[PATCH, TESTSUITE]: Fix extra testsuite failures with -fpic/-fPIC.
Kaveh R. Ghazi
ghazi@caip.rutgers.edu
Tue Sep 1 20:59:00 GMT 2009
> From: "Richard Henderson" <rth@redhat.com>
>
>> For the C test cases, at least, why not fix this instead
>> by making the functions static? If the tests really rely
>> on local binding, that is...
>
> I prefer not to modify the original testcase code if possible. Sometimes
> the author wanted specifically to test whether global functions were
> optimized. Most of the time, global functions should work (and do for
> regular non-pic passes). It's only when I add manually -fpic to the
> RUNTESTFLAGS that the error occurs. So I think adding -fpie is reasonable
> for these cases.
>
> In these particular tests, we do really rely on local binding. I.e. -fpic
> will prevent inlining, auto-detecting pure/const functions and certain
> whole file optimizations because the global functions can be replaced at
> link-time. Specifying -fpie allows GCC to reassert these optimizations
> and the tests all pass (correctly IMHO) without changing the test (by
> making it static) for the non-pic default case.
Ping: http://gcc.gnu.org/ml/gcc-patches/2009-08/msg01167.html
The fortran bits went in separately, so just inline-3.c and
local-pure-const.c are awaiting.
I explained my rationale for preferring -fpie over using static when the
tests are run with -fpic. Is there still any disagreement over this patch
or may I install it?
Thanks,
--Kaveh
More information about the Gcc-patches
mailing list