***michel_mno_afk is now known as michel_mno
***michel_mno is now known as michel_mno_afk
<ArneBab_>sadly this seems to backfire on guile, too: “Like several times <ArneBab_>before, GUILE developers promised to get actively involved only to drop <ArneBab_>out of the discussion once they were provided with instructions, an <ArneBab_>up-to-date branch/source to work with and current problem descriptions.” <davexunit>ArneBab_: I think I can guess who wrote that <davexunit>mark_weaver, civodul, wingo: new action in lilypond land, lilypond may be removed from Debian testing soon. see above ^ <davexunit>ArneBab_: well, now that I've read everything, looks like they are downgrading the severity <ArneBab_>but this should really, really not happen. It gives the impression to all projects that using Guile is a technical liability <ArneBab_>this will take a bit of time. I have to build stable-2.0 first <ArneBab_>if this bug is fixed in 2.0.12, how much work would it be to do the 2.0.12 release soon? <ArneBab_>davexunit: I don’t know how he acts, but if he’s acting strange, I kinda understand it. He likely thought that using Guile would be a great asset for GNU programs, but it proved to add work due to the porting issues. <ArneBab_>I fear that he’s thinking “why did I ever use Guile when I could have invented a small scripting language myself”. This is not the kind of user story which helps adoption. <ArneBab_>but I don’t want to put blame on anyone: I could have invested time to fix this, too, and I did not get push it through, either. I just say it to give an analysis: this is dangerous. <davexunit>ArneBab_: the person in question likes to twist words and say bad things about guile any time is brought up. <davexunit>not to say that he doesn't have some valid points, but he's more of a troll than you think. <ArneBab_>I just care about the valid points and the effects this has on Guile <ArneBab_>I did not like the tone in which he wrote the message <ArneBab_>but as long as lilypond does not work with guile-2.0 that leaves an open flank for guile <ArneBab_>it makes it hard for Guile to show its strengths <ArneBab_>davexunit: can you estimate how much effort is involved in making a release? <davexunit>make distcheck, updating NEWS and ChangeLog, mailing list post, etc. I imagine <ArneBab_>I’m wondering about the etc. - and the ChangeLog might be quite some work, too <ArneBab_>mark_weaver: can I do anything to help you create a 2.0.12 release? <davexunit>ArneBab_: I don't think rushing a release is going to help anything. <ArneBab_>I think a higher frequency of releases helps a lot: It gets improvements into the hands of the people who need them. <ArneBab_>as long as there is no release, it is impossible to refer to a version which supports something I need within distributions. <ArneBab_>I know that there are things which just need time to get them right <ArneBab_>but there are also many other small improvements which get delayed <ArneBab_>it seems that at least in january this wasn’t fixed <jmd>I'm rewriting makeinfo in Guile. <stis>\\me is trying to understand ordered sets and a sane notion of a complement for them <jmd>The complement of a set is also a set. <stis>well say that the world is not decided, then you have something different <jmd>To define it, however you need to know the universal set. <jmd>Sorry. "define" was the wrong word. <stis>lookup the bitwise operators e.g. lognot is the complement <stis>and consider a nuber as a set of bits <stis>It's not that hard to do complement on unordered sets <jmd>I don't see that being ordered of unordered would make a difference. <stis>It does, if you hav A u c(B)m c(.) the complement <stis>you can write it c(B-A) just as well, but then the ordering in A is lost <jmd>Is the set of all sets a subset of itself? <stis>well mathematicall not defining the world has clear problems <stis>but if your Set of sets is all sets that is constructued from finite atoms and finite number of set operations, you are safe <stis>the set of all set is not definable in that space <stis>e.g. the general one with the famous undecidability issues <stis>you see that that in that construction you can count up all combinations, set of all sets contains the rational numbers <stis>guiles bitwise operations does as well not define the world but still manages to work quite well <stis>play with it, kind of amazing <jmd>I was suprised to learn that guile saves real numbers as IEE754. <jmd>Doesn't that defeat the advantages of the arbitray precision arithmetic? <civodul>jmd: there are "inexact" and "exact" numbers <civodul>IOW inexact numbers are not arbitrary-precision <stis>civodul: any hope of getting higher precitions into guile? <civodul>nobody's working on it, but an extension using MPFR and MPC would be welcome <stis>civodul: I would think that complementable sets wouled me a nice addition to guile WDYT? <stis>it can be a higer order lib and use any ordered or unordered set operations as a base and generate complementable set operations associated with those imputs <stis>I find these things interesting anyway. <stis>complement(C) = World - C, with World undecided <ijp>probably sets which also carry around a univesal set <ijp>I'm going to stop paying attention right now <ijp>because it makes no sense <stis>in bitwise operations you do not define the number of bits <stis>still lognot which is a complement for bitsets works just fine <ijp>1. what is the result of (lognot (set 'a 'b)) <stis>well all atoms defined on a computer is countable, you just map them to bits <ijp>fine let me ask a different question, what possible use has this? <stis>well it parameterizes on the world. <ijp>how can you give an answer if the world is generic? <stis>for a complementable set A, with a world X you just do X n A <stis>that will then be an ordinary set <ijp>at that point, all you are doing is flipping and currying the complement function, and needs no special support <stis>and the nice thing is that you can do all your set operations first <stis>and in the very last step apply the world <ijp>now we are getting somewhere: you want a set api with a universal set <ijp>which is what I thought you wanted at first, but then it seemed not <stis>well you can solve this by delaying all operations bt that can be costly <ijp>then you still don't actually need special support <stis>so think of the tool as a way to do smart algebra sa that the delayed expression becomes superfast when you actually have the world <ijp>you define your own complement function, and make the universe a parameter <stis>no, when I do the set operations like union etc the world iss not availabl <stis>so you delay the evaluation <ijp>and now you've lost me again <ijp>why? that's like doing modulo arithmetic without a particular modulus <stis>consider A u B, denote a prefix <stis>then consider X-A u X-B = X-(A n B) <stis>so when doing u, you perform the A n B and store that <ijp>stis: why should set operations build computations, that are ran at the end, rather than make the modulus a parameter? <ijp>the code would look *almost* exactly the same <ijp>and one requires only a slight tweak to code that already works <stis>well you precompile A n B above so that when you apply your world X <stis>it is much faster then doing all set operations all over again for each world <ijp>and most importantly, why should this be added to guile <stis>well, I just through out the idea, it can just as well be a library. I just find it a natural adition to the set operations we already have <ijp>it's not an addition, it's a rewrite <ijp>you can't do it additively <stis>well you can take a set of ordinary set operators and transform them to a set operators including the complement. <stis>so you can have it as a nice higher order function <stis>I view that as addition, but your mile may vary <ijp>compare that to my two line solution <stis>ijp: the hard part is to get complements to behave naturally for ordered sets and setmaps mimicking assoc lists <stis>Else for unordered sets, it's just mostly boiler plate <stis>and you can do it by hand just fine. <ijp>you mean so that complements of ordered sets are ordered sets? that's just false <ijp>at least under the same ordering relation (inb4 well-ordering theorem) <stis>well the complement is unordered but lives in a space of it's own <stis>if you take the complement of a complement set the ordering will be important <stis>this measn that you cannot represent complemetes sets as c(S) <stis>you need to represent them as A u c(B) <stis>in an unordered setup this would be c(B-A) <stis>but now the ordering of A is important <ijp>actually, what are we talking about by "ordered set" <stis>it's not super difficult to get complement working for ordered sets, but normal people would probably not pull that off <stis>if A is ordered then the elements of A could be placed on a list that defines theor order <stis>let the poerator l take a set to this list l(A) <stis>for union A u B = you define the new order as <ijp>okay, that's just weird <stis>well consider sets beeing lists <ijp>I was just going to clarify my earlier remark (which was wrong from a maths POV, as I was thinking of something completely different) <stis>err consider assoc lists <ijp>why are you not just supplying the ordering function along with the universe <stis>A u B is typically just an append, <ijp>and have the universe totally ordered <stis>right a list is such a function and is general enogh for totally ordered sets <stis>you can take a set library and turn it into a ordered set library by adding such a list <stis>I was planning to write such a higher order function as well <stis>take for example wingos intsets. that's just for integers <stis>we could design a general set by using the hashes of the elements from the intset <stis>but that would not be ordered <stis>so you add this list that defines the order and now you can have a functional set that is thread safe <ijp>(none of this discussion should be construed as endorsement) <stis>err from the instset -> to the intset <stis>srfi-1's set operations for lists are ordered last time I checked