Here are some points I'd like to emphasize in addition to the threads above, with the disclaimer that I'm the maintainer of Fay.

Fay tries to be very simple, the code base is small (~4800 LoC). This really lowers the entry barrier for contributions which I think is very important for open source projects. GHCJS is much more complicated since it tries to do so much. For GHCJS I'd be afraid that the developers might eventually abandon the project and then it'd be pretty hard to take over development. Much easier for someone to take over Fay development. I think you can safely expect to find bugs in all compilers, and if you do and you are in a hurry you might have to fix it yourself. Fay has very simple output that is close to both Haskell and JavaScript so it's pretty easy to just add a breakpoint and start debugging.

Even if GHCJS can successfully compile most of hackage, would we want to have these as dependencies in web projects? An output size of 1MiB is nothing when compiling a binary, but for a public website 1MiB can still be quite a lot, add some transitive dependencies and output will explode. Most people don't optimize their packages to have few dependencies, which makes sense since the abstraction usually heavily outweighs code size. So either way you would probably want to write some specific light-weight versions of libraries you want to use, that's one reason both GHCJS and Fay have their own base packages.

One place I think GHCJS can shine is for non standard web applications such as WEBGL games. Nothing stopping you from using Fay for this, but I expect you can really start to leverage GHCJS's threaded runtime here.

I'm not sure if GHCJS or Haste have any out of the box solutions for doing client<->server communication. In Fay you can just serialize any data type automatically on the front and backend to a json format and then deserialize it automatically again on the other end. You can also leverage phantom types (https://github.com/fpco/yesod-fay/blob/master/Language/Fay/Yesod.hs#L41) to typecheck this communication.

It's hard to know how well GHCJS performs in a real web application. FP Complete uses Fay for their IDE and School of Haskell so it has been battle tested, so if you want something that definitely works today I think Fay is the more reliable option.

It is possible that GHCJS will eventually supercede Fay. But if you want something that's easy to use today you can always write Fay, and then convert it to GHCJS later since Fay is a subset of Haskell.

I think GHCJS needs more time to prove itself. I know the developers are hard at work and I hope they will focus on a lot of learning materials. Once you can `cabal install ghcjs` and just get started I think they will see a lot more interest coming their way.

Let me know if you have more questions. If you decide to use Fay in your tutorial I'd be happy to help out, just send me an e-mail!

- Adam



On Wed, Sep 4, 2013 at 2:46 PM, Adam Bergmark <adam@edea.se> wrote:
You might be interested in these two comment threads (and maybe the rest of the comments as well):



On Wed, Sep 4, 2013 at 12:36 PM, Nathan Hüsken <nathan.huesken@posteo.de> wrote:
In my opinion haste is somewhere between Fay and ghcjs. It supports more than Fay, but in difference to ghcjs some PrimOps are not supported (weak pointers for example).

It is a little bit more "direct" than ghcjs, in the sense that it does not need such a big rts written in js.

I like haste :).

What I wonder is how the outputs of these 3 compilers compare speed wise.


On 09/04/2013 11:11 AM, Alejandro Serrano Mena wrote:
I haven't looked at Haste too much, I'll give it a try.

My main problem is that I would like to find a solution that will continue working in years (somehow, that will became "the" solution for generating JS from Haskell code). That's why I see GHCJS (which just includes some patches to mainstream GHC) as the preferred solution, because it seems the most probable to continue working when new versions of GHC appear.


2013/9/4 Niklas Hambüchen <mail@nh2.me>
Hi, I'm also interested in that.

Have you already evaluated haste?

It does not seem to have any of your cons, but maybe others.

What I particularly miss from all solutions is the ability to simply
call parts written in Haskell from Javascript, e.g. to write `fib` and
then integrate it into an existing Javascript application (they are all
more interested in doing the other direction).

On Wed 04 Sep 2013 17:14:55 JST, Alejandro Serrano Mena wrote:
> Hi,
> I'm currently writing a tutorial on web applications using Haskell. I
> know the pros and cons of each server-side library (Yesod, Snap,
> Scotty, Warp, Happstack), but I'm looking for the right choice for
> client-side programming that converts Haskell to JavaScript. I've
> finally come to Fay vs. GHCJS, and would like your opinion on what's
> the best to tackle. My current list of pros and cons is:
>
> Fay
> ===
> Pros:
> - Does not need GHC 7.8
> - Easy FFI with JS
> - Has libraries for integration with Yesod and Snap
>
> Cons:
> - Only supports a subset of GHC (in particular, no type classes)
>
>
> GHCJS
> ======
> Pros:
> - Supports full GHC
> - Easy FFI with JS
> - Highly opinionated point: will stay longer than Fay (but it's very
> important for not having a tutorial that is old in few months)
>
> Cons:
> - Needs GHC 7.8 (but provides a Vagrant image)
>
>
> _______________________________________________
> Haskell-Cafe mailing list
> Haskell-Cafe@haskell.org
> http://www.haskell.org/mailman/listinfo/haskell-cafe



_______________________________________________
Haskell-Cafe mailing list
Haskell-Cafe@haskell.org
http://www.haskell.org/mailman/listinfo/haskell-cafe


_______________________________________________
Haskell-Cafe mailing list
Haskell-Cafe@haskell.org
http://www.haskell.org/mailman/listinfo/haskell-cafe