The arm-apple-darwin10-ghc wrapper is autogenerated during GHC installation and I don't think ghcios changes anything there (e.g. /usr/local/bin/ghc looks nearly identical)

Oh, here's the likely culprit: the cabal.sandbox.config hardcodes the packages.conf.d — e.g. in one of my projects it placed:

package-db: /Users/lukexi/Projects/Thop/.cabal-sandbox/x86_64-osx-ghc-7.8.3-packages.conf.d

On Thu, Nov 13, 2014 at 11:32 AM, Axis Sivitz <axis@asivitz.com> wrote:

I filed a bug: https://github.com/haskell/cabal/issues/2213 

Someone commented that it should basically work.

I’m trying to track down exactly where it goes wrong for me. There’s a problem in the way "cabal exec” interacts with the ghc-ios script.

If I run…

ghc-ios -v3 test.hs

Glasgow Haskell Compiler, Version 7.8.3, stage 1 booted by GHC version 7.8.2

Using binary package database: /usr/local/lib/arm-apple-darwin10-ghc-7.8.3/package.conf.d/package.cache

Using binary package database: /Users/axis/.ghc/arm-ios-7.8.3/package.conf.d/package.cache

It correctly finds the package database.

But if I run…

cabal exec ghc-ios -- -v3 test.hs

Glasgow Haskell Compiler, Version 7.8.3, stage 1 booted by GHC version 7.8.2

Using binary package database: /usr/local/lib/ghc-7.8.3/package.conf.d/package.cache

Using binary package database: /Users/axis/sandbox/.cabal-sandbox/x86_64-osx-ghc-7.8.3-packages.conf.d/package.cache

It correctly uses the sandbox, but not the arm ghc.

I looked into /usr/local/bin/arm-apple-darwin10-ghc, which does the following:

 #!/bin/sh

 exedir="/usr/local/lib/arm-apple-darwin10-ghc-7.8.3/bin"

 exeprog="ghc-stage1"

 executablename="$exedir/$exeprog"

 datadir="/usr/local/share"

 bindir="/usr/local/bin"

 topdir="/usr/local/lib/arm-apple-darwin10-ghc-7.8.3"

 executablename="$exedir/ghc"

 exec "$executablename" -B"$topdir" ${1+"$@“}

I’m not sure what’s going on at the end (what is the -B flag?), but somehow that topdir variable doesn’t make it through the sandbox call.

Thanks for your help.

On Wed, Oct 15, 2014 at 9:30 PM, Axis Sivitz <axis@asivitz.com> wrote:

Oh, I think you’re right, it wasn’t actually installing correctly. I will file a bug with cabal.

Thanks

On Wed, Oct 15, 2014 at 6:58 PM, Luke Iannini <lukexipd@gmail.com> wrote:

Hi Axis,

I'm looking at my notes from my (brief) attempt to try sandboxes with ghc-ios, and it looks like one major issue is that the --builddir flag isn't paid attention to when installing into a sandbox, thus the object files from different architectures get mixed up. That will likely need a patch to cabal (I don't think many of the cabal devs are using multiple architectures/cross compilation) — filing a bug would be great.

Best

Luke

On Wed, Oct 15, 2014 at 1:57 PM, Axis Sivitz <axis@asivitz.com> wrote:

Is anyone using ghc-ios with cabal sandbox? I am having trouble with it.

I can 'cabal-ios install’ fine. That installs the packages properly in the sandbox for arm and x86.

But the ghc-ios script doesn’t use the sandbox environment. And running a simple 'cabal exec ghc-ios …’ clobbers the package-db path:

I get...

"Using binary package database: /…/sandbox/x86_64-osx-ghc-7.8.3-packages.conf.d/package.cache”

and then a bunch of errors like:

Bad interface file: /usr/local/lib/ghc-7.8.3/base-4.7.0.1/Prelude.hi

        magic number mismatch: old/corrupt interface file? (wanted 129742, got 33214052)

I guess the best solution is for cabal’s exec command to respect the arguments that cabal-ios provides. But maybe in the meantime perhaps I can tweak the ghc-ios script to add an argument to look for the presence of sandboxing, unless anyone else has a better idea.

Thanks,

Axis

_______________________________________________
iPhone mailing list
iPhone@haskell.org
http://www.haskell.org/mailman/listinfo/iphone

--

Axis Sivitz

http://asivitz.com

(513) 535-1668

--

Axis Sivitz

http://asivitz.com

(513) 535-1668