DejaGnu nightmares

["J. Kean Johnston" <jkj@sco.com>]

Ok. I formally give up on DejaGnu. Great idea but I like my hair too much to
tear it all out any more. Ignore Robert Lipe's COFF results for EGCS, and
mine. DejaGnu still lies like a cheap rug. I have tried everything I can
think of to get it to handle multilibs on OSR5 properly, to no avail. It
will pass some options down tot eh compiler, but not to libio or libstdc++.
Sometimes it gets -I and -L flags right, sometimes not. I have found a few
inconsistancies in libgloss.exp (when I am more sure about my hacks I may
post a patch for people to peruse) and I now have DejaGnu passing correct
-I and -L flags for libio and libstdc++. But it will take the devil's own
luck to get the damn thing to pass -mcoff or -melf when I want it to.

I have tried running with "--tool_opts=-mcoff", with CC="gcc -mcoff",
CXX="gcc -mcoff", event --target_board="unix{-mcoff,-melf}", and none
of them pass any flags on down to libio. So all libio and libstdc++
tests have been failing (and succeeding in some cases) with the incorrect
versions of the library.

Doing a make install before running the tests only partially improves the
problem. At least it doesnt get the wrong header files now.

So ... can some DejaGnu expert PLEASE tell me how we can make sure that the
libio, libstdc++ and other non-gcc tests get passed the compiler flags
specified, and how to guarantee that library and include paths are correctly
calculated?

Kean.