Using the GHC API on Mac OS X with ghci or runhaskell causes a segmentation fault about doctest

[sol]

I'm not aware of that issue. Does the doctest test suite run flawlessly on your system? Say, the following reports no errors:

cabal configure && cabal build && ./tests/runtests.sh

[sol]

Btw, thx for reporting this!

[kazu-yamamoto]

No errors were reported with runtests.sh. In my environement, compiled "Test.hs" can work well.

[kazu-yamamoto]

BTW. My plan is to integrate doctest to test-framework-th. I have already implemented it and doctest really enpowers test-framework. I will release the package in this week. Thank you for your work on doctest!

[sol]

I think it is critical to make this work with runhaskell. I can not reproduce this on Linux (with GHC 7.0.4). I think it would be nice if we'd have a test case, that provokes this error. Can you try if runhaskell tests/Main.hs from the test-framework-doctest repository provokes this on your system?

Btw: Have you tried with doctest HEAD, or only with the latest release?

[kazu-yamamoto]

I tested both on Linux Mac:

---- 0.4.2 HEAD
Linux OK OK
Mac NG NG

NG means bus error or seg fault or "Too late for parseStaticFlags: call it before newSession".

I guess this bug is specific to Mac.

[kazu-yamamoto]

I forgot to write what I did:

Prelude> :m Test.DocTest
Prelude Test.DocTest> getDocTests [] ["Data/RBTree.hs"]

[kazu-yamamoto]

We can forget test-framework-doctest. Let's just think about doctest.

[sol]

The "Too late for parseStaticFlags: call it before newSession" issue is a limitation of GHC (yes, GHC uses global state, see the code, if you are inclined). You can only run getDocTests once per process. So you have to start a new ghci session after a call to getDocTests. This is why I recommend to test this with runhaskell only.

Maybe you are affected by GHC bug #5031/#5132. Can you try one of the minimal test cases given in the tickets?

[kazu-yamamoto]

Thank you for your info.

Due to this bug, I cannot install GHC 7.4.0 on my Mac at this moment.

http://hackage.haskell.org/trac/ghc/ticket/5726

[sol]

Can you verify that test-framework-doctest works with runhaskell once http://hackage.haskell.org/trac/ghc/ticket/5726 gets fixed? In that case I'd leave the ticket open until we have checked that.

[kazu-yamamoto]

I keep this issue in mind. I will try this with GHC RC2.

[sol]

@kazu-yamamoto any new findings on that?

[kazu-yamamoto]

Good timing. I have tried GHC 7.4.0 rc2 just before but many packages including doctest and xhtml limit version of "base" and "ghc". I was tiered to fix them and stopped this confirm job...

[sol]

@kazu-yamamoto Awesome! I saw you adapted test-framework-doctest to the latest version of Doctest.

Have you tested this with runhaskell/ghci and ghc-7.4.1?

[kazu-yamamoto]

Thank you for reminding. Now I can install all related packages. Unfortunately, runghc generated the following error:

lookupSymbol failed in relocateSection (relocate external)
/Users/kazu/Library/Haskell/ghc-7.4.0.20120126/lib/doctest-0.6.0/lib/HSdoctest-0.6.0.o: unknown symbol _doctestzm0zi6zi0_Pathszudoctest_version1_closure' Set.hs: Set.hs: unable to load packagedoctest-0.6.0'

If I use ghc to compile, the test program works well. Any thoughts?

[sol]

Yes, the library section in the cabal file probably does not list all modules. I'll fix that, and do a maintenance release.

[kazu-yamamoto]

I will test this with GHC 7.4.1.

http://hackage.haskell.org/trac/ghc/ticket/5748

[sol]

@kazu-yamamoto Ok, done.

Can you test with the latest version?

cabal update && cabal install doctest

[sol]

Hmm, I think it would be nice to require that version in the cabal file of test-framework-doctest. Does checking for patch versions work in cabal files? If not, I'd do an other release as 0.6.1.

[kazu-yamamoto]

Bravo! runghc 7.4.1 works well with doctest 0.6.0.1!

[kazu-yamamoto]

I tested with test-framework-doctest which includes my patch exactly.

[sol]

Ok, I think we can finally close this ticket.