I will not expose any private structure of types via instances of the Data class. I can't see what benefits SYB brings that it outweighs such loss of abstraction. And if someone else wants to shoot themselves in the foot by using standalone deriving, that's their problem. However, this might not be an issue, as I believe all the constructors of all or almost all types in time are public. If it is possible for a user of time to hand-write instances of Data using only what time exposes, then it is safe to include those instances in the package. -- Ashley Yakeley Sterling Clover wrote:
Am I to take it that you will not be providing Data instances because you find them "unhaskellish"? The time package is a core part of Haskell functionality, and those Data instances are sorely missed by a wide range of libraries. As the maintainer of a key part of the Haskell ecosystem, I think you are obligated to take into account not only what you find to be the "good bits," but also to provide instances such as Data that, even if you find them "unhaskellish" allow the library to better integrate into the Haskell package ecosystem as it now stands. Furthermore, if such instances are not provided, they will be created with standalone deriving. The instances created by standalone deriving, unlike instances which can be partially hand-written, will provide even less in the way of preservation of the abstract nature of certain data types. In the end, you may only be cutting off your nose to spite your face.
Cheers, Sterl.
On Apr 18, 2009, at 8:26 PM, Ashley Yakeley wrote:
Jeremy Shaw wrote:
At Fri, 17 Apr 2009 11:55:51 -0700, Ashley Yakeley wrote:
Instances of Typeable have been requested for various types. I'll add these for 1.1.3. Sweet! Data as well? Happstack uses Data and Typeable for serialization. It would be nice to be able to use Data.Time instead of System.Time.
I'm not sure. I find the whole SYB approach unhaskellish, as it treats values based on their internal structure rather than their types. But perhaps serialisation is a harmless use of the Data class.
Typeable is also rather unhaskellish. A haskellish approach would be to use type witnesses, but it requires open witness declarations as a language extension. So, Typeable it is.
-- Ashley Yakeley Seattle, WA _______________________________________________ Libraries mailing list Libraries@haskell.org http://www.haskell.org/mailman/listinfo/libraries