require -XGADTs in order to pattern-match GADTs
Without -XGADTs, pattern-matching GADTs causes weird typechecking problems (due to the lack of implied -XRelaxedPolyRec), see this thread http://thread.gmane.org/gmane.comp.lang.haskell.glasgow.user/16064 , and some history in bug #2004 (closed)
I propose that pattern-matching against any constructor that uses sufficient GADT features that it's particularly confused without -XRelaxedPolyRec, should require -XGADTs
(I don't understand the issue well enough to know if what I'm saying makes sense. I guess ordinary data and mere existentials don't need it, even when defined with GADT syntax...?)
Alternately we could just require -XRelaxedPolyRec, but that seems even more confusing (and it wouldn't help make code portable to other compilers, as GADTs are definitely being used). Or if Haskell-prime ever comes around and makes -XRelaxedPolyRec default (is it planning to?), then the issue might be moot. I don't think the argument (against requiring -XGADTs) by parallel to -XOverlappingInstances is particularly strong... -XOverlappingInstances has an implicit effect on a module's class definitions... also I don't necessarily agree with that we made that decision for existing flags either :-)
--Isaac Dupree
Trac metadata
| Trac field | Value |
|---|---|
| Version | 6.10.1 |
| Type | Bug |
| TypeOfFailure | OtherFailure |
| Priority | normal |
| Resolution | Unresolved |
| Component | Compiler (Type checker) |
| Test case | |
| Differential revisions | |
| BlockedBy | |
| Related | |
| Blocking | |
| CC | id@isaac.cedarswampstudios.org |
| Operating system | |
| Architecture |