[Bug go/48020] libgo flag test FAILs on Solaris 2

[ro at CeBiTec dot Uni-Bielefeld.DE]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48020

--- Comment #2 from ro at CeBiTec dot Uni-Bielefeld.DE <ro at CeBiTec dot Uni-Bielefeld.DE> 2011-03-09 16:09:42 UTC ---
> --- Comment #1 from Ian Lance Taylor <ian at airs dot com> 2011-03-09 07:09:58 UTC ---
> I think that the problem is that the functions are not being run in the order
> they appear in the file.  Most likely the Solaris nm program is sorting the
> test functions by name.  The gotest script passes the -p option, which tells

Right, by default it does.

> GNU nm to not sort the functions.  Of course, the gcc middle-end may sort the
> functions anyhow, so this is imperfect?

Sun nm uses -p to produce the portable output format.

> Does the Solaris nm have any option to not sort the symbols?  I don't see one
> on the man page I found on the web.

You could use -v:

     -v    Sorts  external  symbols  by  value  before  they  are
           printed.

which of course matches the order of the functions in the object file.
This seems to be reasonably portable: GNU nm and the vendor nm's on IRIX
and Tru64 UNIX support it.

> I'm not sure what the best fix is going to be here.

I wonder if it might be an option (at least for a DejaGnu-ified libgo
testsuite) to have individual tests per function instead of per
directory?  This way, it might be easier to see which subtest is
failing, and (undesirable) interactions between them would be avoided.

    Rainer