stack ghci tries to load multiple versions of text

[zudov]

I am using stack to build purescript with lts-3.2. My ghc is installed system-wide, and I also have several libraries installed system-wide.

While stack build works flawlessly, stack ghci tries to use text-1.2.1.3
(from lts-3.2) and text-1.2.0.6 simultaneously which causes an error:

[ 13 of 110] Compiling Language.PureScript.Names ( Language/PureScript/Names.hs, interpreted )

Language/PureScript/Names.hs:91:52:
    Couldn't match expected type ‘T.Text’
                with actual type ‘text-1.2.0.6:Data.Text.Internal.Text’
    NB: ‘T.Text’
          is defined in ‘Data.Text.Internal’ in package ‘text-1.2.1.3’
        ‘text-1.2.0.6:Data.Text.Internal.Text’
          is defined in ‘Data.Text.Internal’ in package ‘text-1.2.0.6’
    In the first argument of ‘T.unpack’, namely ‘str’
    In the second argument of ‘splitOn’, namely ‘(T.unpack str)’

zudov@x200 ~/prog/haskell/purescript $ stack exec -- ghc-pkg list text
/usr/lib64/ghc-7.10.2/package.conf.d
   text-1.2.0.6
/home/zudov/.stack/snapshots/x86_64-linux/lts-3.2/7.10.2/pkgdb
   text-1.2.1.3
/home/zudov/prog/haskell/purescript/.stack-work/install/x86_64-linux/lts-3.2/7.10.2/pkgdb

Here is an output of stack ghci --verbose: paste.

I can reproduce it using latest stack release (0.1.3.1) and stack from master branch
(commit 478ca71).

[zudov]

Related issue #127

[chrisdone]

Interesting....

[zudov]

I've also got it appear with both transformers-0.4.2.0 and transformers-0.4.3.0 installed in a global database.

[snoyberg]

My best bet here is that the usage of -package= instead of -package-id is giving GHC too much freedom in choosing specific installed instances of packages, leading to this confusion. @chrisdone is it possible to use GhcPkgIds instead?

[snoyberg]

@zudov I've just pushed a new branch, 867-ghci-package-id, which may fix the problem, can you test it?

@chrisdone See PR #1082

[zudov]

I made sure that the problem is gone, created even more entropy in my global database than I originally had and reproduced it without a patch, then tried it with patched stack and it worked perfectly.

👍 Thanks a lot

[snoyberg]

Cool, thanks for the confirmation. I'm going to close this issue, and any problems with the patch can be discussed in the PR.