I switched to Git bash and the runtime error produced by the library is the same. This error may be produced because the configuration it does not detect the netwiorkin related includes such is socket.h. This does not exist neither in the ghc installation neither in GIT/Mingw 2012/2/7 Holger Reinhardt :

I just use the version of MSys that is included with Git [1]. This puts a "Git bash" icon on your desktop which you can then use to build the network library.

[1] http://code.google.com/p/msysgit/

2012/2/7 Alberto G. Corona

...

Nothing bur a long history of failures. The problem is the configuration and versioning of MinGW and MSys. This  is a nighmare.

2012/2/7 Holger Reinhardt :

...

Oh you are using Cygwin. I'm using MSys so this is why I cannot reproduce your problem. Is there anything preventing you from using MSys?

2012/2/7 Alberto G. Corona

...

The "problem" this time is in "Configure" :

case "$host" in *-mingw32)        EXTRA_SRCS="cbits/initWinSock.c, cbits/winSockErr.c, cbits/asyncAccept.c"        EXTRA_LIBS=ws2_32        CALLCONV=stdcall ;; *-solaris2*)        EXTRA_SRCS="cbits/ancilData.c"        EXTRA_LIBS="nsl, socket"        CALLCONV=ccall ;; *)        EXTRA_SRCS="cbits/ancilData.c"        EXTRA_LIBS=        CALLCONV=ccall ;; esac

Since I´m cross-compiling with cygwin, the variable Host does not contain ¨*-muingw32"  but "i686-pc-cygwin"

changing the case , the library incorporates the lost C coded files.

Now the library links fine win imported, but there is a runtime error:

NetworkSocket.hsc:(948,3)-(1007,23): Non-exhaustive patterns in case

maybe it is due to some other preprocessor directive mismatch

2012/2/7 Holger Reinhardt :

...

Did you also change the files in the /cbits/ folder? Because they also check for HAVE_WINSOCK_H.

2012/2/7 Alberto G. Corona

...

The code is evolving and none of the versions match exactily with the patch, but substituting HAVE_WINSOCK by HAVE WINSOCK2 in these files solves the compilation problem at least in the network 2.3.0.10 version from hackage.

However it produces the same undefined references when this library is imported in my application. It seems that some object code is not included in the final library.  I verified that at least some of these undefined references correspond with  C code in the source, but somehow this is not included in the object library....

2012/2/7 Johan Tibell : > Note that there are two branches on github, master and stable. You > want > the > latter. > > On Feb 7, 2012 8:23 AM, "Alberto G. Corona" > wrote: >> >> This is quite different. >> I don´t know how but I was looking at some other older patch >> around >> the same issue and I supposed that it was the one refered by >> Yohan >> Tibell. >> >> I´ll try your patch. >> >> Thanks!. >> >> 2012/2/7 Holger Reinhardt : >> > Hi, >> > >> > (I submitted the patch that Johan linked to) >> > Network/Socket/Internal.hsc has the following code: >> > >> > #if defined(WITH_WINSOCK) || defined(cygwin32_HOST_OS) >> > type CSaFamily = (#type unsigned short) >> > #elif defined(darwin_HOST_OS) >> > type CSaFamily = (#type u_char) >> > #else >> > type CSaFamily = (#type sa_family_t) >> > #endif >> > >> > You have patched this part to always use 'unsigned short'. But >> > the >> > real >> > issue is that WITH_WINSOCK is not defined, even though it >> > should >> > be. The >> > reason for this lies in include/HsNet.h: >> > >> > #if defined(HAVE_WINSOCK_H) && !defined(cygwin32_HOST_OS) >> > # define WITH_WINSOCK  1 >> > #endif >> > >> > The problem here is that it checks for HAVE_WINSOCK_H, but the >> > configure >> > script never defines this variable. Instead it >> > defines HAVE_WINSOCK2_H. >> > It >> > seems that the network library used Winsock1 in the past and in >> > the >> > transition to Winsock2 someone forgot to change a few of the >> > #ifdefs. >> > >> > My patch just changes all occurences of HAVE_WINSOCK_H >> > to HAVE_WINSOCK2_H. >> > You might want to try that and report back if it works for you. >> > >> > 2012/2/7 Alberto G. Corona >> >> >> >> Hi Johan, >> >> The patch is not for the current version of network and the >> >> code >> >> is >> >> quite different. Basically it is necesary to  define this >> >> variable >> >> as >> >> "unsigned short" that is the thing intended in the patch. >> >> however >> >> I >> >> put it by brute force, without regard of the prerpocessor >> >> directives. >> >> With this change the code compiles well with: >> >> >> >> >> >> >> >> >> >> >> >> >> >> http://neilmitchell.blogspot.com/2010/12/installing-haskell-network-library-... >> >> >> >> However my compiled library lack the methods defined as >> >> foreign. >> >> I´ll >> >> keep trying. >> >> >> >> 2012/2/6 Johan Tibell : >> >> > Hi, >> >> > >> >> > Someone recently contributed a fix that should make network >> >> > build >> >> > with >> >> > 7.4: https://github.com/haskell/network/pull/25 >> >> > >> >> > Can you see if that works for you? I haven't yet had time to >> >> > merge >> >> > and >> >> > release that fix (I'm on vacation.) >> >> > >> >> > -- Johan >> >> > >> >> >> >> _______________________________________________ >> >> Haskell-Cafe mailing list >> >> Haskell-Cafe@haskell.org >> >> http://www.haskell.org/mailman/listinfo/haskell-cafe >> > >> >