From Bugzilla Helper: User-Agent: Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.7.6) Gecko/20050323 Firefox/1.0.2 Fedora/1.0.2-1.3.1 Description of problem: The glibc internal testcases nptl/tst-cancel17 and tst-cancelx17 fail when run under kernel-2.6.11-1.14_FC3, and succeed under kernel-2.6.10-1.766_FC3. The success output of tst-cancel17 is ----- going to cancel tf in-time going to cancel tf2 in-time in-time cancellation succeeded aio_cancel failed going to cancel tf early going to cancel tf2 early early cancellation succeeded ----- and the fail output is ----- going to cancel tf in-time going to cancel tf2 in-time in-time cancellation succeeded aio_cancel failed going to cancel tf early going to cancel tf2 early tf2 cleanup handler not called ----- Note that the last line is different. The process exit() status is 0 versus 1. This happens during "rpmbuild -ba --target i686 glibc.spec" after "rpm --install glibc-2.3.5-0.fc3-1.src.rpm". I built glibc twice, once using gcc (GCC) 3.4.2 20041017 (Red Hat 3.4.2-6.fc3) on 2.6.10-1.766_FC3, and once using gcc (GCC) 3.4.3 20050227 (Red Hat 3.4.3-22.fc3) on 2.6.11-1.14_FC3. nptl/tst-cancel17 fails using either libc.so on 2.6.11-1.14_FC3. nptl/tst-cancel17 works using either libc.so on 2.6.10-1.766_FC3. tst-cancelx17 exhibits similar behavior as tst-cancel17. Version-Release number of selected component (if applicable): glibc-2.3.5-0.fc3-1 How reproducible: Always Steps to Reproduce: 1. rpm --install glibc-2.3.5-0.fc3-1.src.rpm 2. cd SPECS; rpmbuild -ba --target i686 glibc.spec >rpm.out 2>&1 3. look at rpm.out transcript, and in BUILD/glibc-2.3.5/build-i686-linuxnptl/nptl/tst-cancel17.out Actual Results: nptl/tst-cancel17 and tst-cancelx17 fail under kernel-2.6.11-1.14_FC3. They both succeed under kernel-2.6.10-1.766_FC3 Expected Results: Tests should succeed on both kernels. Additional info:
nptl/tst-cancel17 also fails with glibc-2.3.5-1.src.rpm on ppc (Mac Mini) under kernel-2.6.11-1.1240_FC4.
The problem is that recent kernels no longer block when you write more than fpathconf (fds[1], _PC_PIPE_BUF) bytes into a pipe from which nobody is reading. The tests will be eventually changed to do something that will be guaranteed to block, at least for the time being. Still, not sure why you are reporting it here.
I report it here because: 0. Distributed package does not pass its own regression test in then-current environment. 1. Fedora was the last to touch glibc-2.3.5-1.src.rpm . 2. Problem has not been reported before in bugzilla. 3. Reponsibility of kernel was/is uncertain (cause of problem was not reasonably known when reported). 4. GNU bug reporting process is exceedingly weak: Must be reported against GNU-only version (.src.rpm is laughed at), must be reported against CVS version (no support for released version), reports are _usually_ ignored (no reponse for months, if ever).