Zoffix Znet

This document is the June, 2018 progress report for The Perl Foundation's Perl 6 Constant and Rationals Grant.

Tangibles

The bonus deliverable "Perl 6 Numerics" Language documentation page was merged to master. It describes all of the available Perl 6 numerics, their interactions, suitability, and hierarchy.

The bulk of work on constants also has been merged to post-release-2018.06 branch, which will be merged to master after this month's release. I wrote 200 spec tests, available in S04-declarations/constant-6.d.t spec file, and about 500 words of documentation to cover this work.

Rationals

Since my last report, I first continued working on Rationals, focusing on three pieces of work that currently reside in car-grant-unreduce branch

1. Fixing the rare data race and doing some optimizations
2. Fixing bad math in some ops with Zero-Denominator Rationals (ZDRs)
3. Attempting a trial implementation where ZDRs are marked with a role, allowing us to improve performance of some operators.

The (1) was successful and I already was able to optimized some ops due to Rationals being always reduced now: == was made 28% faster and === was made 52% faster. I also made creation of Rationals 19% faster and argless Rational.round 4.7x faster (used by .Str and .base).

The plan for (2) was to try normalization to <1/0>, <0/0>, <-1/0> and the hope was that alone would fix all math problems. However, after that change some issues still remained. Also, doing this normalization created a new problem where code like say 42 / 0 would throw an Exception with message "division of 1 by 0 attempted" which is quite confusing for users who did not internalize that / op with Int objects is really just a Rat constructor. The 6.c spec actually covers this exact scenario and expects the thrown Exception to report value 42 for numerator, thus blocking this change.

The first attempt at (3) ended with a dead-end. Marking ZDRs with a role created an extra dispatch ambiguity with some operators like cmp. We already had to disambiguate that op with is default marker to disambiguate between Rational and Real candidates, so now we'd need an is default of defaults trait :). I gave up on this for now, but will likely revisit and try again.

As you can see, the Rationals work was a mixed bag, so to experiment with the problem space a bit I decided to implement my own rationals outside of core from scratch. The work is available in zoffixznet/perl6-Rashnl repo, which might become an ecosystem module if it offers significantly better rationals.

I doubt the entirety of Rashnl could be made core, as my current approach does not involve separate roles at all and just has a single class with a flag for fattiness that marks what in core is a FatRat type. The primary purpose of that work is to experiment with code and learn some lessons that could be applicable for core Rationals and achieve the goals of this Grant.

Due to that detour for the Rationals work and more time to think required, I switched to the work on constants for the time being…

Constants

The bulk of the work on constants is now completed and has been merged to post-release-2018.06 branch, which will be merged to master after this month's release.

I wrote 200 spec tests, available in S04-declarations/constant-6.d.t spec file, about 500 words of documentation, and 8 commits of the implementation.

Except for native types, the type constraints are now enforced on constants. The auto-coercive behaviour for %- sigilled constants was blocked by one test in 6.c specification and so that behavior has been added to 6.d language and currently requires the use of use v6.d.PREVIEW pragma to enable (6.c behavior is to simply throw without any attempts to coerce, making constant %foo = :42foo, :70bar fail, because it's a List).

One of the remaining things for constants is improving error reporting for unsupported constructs. Most likely I will implement support for coercers—even though they're currently not available on variables, there should be no problem with executing them during constant creation.

The other remaining item is natively-typed constants. Currently, my int const foo = 42 actually does not create a natively-typed constant at all. I hope the knowledge I'll gain while implementing the Grant's bonus work for support of native-typed unsigned attributes will help me in implementing the natively-typed constants as well.

No Report for July / Further Work

As has been discussed with and approved by my grant manager, I plan to now take a month off working on this grant, so there will be no report in mid-July, and the next report will be in mid-August. I plan to take this time to work on resolving Rakudo's outstanding spectest issues on Windows and possibly resolving some of the open Issues with our community websites.

After that period, I will resume working on the grant, finishing the remaining work on constants, and then going back to Rationals, and finally finishing the bonus work with the natively-typed entities.