> > @r = @r , 'd'; > > Okay, that makes sense. So the circular reference I thought I > was seeing is really there, and it's working as designed. > > There isn't anything very useful in this behavior though, is there?
Yes. Here are some relevant results from a search for "self referential" in the #perl6 and #raku logs. 2006 https://colabti.org/irclogger/irclogger_log/perl6?date=2006-01-06#l567 > stevan: dunno about ML (or whatever TaPL uses), but IIRC scheme's letrec is a form of let which allows things to be self referential 2012 https://colabti.org/irclogger/irclogger_log/perl6?date=2012-01-01#l107 my $ff=1; constant @primes = 2, 3, 5, -> $n is copy { repeat { $n += 2 + ($ff=!$ff)*2 } until $n %% none @primes ... * > sqrt $n; $n; } ... *; say @primes[^10]; # OUTPUT«2 3 5 7 11 13 17 19 23 29» 2012 https://colabti.org/irclogger/irclogger_log/perl6?date=2012-10-05#l582 > TimToady: I'd like to get to the point where we can do hamming numbers self referentially constant @hamming = 1, dedup (@hamming X* 2) M (@hamming X* 3) M (@hamming X* 5); 2014 https://colabti.org/irclogger/irclogger_log/perl6?date=2014-11-02#l823 > lizmat: my main reason for not pursuing self referential structures in .raku > lizmat: was that I didn't have a solution for representing them > lizmat: thoughts on that are very welcome .... > moritz: lizmat: I fear it must be printed out exactly like that 2015 https://colabti.org/irclogger/irclogger_log/perl6?date=2015-08-17#l1558 > vendethiel: TimToady: I read the backlog a bit, but I'm not sure I read the answer to this question: do you want "my @a = 1, @a" to be (1, 1, 1 ...) (à la haskell)? > TimToady: not sure it makes sense with 'my' due to assignment semantic, but would certainly be nice if 'constant' could be self referential > TimToady: and we might intuit a lazy if we see the same symbol on the left > and right > TimToady: haskell obviously never needs to have explicit lazy, since > everything already is... . > Also (the immediate point, really) this isn't described very well > in the documentation. I'd say the first focus is best `.raku`. But perhaps it's already optimal. Then again, I imagine "thoughts on that are very welcome" still applies. It's always appropriate to consider improving the doc, especially imo if a draft of an improvement is written by, or in cooperation with, folk who encountered a problem.. > Just to be thoroughly clear, since with a hash this works ... > My first guess was that for arrays, this would work: > But here, while @all_at_once is just: [a b c] the array > @in_stages has a peculiar structure containing circular references. Each hash in a list of hashes is automatically flattened: my %hash = { :a, :b }, { :c, :d } say %hash; # {a => True, b => True, c => True, d => True} Whereas each array in a list of arrays is not: my @array = [ :a, :b ], [ :c, :d ]; say @array; # [[a => True b => True] [c => True d => True]] This is the "class dependent" aspect of the semantics of `=`. You could create a new array type SteamRolled and bind that: my @array is SteamRolled = [ :a, :b ], [ :c, :d ]; say @array; # [a => True b => True, c => True d => True] Then using the `,=` op would do as you desire. > say @r; # ( [<CIRCULAR REFERENCE> d]) "Thoughts are welcome". :) > And not to quibble too much, but if you didn't want this to do > that to the array, then it *is* "mangling" it. I hear you don't feel sufficiently heard about it being "mangling", but I just wanted to be clear it was not doing what you may have thought was going on, namely exhibiting an unintended bug. > I think that means you're never going to want to do this. > It could be it should even be warning or erroring out. Per the above IRC discussions, self referential data is useful. That said, Larry's comment above about perhaps outlawing it for `my` variables and instead only allowing it for `constant`s is the way to go. > > It isn't really anything to do with "class dependent" behaviour. > > In which case, the documentation could use some improvement. I am always supportive of at least discussing improvement of any aspect of Raku, especially in response to someone experiencing a surprise or confusion. My comment above was meant to be about `,=`. The difference in behaviour between arrays and hashes for what they do when a list is assigned *is* class dependent behaviour. > (Actually, saying that some behavior in Raku is class dependent is > kind of redundant... it's all class dependent in one way or another.) Right. But I think *something* needs to be said about how the LHS gets to decide what happens when you use `=` based on the class of the LHS.. This is different from what happens with, say, `:=`, where the overall behaviour is essentially the same regardless of the class of the LHS. > > Perhaps more importantly, what improvement do you propose? > > Well, I just mentioned a few. I always want to see more examples > in the docs. Write them, collaborating with those working on the doc to improve it, and I predict your wishes will come true. :) > What else would you use ,= with besides a hash? Any data structure / composite class. > It could be my confusion here is another form of something else > that keeps tripping me up. Merging two hashes together is > pretty simple: > > my %h_all = %h1, %h2; > > But doing the same thing with arrays is going to require slipping > the values out of the containers: > > my @a_all = | @a1, | @a2; > > So the two cases aren't all that parallel. Exactly. Hashes autoflatten by default, because that's the only sensible thing to do. Arrays do not, because doing flattening by default makes it awkward to build data structures. ---- To be clear, imo someone's sense that something isn't right is always entirely valid. and usually a useful thing to bring to the table to improve both their own understanding and Raku. What wasn't so clear to me was how best to discuss it, so I've ended up writing very dry stuff. I hope that I haven;t sounded dismissive, because I am absolutely not intending to do aught but encourage you to stay with what you are journeying thru in regard to the impact of trying `,=` with an array, and having in mind its use with a hash, and figuring out how to improve Raku semantics, gists, doc, or whatever else is appropriate, Much love to all, raiph
