Inconsistent treatment of unsupported language features in L.H.TH.Desugar.Sweeten

[Ryan Scott]

In most parts of th-desugar, any code that deals with an unsupported language feature will simply throw an error. Here is an example from L.H.TH.Desugar.Core:

th-desugar/Language/Haskell/TH/Desugar/Core.hs

Lines 609 to 610 in 67750a0

 dsPat (ViewP _ _) = 
   fail "View patterns are not supported in th-desugar. Use pattern guards instead." 

And from L.H.TH.Desugar.Sweeten:

th-desugar/Language/Haskell/TH/Desugar/Sweeten.hs

Lines 59 to 61 in 67750a0

 #if __GLASGOW_HASKELL__ < 709 
 expToTH (DStaticE _)         = error "Static expressions supported only in GHC 7.10+" 
 #else 

However, this convention is not applied uniformly in L.H.TH.Desugar.Sweeten. There is one particular function, pragmaToTH, that returns a Maybe rather than erroring out on unsupported language features:

th-desugar/Language/Haskell/TH/Desugar/Sweeten.hs

Lines 380 to 384 in 67750a0

 #if __GLASGOW_HASKELL__ < 709 
 pragmaToTH (DLineP {}) = Nothing 
 #else 
 pragmaToTH (DLineP n str) = Just $ LineP n str 
 #endif 

This has a ripple effect on several other type signatures in L.H.TH.Desugar.Sweeten. For instance, the type of decToTH is DDec -> [Dec] instead the more sensible DDec -> Dec.

I propose that we change pragmaToTH to error out on unsupported pragmas, which would allow simplifying its type to DPragma -> Pragma, enabling other simplifications elsewhere in the module.