On 11/11/2011 16:40, Colin Paul Adams wrote:

Simon> On 11/11/2011 15:06, Bas van Dijk wrote:

...

I'm not sure to what extent we want to do this. So far we've only had to copy a few individual functions and fix export lists. The change to Num recently meant that we have diverged from Haskell 98 and Haskell 2010 in a fairly fundamental way, and to recover compatibility we would have to copy a lot of code into haskell98 and haskell2010.

What's wrong with copying a lot of code? It's not code that's going to need maintenance is it?

I don't have a big problem with it, no. It's just the time required to actually do it - depending on how deep the divergence is, you might have to pull a lot of the GHC.* modules in the base package over, and the intermodule dependencies can be quite tricky. There are the lesser issues of the extra compile time and disk space for the duplicated library code too. Also, the more copying you do, the less it becomes possible to have a program that consists of a combination of haskell2010 modules and modules that import the base package. Essentially it becomes impossible to have libraries written in pure Haskell 2010. Again, I don't know how much we care about that. Cheers, Simon

On Fri, Nov 11, 2011 at 5:00 PM, Bas van Dijk wrote:

On 11 November 2011 17:50, Simon Marlow wrote:

...

Also, the more copying you do, the less it becomes possible to have a program that consists of a combination of haskell2010 modules and modules that import the base package.  Essentially it becomes impossible to have libraries written in pure Haskell 2010.  Again, I don't know how much we care about that.

What is the purpose of haskellYYYY if it becomes impossible to combine (either directly or indirectly) base and haskellYYYY?

Is it just so that book authors can base their books on a stable API? If so, why can't they just base their books on some version of base exploiting the PVP?

Cheers,

Bas

My understanding was that base is large and fairly intricate (contains use of ExistentialQuantification, RankNTypes, and so forth) whereas haskell2010 is comparatively simple to implement, so there's more hope that it could be provided by a non-GHC compiler. In practice I suspect that it's not often used in this way, but I like the idea. Ben