bulat.ziganshin:

Hello Andrew,

Wednesday, October 14, 2009, 10:28:45 PM, you wrote:

...

I had always *assumed* that there was something like a hundred core developers

only 10 and only in binary system :)))

Simon Peyton-Jones works on "front-end", i.e compiling Haskell down to simple core language, and Simon Marlow does "back-end" i.e. compiling this to actual machine code and dealing with idiosyncrasies of all target systems. and Ian Lynagh is "Build engineer" maintaining releases, bug tracking and so on

there are dozens of people doing interesting things with GHC and some of them are even going to main tree. and there are interns that implements some rather small and isolated parts of GHC. and dozen or so of people porting GHC to their environment, building packages and so on

About 1000 people have worked on libraries on Hackage. That's about two orders of magnitude more than work on GHC.

Hello Andrew, Thursday, October 15, 2009, 12:54:37 AM, you wrote:

Does anybody actually "get paid" to develop GHC? Or is this all people

SPJ, SM and Ian are paid by MS Research. Other people involved in core development are mainly scientists (afaik) -- Best regards, Bulat mailto:Bulat.Ziganshin@gmail.com

Don Stewart wrote:

bulat.ziganshin:

...

Hello Andrew,

Thursday, October 15, 2009, 12:54:37 AM, you wrote:

...

Does anybody actually "get paid" to develop GHC? Or is this all people

SPJ, SM and Ian are paid by MS Research. Other people involved in core development are mainly scientists (afaik)

Perhaps these are the "three people" the speaker was referring to then. Seems like there are plenty of other people involved though...

Besides MSR and the university groups, the IHG funds Well-Typed to do development, as well.

...not least of all these. ;-) But still, it seems the total GHC manpower effort is significantly smaller than I imagined. (Which just makes it more impressive that GHC is so good...) Ah, MSR, the only known Haskell-related enterprise based in the UK. Sometimes I have these delusions that I should go ever there and see if they'll hire me...

Daniel Peebles wrote:

Well-Typed is in the UK too :)

Really? Cool! I wonder where... Oh, Oxford. So also not far from me. However, given that I can't even construct a simple sentence correctly...

On Thu, Oct 15, 2009 at 1:52 PM, Andrew Coppin wrote:

...

Sometimes I have these delusions that I should go ever there and see if they'll hire me...

...I think we can safely conclude that nobody is going to be interested in hiring me. :-/

On Wed, Oct 14, 2009 at 11:06 PM, Bulat Ziganshin wrote:

Hello Andrew,

Thursday, October 15, 2009, 12:54:37 AM, you wrote:

...

Does anybody actually "get paid" to develop GHC? Or is this all people

SPJ, SM and Ian are paid by MS Research. Other people involved in core development are mainly scientists (afaik)

Taking the opportunity to thank very much both Simons and Ian for the work they do and the enthusiasm they show. You guys rock.

Martijn van Steenbergen wrote:

David Virebayre wrote:

...

Taking the opportunity to thank very much both Simons and Ian for the work they do and the enthusiasm they show. You guys rock.

I heartily second that!

I won't disagree with that sentiment. ;-) According to a paper I just read, GHC is "about 150,000 lines of code". It's pretty impressive that such a small team could write and actually maintain such a big thing. (At least, it impresses me.) Speaking of which... One of the things I like about the GHC team is that they keep putting out really interesting papers. Most of them are pretty easy reading, even for somebody as stupid as me. Things like the papers on STM, parallel GC, stream fusion, DPH, and low-level performance tuning of the GHC RTS and STG implementation. For example, I've just been reading a paper which suggests that turning off vectored returns yields a net performance win. One slightly frustrating aspect of all this is that these papers describe interesting techniques or new optimisations, but it's usually quite hard to track down when/if these things made it into the production GHC releases... Sometimes you can tell from the paper (e.g., there was recently a paper about a bunch of things being done to the parallel RTS to remove bottlenecks from it, and IIRC it stated that this work will be released in GHC 6.12.x) Sometimes the GHC release notes will tell you (e.g., GHC 6.10.1 says "there's a new flag to turn on parallel GC", which pretty much means that this is the first version to support this feature). But sometimes it's really hard to cross-reference this stuff. It would be nice if somebody could collect together all the papers about GHC, and which (if any) GHC release versions implement this stuff...