hslibs/ and hugs98

[Alastair David Reid]

>>>>> "Sigbjorn" == Sigbjorn Finne <sof@galconn.com> writes:

> Hi everyone, Alastair's recent checkin gave Hugs98 access to a great
> many of the modules provided by fptools/hslibs, which is great. What
> about adding these derived modules to hugs98/lib in the CVS
> repository?
>
> I know very well there are downsides to replication, but having a
> CVS tree that's self-contained does have its merits.

I lean against this because:

1) It will encourage divergence to happen again.
   The only way I know to avoid this is to run off a single
   set of (ifdefd) sources - and even that isn't perfect.

2) Johan's recent checkin of the hierarchial module names should
   put us very, very close to being able to use the new library
   repository.  (SimonM has the Prelude and the lang stuff 
   pushed over to the new repo and working with ghc (not committed
   yet, I think).  If his batteries last, I'm sure he'll have all the
   other libs done by the time he steps off the plane tomorrow.
   If I'd realised he was this close, I'd have held off on my commit.)

I find the "single repository" argument unconvincing.  (One could also
argue that not having to run cpp, sh, perl or whatever my script
depends on is important - I totally buy that argument for Hugs
releases but not for anyone using cvs.)

There may be some merit in insulating ourselves against any breakage
that may be introduced in the main library.  But Andy Gill suggested a
simpler way of doing that based on nightly (or more frequent) testing
and a "last working/stable version" tag which is automatically
advanced by the testing mechanism.  (We also hope that social pressure
will keep the rate of breakage significantly lower than in, for
example, the GHC module - mere technical approaches can only go so
far.)

-- 
Alastair Reid        reid@cs.utah.edu        http://www.cs.utah.edu/~reid/