MasterDuke17

The failing NQP test was added in https://github.com/Raku/nqp/commit/0bc35f4476d151920c40e53856f3cae5603f0625 and the failing roast tests were added in https://github.com/Raku/roast/commit/0a7bd6fe68cca065d65b810c2786491da47cfed2. @samcv do you have any opinion on this?

> I think my tests added in [Raku/roast@0a7bd6f](https://github.com/Raku/roast/commit/0a7bd6fe68cca065d65b810c2786491da47cfed2) shouldn't rely on any functionality with padding Ah, but they do. At https://github.com/Raku/roast/commit/0a7bd6fe68cca065d65b810c2786491da47cfed2#diff-b233fcf9e1bc5b933625f6adb84948504ec0e26ec56086da955390d3754a65d5R76-R82 the bitwise ops are only done up to the...

That error comes from [MVM_nfg_codes_to_grapheme](https://github.com/MoarVM/MoarVM/blob/master/src/strings/nfg.c#L250). It's being called from [grapheme_composition](https://github.com/MoarVM/MoarVM/blob/master/src/strings/normalize.c#L618). If I instead have `grapheme_composition` call [MVM_nfg_codes_to_grapheme_utf8_c8](https://github.com/MoarVM/MoarVM/blob/master/src/strings/nfg.c#L255), then there's no error. Raku thinks what it slurps is only one char...

> Could you re-iterate the advantages of this change? Would be nice copy for the RWN when this is merged :-) Yeah, I'll add some more explanation and benchmarks before...

> Does this include the option of using a system gmp as well as building one? (Both matter, the former is probably a requirement for packaging.) Theoretically yes, but I...

> Theoretically yes, but I haven't officially tested it yet (I unofficially did when at first I couldn't get it to use the .a we built). I'll give that a...

> I'm now getting two tests fail in a `make m-spectest`, but that happens with `--has-gmp` and without (I think it's something with a correction to MVM_bigint_div_num for the rebase...

This happens because 2147483647 is the cutoff (https://github.com/rakudo/rakudo/blob/master/src/Perl6/Optimizer.nqp#L1215) where iterating a range can't be optimized (https://github.com/rakudo/rakudo/blob/master/src/Perl6/Optimizer.nqp#L2498-L2579). 2147483647 is probably the cutoff because that's the value below which an integer is...

Still the same behavior for me with `This is Rakudo version 2020.08.2-35-g09e4f2333 built on MoarVM version 2020.08-77-g51d399cb5 implementing Raku 6.d.`

I have a hacked up version of https://github.com/japhb/perl6-bench running locally. I'm not an expert user of it, but once I figure out how to use MoarVM branches I can try...