Improve `mkInteger` interface
mkInteger is the current operation provided by the
Integer libraries to construct (large) integer values. The current type-signature is
mkInteger :: Bool -- sign-bit -> [Int] -- absolute value in 31 bit chunks, least significant first -> Integer
A somewhat pathological example of why this representation is not so nice is the following simple CAF
c :: Integer c = 0xf000000000000000
that (this is for a 64bit arch!) gets compiled into the following STG:
==================== STG syntax: ==================== sat_sJQ :: GHC.Types.Int [LclId, Str=DmdType] = NO_CCS GHC.Types.I#! ; sat_sJR :: [GHC.Types.Int] [LclId, Str=DmdType] = NO_CCS :! [sat_sJQ GHC.Types.]; sat_sJP :: GHC.Types.Int [LclId, Str=DmdType] = NO_CCS GHC.Types.I#! ; sat_sJS :: [GHC.Types.Int] [LclId, Str=DmdType] = NO_CCS :! [sat_sJP sat_sJR]; sat_sJO :: GHC.Types.Int [LclId, Str=DmdType] = NO_CCS GHC.Types.I#! ; sat_sJT :: [GHC.Types.Int] [LclId, Str=DmdType] = NO_CCS :! [sat_sJO sat_sJS]; Foo.c :: GHC.Integer.Type.Integer [GblId, Str=DmdType] = \u srt:SRT:[0Y :-> GHC.Integer.Type.mkInteger, sJT :-> sat_sJT]  GHC.Integer.Type.mkInteger GHC.Types.True sat_sJT;
Moreover, determining how much space to allocate for the resulting
Integer is a bit work as it for one, you need to traverse the list twice, and then you can't simply derive the exact allocation amount by the list-length alone due to the odd 31-bit chunks.
Instead a more "natural"
mkInteger would be desirable, possibly in the style of
unpackCString#, in terms of a packed/unboxed vector of machine-word-sized chunks. A better
mkInteger could then take a
ByteArray# or a
Addr# + length instead, e.g.
mkInteger :: Int# -- signum(n) = sign of encoded Integer -- abs(n) = number of machine-word sized chunks -> Addr# -- pointer to start of machine-word sized chunks, -- least-significant chunk first -> Integer