April 2020 - Haskell-Cafe

Re: [Haskell-cafe] When did it become so hard to install Haskell onWindows?
by Anthony Clayden 26 Apr '20

26 Apr '20

Thank you Ben > I'd urge everyone to keep in mind that our Windows support exists almost entirely due to the efforts of Tamar; we all owe Tamar a debt of gratitude for all of the work he has done over the past years. We could certainly use more people who are able and willing to contribute resources to help improve GHC's Windows story. I've tried and tried, to the point where I was clearly considered a nuisance. I've now given up. And GHC has just got worse since. I suggest Tamar downloads/installs Hugs [*], then provides (and documents) something as easy as that. Wasn't there a GHC equivalent to WinHugs? That is: something as easy for casual/recreational users/learners of Haskell, not aimed at Microsoft afficianados. [*] Sadly, I apologise in advance for this: https://mail.haskell.org/pipermail/hugs-bugs/2018-July/001914.html >>* Increasingly, GHC HQ is a cult/elite that doesn't want any new members.* > To the contrary, we both want and need new contributors! Not my experience. I was told in polite but unambiguous terms to stop contributing. > I would ask anyone who feels that GHC development is exclusionary to please be in touch; I am very interested to know how we can improve. I'd like to cite as evidence the survey results Francesco links to: I don't think I ever got invited to contribute -- where would that have been? There wasn't an invite on the Hugs-users list. Note I don't use Reddit because I'm not a brogrammer. The survey is kinda self-fulfilling: most people don't use GHC on Windows, because it's too hard to install. GHC is exclusionary against casual/recreational Haskellers (who are more likely on Windows) and those more interested in using it to explore PL Theory. This is the opposite of how it was when I first grew to love Haskell around a decade ago. Just look at the buzz on the cafe back then. AntC

5 4

Re: [Haskell-cafe] When did it become so hard to install Haskell onWindows?
by Anthony Clayden 26 Apr '20

26 Apr '20

@lonetiger, I take it you've never had to support software installs on shared machines, or students' machines where you can't be too sure of the config? The reasons against using powershell and curl (or equiv) I would have thought are pretty obvious: you don't know what you're getting/what it might be downloading. You might mis-type some command and cause havoc on your machine or on your network. With Hugs you download the packed-up .exe; you disconnect from the internet and run virus checks; you run the install while your machine is still in quarantine. With Hugs install you (usually) don't need to go into Admin mode. Just because some long-winded process is "completely standard on Windows", doesn't mean it's fit for use. That's a general learning about stuff from Microsoft. AntC ➢ Then I have to know what powershell.exe is, use an administrative prompt, and enter scary commands in it. > Powershell has been the standard shell in Windows for well over the past decade. Every single script from Microsoft or third parties come with powershell for automation. It’s understandable that you may not know it since your primary platform isn’t Windows. But it’s been included in every single Windows version for the past 13 years. > An administrative prompt is nothing different than running sudo or clicking on that installer that you *assumed* not to be scary because you didn’t see the actions it was performing. That scary looking command is nothing but a curl command allowing the one time execution of a script from a remote source. As in a script that’s not physically on your machine. > So what exactly makes this scary? Is it because > Set-ExecutionPolicy Bypass -Scope Process -Force; [System.Net.ServicePointManager]::SecurityProtocol = [System.Net.ServicePointManager]::SecurityProtocol -bor 3072; iex ((New-Object System.Net.WebClient).DownloadString('https://chocolatey.org/install.ps1')) > Is more verbose than > curl -sSL https://path.to.some.script/ | sh > or because the technologies used while completely standard on Windows aren’t known to the casual user?

3 2

Need help with advanced type-level programming
by Evgeny Permyakov 26 Apr '20

26 Apr '20

Consider following code data Tag :: (Symbol -> Constraint) -> Symbol -> * where Tag :: (KnownSymbol s, con s) => Tag con s data TagList :: (Symbol -> Constraint) -> [Symbol] -> * where LNil :: TagList con '[] LCon :: Tag con s -> TagList con ss -> TagList con (s ': ss) data TagSet :: (Symbol -> Constraint ) -> [Symbol] -> * where SNil :: TagSet con '[] SSin :: Tag con s -> TagSet con '[s] SCon :: CmpSymbol ht mt ~ LT => Tag con ht -> TagSet con (mt ': ts) -> TagSet con (ht ': mt ': ts) How one would write a function converting a TagList into a TagSet , dropping duplicate tags ? I tried a few approaches, but failed. I clearly lack understanding how to fit it into GHC type system. I know there is a library type-level-sets, but it isn't what I would like and I don't understand how it fits into GHC type system, so it appears to be magical to me. Similarly, it would be greatly appreciated if someone could show how to implement following typeclass class CmpTag (s1 :: Symbol) (s2 :: Symbol) where cmpTag :: Tag con s1 -> Tag con s2 -> OrdTag con (CmpSymbol s1 s2) s1 s2 data OrdTag :: (Symbol -> Constraint) -> Ordering -> Symbol -> Symbol -> * where TLT :: CmpSymbol s1 s2 ~ LT => Tag con s1 -> Tag con s2 -> OrdTag con LT s1 s2 TEQ :: (CmpSymbol s1 s2 ~ EQ, s1 ~ s2) => Tag con s -> Tag con s2 -> OrdTag con EQ s1 s2 TGT :: CmpSymbol s1 s2 ~ GT => Tag con s1 -> Tag con s2 -> OrdTag con GT s1 s2

3 3

Getting the profile results for GHC when running in Cabal.
by Sean Parsons 25 Apr '20

25 Apr '20

Hi there. I've built a GHC with profiling enabled as guided by the wiki and that works fine standalone. I found the option for specifying a certain GHC to Cabal. But I'm a little baffled as to how to get the results of the GHC profile when building a library with Cabal. My goal is to try to get a baseline profile, then have a poke at the innards of GHC and then see if whatever I've changed has made an improvement to the build speed. Does anyone have any idea what the best way to do that is? Thanks, Sean.

2 1

base library on hackage
by Ben Franksen 24 Apr '20

24 Apr '20

It seems the latest version uploaded there is 4.12. Where can I find docs for later versions?

3 2

parsing Haskell expressions
by Johannes Waldmann 24 Apr '20

24 Apr '20

Dear Cafe, for a language that has (subset-of) Haskell expression syntax, but different global declaration syntax, we are using parseExp from haskell-src-exts. I have two question on this. I am asking here because https://github.com/haskell-suite/haskell-src-exts#maintenance says that it's no longer active. (1) can we make parseExp self-delimiting? The type is parseExp :: String -> ParseResult (Exp SrcSpanInfo) but I'd want "maximum munch" and the part that was not eaten, as in :: String -> ParseResult (.. , String) I tried parseExp "x } foo bar" ParseFailed (SrcLoc "<unknown>.hs" 1 3) "Unexpected }" so I could extract the error location, and then re-parse that prefix. That's feels a bit strange. (2) perhaps we don't need (1) if we do the chunking on the outside. E.g., we run a lexer and look for delimiters (keywords of the host language that then are forbidden in inner expressions). We want to use lexTokenStream :: String -> ParseResult [Loc Token] and then I'd want parseExp' :: [Token] -> ... Else, we'd have to reconstruct the slice of the original input, which again, feels a bit strange and (because) it duplicates work. It is suggested to use http://hackage.haskell.org/package/ghc-lib-parser instead. Would this help with the above? But then of course (3) is there a transition strategy (e.g., wrapper modules)? Thanks - J.W. NB: this is the issue from the project's tracker: https://gitlab.imn.htwk-leipzig.de/waldmann/cyp/issues/45

2 1

Re: [Haskell-cafe] Better names for
by Olaf Klinke 24 Apr '20

24 Apr '20

> > fmap f. If I could re-define Haskell from scratch, I'd probably > >name > > the Functor type class function 'map' > > As it was in Haskell 1.4 That's interesting, I did not know this. Thanks for sharing. Are there any public records as to why the decision was made to re-name 'map' to 'fmap'? I can not find any mention in the Haskell98 report(s). There is a stackoverflow thread (#6824255) which mentions paedagogical reasons, simplifying error messages. I sometimes use the current naming 'map' to specialize a polymorphic function to work exclusively on lists where any functor could have been used. Other than that, I can not quite see the use in keeping 'map' around when it is equivalent to fmap. What reasons are there apart from backwards compatibility? Are there re-write rules that make 'map's (operational) semantics different to 'fmap'? Olaf

3 2

Does GHC recognise that CmpSymbol s1 s2 ~ EQ implies s1 ~ s2 ?
by Evgeny Permyakov 24 Apr '20

24 Apr '20

By semantic CmpSymbol s1 s2 ~ EQ should be equivalent to s1 ~ s2 for s1 , s2 :: Symbol Does GHC recognize it? If not, is there some trick to obtain evidence of the second constraint from the first ? I'm talking about context of type class instance declarations.

2 1

Re: [Haskell-cafe] Better names for
by Olaf Klinke 23 Apr '20

23 Apr '20

> I've heard several times that some Haskell type classes and methods > are > ill-named. > > For instance, a functor can be both covariant and contravariant, > while > Haskell `Functor` typeclass i just covariant. Would `Cofunctor` be a > better name for it? And what about the `fmap` method? Would it be > better > to call it `comap`? How about `Bifunctor`, `Profunctor` and others? Hi Marc, part of the answer is: For historical reasons and to preserve backwards compatibility. The term "Functor", however, comes from category theory. MacLane [1] defines the term "Functor" to refer to covariant functors. The notion of "contravariant" is introduced via the concept of "opposite category" [2] which has the same objects but direction of arrows reversed. Thus a contravariant functor F: C -> D is just an ordinary functor C -> D^op. Category theory has no name for the action of a functor on arrows, Haskell had to make up a name, which ended up to be "fmap". Sometimes the notation F(f) is used in category theory where in Haskell we write fmap f. If I could re-define Haskell from scratch, I'd probably name the Functor type class function 'map' and omit the map function on lists, since it is the same as the Functor instance. As for your suggestion of 'comap': The prefix co- in category theory is mostly used in conjunction with the opposite construction [3], whence a category theorist would believe 'comap' to mean the opposite of 'map'. Regards, Olaf [1] Saunders MacLane: Categories for the Working Mathematician. ISBN 7- 5062-6008-5 [2] See Op in Data.Functor.Contravariant [3] See categorical concepts like limit vs. colimit, cone vs. cocone.

2 1

Better names for
by Marc Busqué 22 Apr '20

22 Apr '20

Hi there! I've heard several times that some Haskell type classes and methods are ill-named. For instance, a functor can be both covariant and contravariant, while Haskell `Functor` typeclass i just covariant. Would `Cofunctor` be a better name for it? And what about the `fmap` method? Would it be better to call it `comap`? How about `Bifunctor`, `Profunctor` and others? Thanks, Marc Busqué http://waiting-for-dev.github.io/about/

3 3