On Tuesday, May 13, 2003, at 01:14 AM, André Pang wrote:
I suspect the compilation isn't working because the 'myDouble' symbol in c_double.o isn't being loaded _at splice time_ into the list of symbols that GHC knows about -- which is also why the exception occurs. I've tried all sorts of kludges to convince GHC to load the symbol from the foreign object file, like doing "ar rcs libcdouble.a c_double.o" and then putting -L. -lcdouble on the commandline, passing additional linker/compiler flags with -optl and -optc, etc, and none of them have worked so far.
This was tested with GHC on Mac OS X built from CVS 9 March 2003; my apologies for a fairly out-of-date GHC build, but I'm having trouble building the recent GHCs on Mac OS X. Can somebody verify that this also occurs on Linux with a recent build of GHC, and hopefully fix it or give me a clue on where to start looking for the bug?
I just built a GHC from a few hours ago (and updated the types for Templates.hs so that it outputs an ExpQ instead of Expr), and it exhibits the same "can't find symbol" bug, although it doesn't throw an exception now: 14:42 ~/th/ffi-bug % /usr/local/packages/fptools.2003.05.13/bin/ghc-5.05 -fglasgow-exts --make SplicerFFI.hs Chasing modules from: SplicerFFI.hs Compiling TemplatesFFI ( TemplatesFFI.hs, ./TemplatesFFI.o ) Compiling Main ( SplicerFFI.hs, ./SplicerFFI.o ) Loading package base ... linking ... done. Loading package haskell98 ... linking ... done. Loading package haskell-src ... linking ... done. TemplatesFFI.o: unknown symbol `_myDouble' 14:42[1] ~/th/ffi-bug % I tried replacing with the "--make" option with "-c c_double.o" and the result's the same. So I guess this hasn't been fixed yet in the HEAD branch -- maybe there's a small chance it'll work on Linux? -- % Andre Pang : just.your.average.bounty.hunter