WinXP, ghc-6.0.1, HOpenGL-1.05 .. problems
Having upgraded to ghc-6.0.1 since I last used HOpenGL, I once again embarked on re-installing.. - ghc-6.0.1 seems to lack GL/glut.h? copied that in.. - tons of warnings (most but not all about deprecation) - no errors stopping configure/make depend/make/make install.. there seem to be hidden errors, though, starting in make depend right after the green-card calls: cut: /cygdrive/c/ghc/ghc-6.0.1/bin/ghc: No such file or directory (`which ghc` on cygwin delivers the filename without .exe, but even if I add the suffix for cut, and try it out on the commandline, the filter doesn't find anything; I've got a feeling of deja vu on this one, btw..) - no Graphics.UI.GLUT in sight after install? I remember that ghc-5.04.2 with HOpenGL-1.04 and the usual patches worked without (visible) problems, so before I look for what is going wrong on my system, I just wanted to check whether the Subject-configuration is known to work on other people's systems? Cheers, Claus PS. Perhaps I shouldn't be looking at these old versions anyway, but this is still what HOpenGL users are most likely to have. Are there any plans for when the cvs-version will take over? And do I have a chance of building the cvs-version of HOpenGL without building any of the other stuff in fptools (ie. can I just use ghc-6.0.1 binary release to build it)?
Claus Reinke wrote:
[...] - ghc-6.0.1 seems to lack GL/glut.h? copied that in..
I've already complained about that (=> Sigbjorn).
- tons of warnings (most but not all about deprecation)
These are all harmless and are caused by two things: * The FFI syntax has changed a bit nowadays. * Haskell's FFI has no way to state the constness of arguments. This is a pity and even a 80% solution would be better than nothing.
- no errors stopping configure/make depend/make/make install..
there seem to be hidden errors, though, starting in make depend right after the green-card calls:
cut: /cygdrive/c/ghc/ghc-6.0.1/bin/ghc: No such file or directory
(`which ghc` on cygwin delivers the filename without .exe, but even if I add the suffix for cut, and try it out on the commandline, the filter doesn't find anything; I've got a feeling of deja vu on this one, btw..)
Comment from Makefile.mk: # SUP: One of my greatest and most evil hacks... :-} What happens when you comment out the evil line? Does "make depend" still work then?
- no Graphics.UI.GLUT in sight after install?
*aaargl* I forgot to include a patch for this in 1.05, see: http://haskell.org/pipermail/hopengl/2003-May/000389.html http://haskell.org/HOpenGL/releases/install.patch
[...] Are there any plans for when the cvs-version will take over?
Initially I hoped that the library infrastructure project would be usable "soon", but now I think that as an intermediate measure a home-grown solution will be needed. Alastair did similar things for his subprojects, BTW. To make things easy, some restructuring of the autoconf stuff in fptools will be necessary, but I'd prefer to wait until the next Hugs and GHC releases are out.
And do I have a chance of building the cvs-version of HOpenGL without building any of the other stuff in fptools (ie. can I just use ghc-6.0.1 binary release to build it)?
Alas, no, at least not easily. You could try to get the HEAD of fptools, do the usual autoreconf + configure, and then issue "make depend && make" in the OpenGL and GLUT subdirectories. The hard part will be getting the make variables right for the Haskell compiler to use, the installation directories, etc. I haven't tried it yet, but somehow this should work. Cheers, S.
* Haskell's FFI has no way to state the constness of arguments. This is a pity and even a 80% solution would be better than nothing.
The Right Thing is to generate C wrapper functions which use const as appropriate. GreenCard can do this and I believe HDirect can too. However, the 80% solution is a lot less work. Compiling the C code generated by your Haskell compiler with -Dconst is a good way to suppress these warnings. Of course, there is a risk that some of the warnings are about a real problem... -- Alastair Reid www.haskell-consulting.com
> >[...] - ghc-6.0.1 seems to lack GL/glut.h? copied that in.. >I've already complained about that (=> Sigbjorn). ah well, fewer and fewer files missing with each round;-) > > - tons of warnings (most but not all about deprecation) >These are all harmless and are caused by two things: > * The FFI syntax has changed a bit nowadays. > * Haskell's FFI has no way to state the constness of arguments. This > is a pity and even a 80% solution would be better than nothing. okay (there's also one about missing ghci lib when processing .conf; is that an intentional limitation?). >Comment from Makefile.mk: > # SUP: One of my greatest and most evil hacks... >:-} What happens when you comment out the evil line? Does "make depend" >still work then? well, one can do better than that (unless you really love this hack so much:): replace it with a call to "ghc --print-libdir" .. wasn't it you who asked Simon M for that feature several ghc-versions ago?-) > > - no Graphics.UI.GLUT in sight after install? >*aaargl* I forgot to include a patch for this in 1.05, see: same patch as for 1.04, then? okay, will try tomorrow. btw, is there any reason to preserve the unpatched releases? I'd much prefer to have HOpenGL-1.05-pl1 (with patch applied and missing include file included), etc. on that page, keeping the separate patches only for early adopters. With the current system, occasional HOpenGL visitors download the unpatched release and fail, or start patching but forget to copy includes, and so on. The result is that I often hear people telling me that HOpenGL doesn't work for them (which just means they had too many problems and gave up instead of asking for help on this list). > > [...] Are there any plans for when the cvs-version will take over? >Initially I hoped that the library infrastructure project would be >.. >until the next Hugs and GHC releases are out. Thanks, I just wanted to check when I should prepare for the jump. Will stick with the ready-made package for now. > > And do I have a chance of building the cvs-version of HOpenGL without > > building any of the other stuff in fptools (ie. can I just use > > ghc-6.0.1 binary release to build it)? >Alas, no, at least not easily. Sigh.. that has always been my biggest gripe with fptools: too monolithic. Thanks, Claus
participants (3)
-
Alastair Reid -
C.Reinke -
Sven Panne