Unfortunately, that pragma pretty much makes the difference between vector being fast, getting benefits out of stream fusion for all those intermediate stream types and not.

It might be a much nicer story if stage-1 builds simply ignored ANN pragmas entirely.

As it is I get complaints from distribution maintainers with wide distributions about any use of module or function annotations for HLint, and this particular ForceSpecConstr annotation has come up as a source of pain for Austin before, though I've forgotten the details as to why.

-Edward

On Sun, Mar 30, 2014 at 6:05 PM, Nikita Karetnikov <nikita@karetnikov.org> wrote:

The latest Hackage version and HEAD fail to build on mips64el due to the
ANN pragma in ‘Data/Vector/Fusion/Stream/Monadic.hs’:

[ 5 of 21] Compiling Data.Vector.Fusion.Stream.Monadic ( Data/Vector/Fusion/Stream/Monadic.hs, dist/build/Data/Vector/Fusion/Stream/Monadic.o )
ghc: panic! (the 'impossible' happened)
(GHC version 7.6.3 for mipsel-unknown-linux):
Cant do annotations without GHCi
{Data/Vector/Fusion/Stream/Monadic.hs:104:19-33}
base:GHC.Exts.ForceSpecConstr{d r2i8}

Please report this as a GHC bug: http://www.haskell.org/ghc/reportabug

Here’s the relevant part of the code:

data SPEC = SPEC | SPEC2
#if __GLASGOW_HASKELL__ >= 700
{-# ANN type SPEC ForceSpecConstr #-}
#endif

Is it possible to use CPP along with some configure flag to determine if
GHCi is available, and if it is not, skip the said pragma? I tested the
Hackage version, and it builds fine if I comment out the mentioned
lines.

Does the lack of the pragma make a big difference? (The relevant
section of the GHC manual is very brief.)

_______________________________________________
Libraries mailing list
Libraries@haskell.org
http://www.haskell.org/mailman/listinfo/libraries