------- Comment #20 from fxcoudert at gcc dot gnu dot org 2007-09-11 11:06 ------- (In reply to comment #19) > More seriously, the test fails for the above values, but this does not mean > that it does not fail for other values.
Sorry, I was posting in a haste and was to affirmative. What I meant is: the AIX and powerpc-glibc routines are only failing for a finite number of values. That's dead annoying, but it will still be working most of the time. If I were using one of these platforms for production work, I'd be dead annoyed and complain heavily to the vendor (especially for AIX, which is not free), but as I'm not, there's only so much I can do. > So before concluding that only four values give a > wrong results, more tests should be performed with these known candidates > for failures. Well, tests certainly need to be conducted and the source code reviewed. But this needs to be done the packages who have incorrect behaviour: GNU libc and AIX libm. > This is one of my prejudice against, not the test suite itself, but its use: > incorrect conclusions are drawn from incomplete test cases. Well, you're welcome to contribute to the testsuite to extend its coverage. I'm glad it's there, because it keeps us from regressing too much (we *still* regress, and any regression is one too many, and the testsuite could be extended, but we're better off than if it weren't there). Also, to answer your affirmation that XFAILed tests are of no use to users, it's wrong. The testsuite is primarily designed for GCC developers to check on the status of their code, which is why the default reports only contain FAILing and XPASSing testcases. But it can also be used by users to check on what their compiler limitations are, in which case you need to check for XFAILing tests also. These are kept in the test summary, they're just not output in default reports. XFAIL is used to signal a functionality that's not working, due not to defect in GCC but in other software (such as the libc). -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33271