I tried to build the FreeBSD port of 2.0.2 on FreeBSD 6.2 today, it eventually
failed as given below. The same symptom appeared once before on Linux, and the
following thread documents patches available to address in the Debian bug
tracker.  I'm not sure of the current state of things. 

Thanks again to the developers for your work on darcs! In continues to be a 
valuable and enjoyable tool to use.

Reference:
http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=470376

####
checking for ghc... ghc
found The Glorious Glasgow Haskell Compilation System, version 6.8.2
checking whether to profile... no
checking whether to use the threaded runtime... yes
checking GHC.Handle.openFd... NOT old API
checking GHC.Handle.openFd new API... NOT 'new' API
checking GHC.Handle.fdToHandle' API... Segmentation fault (core dumped)
failed
configure: error: Couldnt figure out how to call GHC.Handle.openFd!

Hi Mark, could you attach your config.log?
It seems to be a different issue than the one you posted in your link.

Attaching config.log. I can also try again with 6.8.3, as another way to
potentially resolve the issue. I'm not sure if compatibility with 6.8.2 remains
important. 

Thanks for the blazing fast response, Eric!

Attachments

• config.log

If I understand the log correctly, the following test program segfaults when you
run it under FreeBSD (compiled via GHC 6.8.2, e.g. ghc --make conftest), is that
correct?

--conftest.hs
import GHC.Handle ( openFd )
import IO ( IOMode(..))

main = openFd 1 Nothing False "stdout" WriteMode True

On Thu, 07 Aug 2008 13:20:54 -0000
Eric Kow <bugs@darcs.net> wrote:

> 
> Eric Kow <eric.kow@gmail.com> added the comment:
> 
> If I understand the log correctly, the following test program segfaults when you
> run it under FreeBSD (compiled via GHC 6.8.2, e.g. ghc --make conftest), is that
> correct?
> 
> --conftest.hs
> import GHC.Handle ( openFd )
> import IO ( IOMode(..))
> 
> main = openFd 1 Nothing False "stdout" WriteMode True

The result is this:

$ghc --make conftest
[1 of 1] Compiling Main             ( conftest.hs, conftest.o )

conftest.hs:1:20: Module `GHC.Handle' does not export `openFd'

####    

    Mark (who is compiling 6.8.3 now)

Whoops! I meant

import GHC.Handle ( fdToHandle' )
import IO ( IOMode(..) )

main = fdToHandle' 1 Nothing False "stdout" WriteMode True

On Thu, 07 Aug 2008 14:00:36 -0000
Eric Kow <bugs@darcs.net> wrote:

> 
> Eric Kow <eric.kow@gmail.com> added the comment:
> 
> Whoops! I meant
> 
> import GHC.Handle ( fdToHandle' )
> import IO ( IOMode(..) )
> 
> main = fdToHandle' 1 Nothing False "stdout" WriteMode True

Actually, that passed:

$ ghc --make conftest
[1 of 1] Compiling Main             ( conftest.hs, conftest.o )
Linking conftest ...

####

/me scratches head.

And what happens when you run it?

6.8.3 has finished the 'make' process now. I can proceed with 'make install',
but can keep 6.8.2 if it continues to be interesting to debug.

> And what happens when you run it?

It produces no output, but doesn't otherwise "die" in any visible way.

Weird:
- so you still segfault when configuring on GHC 6.8.2
- what about 6.8.3?

Another thought, perhaps the segfaulting can be reproduced if you use the same
compile options as the configure script:
 ghc  -threaded  -optl-L/usr/local/lib -optl-L/usr/local/lib/ -optl-pthread -o
conftest conftest.hs

On Thu, 07 Aug 2008 15:28:21 -0000
Eric Kow <bugs@darcs.net> wrote:

> 
> Eric Kow <eric.kow@gmail.com> added the comment:
> 
> Another thought, perhaps the segfaulting can be reproduced if you use the same
> compile options as the configure script:
>  ghc  -threaded  -optl-L/usr/local/lib -optl-L/usr/local/lib/ -optl-pthread -o
> conftest conftest.hs

Bingo. That compiles, but segfaults when I run it.

So, is there is more to test with 6.8.2 at this point, or should I proceed
with the upgrade to 6.8.3 (at which point, I doubt 6.8.2 will continue
to be available ).

Clearly it's a GHC/FreeBSD issue at this point and not a darcs bug.

    Mark

Hurrah!  Please try with ghc 6.8.3

If it fails too, I guess we should bump this ticket up to Ian (by virtue of him
being a GHC guy).

Or maybe the GHC folks would just rather you filed a report on
  http://hackage.haskell.org/trac/ghc/report

> Hurrah!  Please try with ghc 6.8.3
> 
> If it fails too, I guess we should bump this ticket up to Ian (by virtue of him
> being a GHC guy).
> 
> Or maybe the GHC folks would just rather you filed a report on
>   http://hackage.haskell.org/trac/ghc/report

It fails the same way with 6.8.3.

I compiled darcs 2.0.0 on the same machine with the 6.8.2 compiler, so maybe
something changed in darcs recently with triggered it, although it continues to
look like a GHC bug.

    Mark

Hmm. Ticket filed, then:
  http://hackage.haskell.org/trac/ghc/ticket/2502

Adding Ian anyway.

I think I've found the fix, but need to test it. There was a patch in the
FreeBSD port of 1.0.9 for what appears to the same issue. When the port was
updated for 2.0.0, the patches were removed the port. I believe this is because
it builds fine on FreeBSD 7.x. I suspect bringing those old patches back to life
for my 6.2 machime may fix it. 

Reference:
http://www.freshports.org/commit.php?category=devel&port=darcs&files=yes&message_id=200801041844.m04IiHSo034439@repoman.freebsd.org

I looked at this a little more today, but couldn't recreate a patch to
autoconf.mk.in that applied cleanly. I was trying to put back this patch:

-GHCFLAGS         := @GHCFLAGS@ @CPPFLAGS@ @DEFS@
+GHCFLAGS         := -package containers @GHCFLAGS@ @CPPFLAGS@ @DEFS@

I don't really know what that does. I think I may stop at this point and wait
for feedback that I'm on the right track, or ask the FreeBSD port maintainer for
help.

The use of -package containers does not seem particularly relevant here.

By the way, have you also tried building darcs from the darcs repository?

> The use of -package containers does not seem particularly relevant here.
> 
> By the way, have you also tried building darcs from the darcs repository?

I have not. I'll consider that well. Thanks for the suggestion!

    Mark

Building straight from the repo, actually worked cleanly and successfully, with
the exception that I had to tell the makefile that autoconf was named
"autoconf-2.61" on my system. 

That makes this squarely a FreeBSD packaging issue in my eyes. I have already
notified the package maintainer about this, and pointed him back to this ticket.

If I understand correctly, this is no longer an issue since we've switched to
Cabal entirely.