
On Sat, Jun 20, 2009 at 5:47 AM, Matthias Kilian
On Sat, Jun 20, 2009 at 02:29:02PM +0200, Matthias Kilian wrote:
Sat Jun 20 14:21:59 CEST 2009 Matthias Kilian
* Fix detection of libiconf. It's not enough to try to link against libiconv and look for a symbol iconv_open, because iconv may be installed in a way that internally renames iconv_foo to libiconv_foo in the library and adds corresponding #define iconv_foo libiconv_foo to iconv.h. This patch is needed on OpenBSD (but it's not enough, since cabal suffers from the same problem).
Oops! The last sentence is a little bit misleading. The real problem seems to be the way haskeline's Setup.hs tries to check for libiconv (it doesn't use /usr/local/lib and /usr/local/include for libraries and include files even if you pass them via environment in LDDFLAGS and CPPFLAGS to configure for a complete ghc buidl).
Are you building ghc-6.10 or 6.11? For 6.10.3, you can specify extra library/include folders using EXTRA_CABAL_CONFIGURE_FLAGS in your mk/build.mk: http://www.haskell.org/ghc/download_ghc_6_10_3.html#sources For 6.11, ghc doesn't use Haskeline's Setup.hs file anymore (since the base library now also uses iconv), so I wouldn't expect you to have that problem. But if I'm wrong, please file a report at hackage.haskell.org/trac/ghc and attach a build log, and I'll take a look at it. You can tell whether Haskeline's Setup.hs was used by searching the log for the phrase "checking whether to use -liconv".
Is there any reason for haskeline not using autoconf?
Personal preference; and it's not present on a default Windows installation. Best, -Judah