Haskell Platform call for consensus: add transformers and revise the mtl package to depend on it

[Isaac Dupree]

On 11/03/10 03:39, Johan Tibell wrote:
> On Wed, Nov 3, 2010 at 5:50 AM, Isaac Dupree
> <ml at isaac.cedarswampstudios.org>  wrote:
>> mtl-2.0 was just recently uploaded.  If it's easy, I would cautiously wish
>> to wait a month before committing-for-sure to mtl-2 in the platform, and see
>> if there are unexpected transition difficulties.  Any issues that users
>> notice will surely get mentioned on the mailing-lists.  Though, the proposal
>> page (and past experience with transformers/monads-fd) does seem to have
>> investigated the difficulties pretty well; and if the difficulties are not
>> much more than estimated, then they seem worth the costs to me.
>
> We have about a month and a half to figure it out. During that time
> we'll try to migrate all HP packages to mtl-2 (and perhaps some other
> important packages.) If that doesn't cause problems, we can go ahead
> with mtl-2 in the next. Release, if it does cause problems, well then
> mtl-2 will not make it. Put another way, all accepted proposals are
> conditionally excepted on actually being possible to implement. :)

Right, IIRC we've done that in the past -- the "implementer's" role just 
wasn't explicitly acknowledged (in the sense that there's *always* a 
nontrivial chance that something can go wrong, and a corresponding 
"conditional acceptance" that we probably should acknowledge as such, 
rather than blame implementers for not being super perfect. Interesting.).

(thinking aloud..) I realized this proposal is somewhere between "adding 
a package" and "upgrading a package".  It adds transformers.  But it 
also upgrades mtl.  But mtl-2.0 didn't exist on Hackage at the time of 
the proposal -- because the proposal was the place to decide *whether* 
we want that to be the future of mtl.  (Usually a package proposed for 
HP should exist on Hackage first, and this will lead to natural testing, 
compatibility-fixing, etc., already.  Although when we accept a package 
on the condition of lots of API changes -- as we might do with 'text' -- 
this also lacks natural testing-time similar to the mtl-2.0 situation.)

>> (Is there time pressure? Given
>> http://trac.haskell.org/haskell-platform/wiki/ReleaseTimetable I am not
>> sure.  Incidentally it suggests that we technically missed the proposal
>> deadline by one day depending on timezones - Nov 1 vs Nov 2 - which I'm not
>> inclined to worry about.  The page history says Don Stewart wrote that
>> timeline mid-July.)
>
> That's my (and the rest of the steering committee's) fault for not
> calling for consensus sooner when the discussion died out. My
> apologies.

Oh, you're quite right.  I'd forgotten we made some roles that only 
steering-committee members can carry out ( 
http://trac.haskell.org/haskell-platform/wiki/AddingPackages ).  I 
should know, because I *am* one of the steering committee!

-Isaac