256: Initial numbers generated by Util/random.sml
have unexpected regularities
* fix issue 258: Thompson NFA regex engine doesn’t advance stream position when matching
* fix issue 260: 64-bit word literals are parsed incorrectly on 32-bit systems
* fix issue 264: Fix VectorSlice.all and ArraySlice.all
* fix issue 266: config: restore PPC for 10.4 and 10.6
* fix issue 272: Calling C function via NLFFI binding may give
result outside range of its C return type
* fix issue 278: CharBuffer.addSlice does not properly handle starting offsets
* fix issue 279: HashSetFn’s calculation of `maxSize can cause
Overflow during functor instantiation
- remove smlnj-kernel-6.x.patch
* fixed in upstream: f07eb71cca7c647d28dbaf195f1e61e7e23cfc6f