Choosing the new Stan compiler's implementation language

[Bob_Carpenter]

@bgoodri

I think you are making too much of this.

Probably. I just want to make sure we’re doing the right thing.

@bgoodri

If A is under BSD and B is GPL but includes A, then anyone is allowed to download A from its website and do whatever with it.

Sure, anything consistent with the BSD. That means keeping the original license. What you can’t do is download A, then redistribute it under a different license, such as GPL.

@bgoodri

And anyone is allowed to download B and redistribute derivative works of B under the GPL.

Right. As long as A keeps its BSD license.

My only concern is what we should put at the top level to describe A. Should we just say that A is GPL-ed or should we say A is GPL-ed but it relies on components under other licenses compatible with GPL. Exactly what @increasechief points out here:

@increasechief

"All material here is released under the GNU General Public License. All material that is not executable, including all text when not executed, is also released under the Creative Commons Attribution 3.0 International (CC BY 3.0) license or later.
…
Like almost all software today, this software depends on many other components with their own licenses. Not every component is (GPL >= 2), but all components are FLOSS.

I had to look up the “L” there—it’s just “free” in another language (“libre”). Seems redundant to me.

We can’t quite do this because our logos aren’t consistent with CC BY, but this is pretty much what we say now.

[coatless]

Folks, can we move the discussion of an all-in-one installer outside of the compiler implementation language thread? I feel like it is getting off track very fast.

[Bob_Carpenter]

I think we can close this discussion altogether as we’ve decided to go with OCaml.

[Bob_Carpenter]

@ChrisChiasson

edit: whoops, I didn’t read the last post before I replied… sorry

No worries. No discussion’s ever truly closed with open-source software.

[idontgetoutmuch]

An implementation already exists in Haskell: https://github.com/diffusionkinetics/open/blob/master/stanhs/lib/Stan/AST.hs

This was presented at StanCon 2018: https://glutamate.github.io/stancon/stancon.html

If you are thinking of OCaml, why not Haskell?

[torkar]

If you search the thread for Haskell you might get background information on why Haskell was not picked.

[idontgetoutmuch]

There was some information but I couldn’t find a summary or the rationale for the final choice just a bald statement. But the choice seems to have been made.

[seantalts]

1. That is not an even half-complete implementation of a Stan parser (which is also less than half of a complete compiler). See how it only has ints and reals, hardcodes all functions, doesn’t represent statements at all, etc? Check out this Stan AST in OCaml to see what a more complete version should contain.
2. The search “feature” on discourse is absolutely awful, my biggest gripe with it. Here are some posts where it is discussed in a lot of detail:
   Choosing the new Stan compiler’s implementation language
   Choosing the new Stan compiler’s implementation language
3. In another post, I made a request for someone who knows both OCaml and Haskell to attempt to compare the two languages using the desiderata in the original post, but no one was able to and the general consensus seemed to be that OCaml is basically almost as good, even to a Haskell person.

[idontgetoutmuch]

Thanks for replying. And sorry to be a Jonny-come-lately to the discussion when I now see there has been a lot of discussion. It’s a shame I didn’t see your request for the comparison. I am wearing an OCAML t-shirt as I write this :-) I don’t think it’s worth writing anything now the choice has been made.