On 05/09/2011 11:06 PM, Tom Tromey wrote:
> As long as the change is done cleanly, and there is some mechanism for
> avoiding the troublesome code on platforms not supporting _Complex (and
> in particular making the rest of the tests in these files continue to
> execute on such a platform), then it doesn't matter to me where the
> tests end up.
> 

As suggested by Tom, the whole test cases to _Complex type should be run
on the platforms that support _Complex.  As a result,
support_complex_tests is added in lib/gdb.exp, which determines whether
to run tests to _Complex according to proper target triplets.  As I
comment in support_complex_tests, the list of target triplet is still
very conservative, because I don't want break existing tests on some
`complex-type-unsupported' platforms that I am not aware of.  People who
familiar with a certain platform can enrich this list later.

All the _Complex stuff in *.c are isolated by checking macro
TEST_COMPLEX.  If support_complex_tests returns 1, TEST_COMPLEX is
defined in compilation commandline.  For the platforms doesn't support
_Complex, the rest of test case will run as usual.

The other part I am not sure is about PROTOTYPES.  If I understand it
correctly, these stuff are used to be compatible with old C standard,
while _Complex types are introduced in C99.  We may don't need prototype
stuffs for functions of _Complex args/return type.  In my patch, I still
follow the existing manner.

Run funargs.exp on i686-pc-linux-gnu, no new fails.

-- 
Yao (齐尧)