[originally posted to darcs-users; reposted to the bugs list.]

Hi,

I've experienced an odd build failure of darcs-2.0.0 on OS X (10.5.2)   
with
ghc 6.8.2.  The hardware is a MacBook Pro (2.6 GHz Intel, 4 GB RAM).

Running

$ ./configure
$ make

gives a successful build.  But

$ ./configure --with-docs
$ make

fails with

[ghc] src/Darcs/Patch/Read.o
[ghc] src/Darcs/Patch/Non.o
[ghc] src/Darcs/Patch/Real.o

src/Darcs/Patch/Real.lhs:100:26:
     GADT pattern match in non-rigid context for `NilRL'
       Tell GHC HQ if you'd like this to unify the context
     In the pattern: NilRL
     In the pattern: Just NilRL
     In a case alternative:
         Just NilRL
           -> ((\ imp_funny_name
                    -> imp_funny_name
                         ("src/Darcs/Patch/Real.lhs", 100, "15:47:58",
"Apr 30 2008"))
                 _bug)
                "found no patches in mergeUnravelled"

src/Darcs/Patch/Real.lhs:116:27:
     GADT pattern match in non-rigid context for `NilFL'
       Tell GHC HQ if you'd like this to unify the context
     In the pattern: NilFL
     In a case alternative: NilFL -> Just p
     In the expression:
         case sort_coalesceFL $ effect p of
           NilFL -> Just p
           _ -> Nothing

src/Darcs/Patch/Real.lhs:135:32:
     GADT pattern match in non-rigid context for `NilFL'
       Tell GHC HQ if you'd like this to unify the context
     In the pattern: NilFL
     In a case alternative: NilFL -> Just (NilFL, mapFL_FL Normal xs)
     In the expression:
         case sort_coalesceFL $ effect goneby of
           NilFL -> Just (NilFL, mapFL_FL Normal xs)
           _z -> Nothing

src/Darcs/Patch/Real.lhs:257:26:
     GADT pattern match in non-rigid context for `NilRL'
       Tell GHC HQ if you'd like this to unify the context
     In the pattern: NilRL
     In a case alternative: NilRL -> []
     In the expression:
         case reverseFL $ xx2patches ix xx of
           NilRL -> []
           Normal q :<: qs -> [Non (reverseRL qs) q]
           _ -> []

src/Darcs/Patch/Real.lhs:371:82:
     deps  not in scope because it has a wobbly type (solution: add a
type annotation)
     In the first argument of `reverseRL', namely `deps'
     In the first argument of `Non', namely `(reverseRL deps)'
     In the expression: Non (reverseRL deps) pp

src/Darcs/Patch/Real.lhs:428:50:
     GADT pattern match in non-rigid context for `NilFL'
       Tell GHC HQ if you'd like this to unify the context
     In the pattern: NilFL
     In a pattern guard for
        a case alternative:
         NilFL <- sort_coalesceFL $ effect c
     In a case alternative:
         ([], NilFL, Non c y)
           | NilFL <- sort_coalesceFL $ effect c
           -> Just (Normal y :> Conflictor a1'nop2 (y :>: NilFL) p2)
make: *** [src/Darcs/Patch/Real.o] Error 1

However, if I disable type witnesses, everything is fine.  So

$ ./configure --with-docs --without-type-witnesses
$ make

succeeds.

I can understand that type witnesses are new enough to possibly cause
problems on some platforms, but I'm baffled by the interaction with "-- 
with-docs".
Shouldn't "--with-docs" just cause the documentation to be built, and
not change any other build settings?

Best Wishes,
Greg

Weird.  Could we see the config.log outputs from both configures?

Hm.  The "jules" machine in this buildbot status page is also Mac OS 10.5.2 with
ghc 6.8.2.20080211 (according to the builder description from clicking on the
name "jules").

That one is able to configure and build fine.  So perhaps this issue is already
fixed in darcs trunk?

On Mon, May 5, 2008 at 11:52 AM, Zooko <bugs@darcs.net> wrote:

>
>
> Hm.  The "jules" machine in this buildbot status page is also Mac OS
> 10.5.2 with
> ghc 6.8.2.20080211 (according to the builder description from clicking on
> the
> name "jules").
>
> That one is able to configure and build fine.  So perhaps this issue is
> already
> fixed in darcs trunk?

The problem is not OS X specific.  The problem is with --with-docs.

Jason

Attachments

• unnamed

Okay, I added --with-docs to the configure script on the buildmaster:

http://allmydata.org/buildbot-darcs-master.cfg

Zooko: thanks again for buildbot.  

Checking http://buildbot.darcs.net , I see that you are calling configure
--with_docs (with an underscore, and not a hyphen).  Could we see what happens
with --with-docs?

On May 7, 2008, at 7:03 AM, Eric Kow wrote:

> Zooko: thanks again for buildbot.

You're welcome!  I'm very pleased with the warm fuzzy feeling that  
darcs build and darcs unit tests are getting automatically checked on  
various platforms with each checkin.

> Checking http://buildbot.darcs.net , I see that you are calling  
> configure
> --with_docs (with an underscore, and not a hyphen).  Could we see  
> what happens
> with --with-docs?

Oops.  Fixed.  Thanks for the report.  I will restart the  
buildmaster, so that it uses the new config, next time there are no  
builds in progress.  (Could be a while.)

Regards,

Zooko

gwright: the buildbot indicates that ./configure --with-docs works fine (i.e. no
compile errors whilst building)

Would you be willing to try this again with the darcs darcs repository?
Otherwise, I shall assume it's something we've already fixed and close this as
resolved-in-unstable.

Thanks!

Yes, this has been fixed already.