From: Christopher F. <cg...@re...> - 2003-03-05 04:18:31
|
On Tue, Mar 04, 2003 at 09:57:33PM +0000, Danny Smith wrote: >Chris, I apologise for letting this slip. No problem! You're doing me a huge favor! >I've been sidetracked by easier problems and by cricket and I'm stuck >on getting this out of -mno-cygwin-include path: >#include <...> search starts here: > /usr/lib/gcc-lib/i686-pc-cygwin/3.4/include > >I don't know if it is necessary to get it out of -mno-cygwin include >path (except for aesthetic reasons), since this is the GCC compiler's >includes. Unless cygwin and mingw chose different floating point >models (the gincluded float.h) there should be no difference between >cygwin and mingw in GCC_INCLUDE_DIR. If, for some reason, they do use >different float point models (eg 53 vs 64 bit mantissas for double) I >would be worried about problems with builtin math functions using >-mno-cygwin, since cygwin's cc1.exe etc are built with the >/usr/lib/gcc-lib/i686-pc-cygwin/3.4/include headers. Ok. That's one of the only non-symlinks in cygwin's version of gcc-lib/i686-pc-mingw32/3.x/ . I wasn't sure how different the two versions were, so I just used the include directory from your gcc release. I think libgcc.a has got to be different, though, doesn't it? Does gcc find the right version when you use -mno-cygwin now? >The other concern (yours more than mine) is long-term maintenance of >the wicked, tricksey code in cygwin.h and cygwin2.c. Yeah, I know. If you can get it in, I'll try to keep it fresh. Someday I will have to lobby the gcc developers for a better way to do what this does. Aren't you amazed that it actually works at all, though? :-) >But something has to be done, so I'll refresh the patch soon. Thanks. cgf |