hppa glibc 2.3.2 (Builders wanted!)
parisc,
Thanks to the help of many people (lamont, randolph, james, willy, dave, grant ... and more) I've gotten glibc 2.3.2 into a working state. This set of patches represents our current state.
There are a number of failures right now, and I'm trying to narrow them down to into categories before release, I would be much appreciated to have other people build glibc and report back their list of errors.
===============
WARNING
===============
This glibc should NOT be installed on any system you value, there are still some suspect high priority problems. I am merely looking for people to build and run the testsuite on their setup. Thanks :)
===============
WARNING
===============
The recipe:
0. Make some room!
mkdir glibc-cvs
cd glibc-cvs
1. Get fresh source and store it for a rainy day.
cvs -z 9 -d :pserver:[email protected]:/cvs/glibc login {enter "anoncvs" as the password}
cvs -z 9 -d :pserver:[email protected]:/cvs/glibc co libc
tar cvf libc.tar libc
2. Download the newest patches (updated this morning).
wget http://www.baldric.uwo.ca/~carlos/glibc-2.3.2-patches.tar.gz
3. Download some scripts.
wget http://www.baldric.uwo.ca/~carlos/glibc-build.sh
wget http://www.baldric.uwo.ca/~carlos/glibc-upnpatch.sh
4. Run the scripts in this order.
./glibc-upnpatch.sh (Look for rejects)
./glibc-build.sh hppa (Build it!)
5. The latter script will output the error from the 'make -k check' phase, please report those back to the list.
---
Current errors and their status:
make[2]: *** [/glibc-cvs/build-hppa/iconvdata/bug-iconv3.out] Error 1
Priority: Unknown.
o Unknown, problem with our libc_lock functions, this test starts up the iconv code in a non-threaded environment, then dlopen's libpthread to make it reentrant and then starts the iconv code back up again to see if the libc_lock functions work... it deadlocks.
make[1]: *** [iconvdata/tests] Error 2
make[2]: *** [/glibc-cvs/build-hppa/math/test-fenv.out] Error 1 make[2]: *** [/glibc-cvs/build-hppa/math/test-float.out] Error 1 make[2]: *** [/glibc-cvs/build-hppa/math/test-double.out] Error 1 make[2]: *** [/glibc-cvs/build-hppa/math/test-ifloat.out] Error 1
Priority: Low.
o We've had these for ever. Printf problems?
make[1]: *** [math/tests] Error 2
make[2]: *** [/glibc-cvs/build-hppa/stdlib/tst-strtod.out] Error 1 make[2]: *** [/glibc-cvs/build-hppa/stdlib/bug-strtod.out] Error 1
Priority: Medium.
o Recent regeressions in 2.3.2, reasons unknown, Randolph had given them a look but it might be mmap related, not tested on a kernel with jejb's fix. Going to try that today.
make[1]: *** [stdlib/tests] Error 2
make[2]: *** [/glibc-cvs/build-hppa/libio/tst-mmap-eofsync.out] Error 1 make[2]: *** [/glibc-cvs/build-hppa/libio/tst-mmap-fflushsync.out] Error 1
Priority: High.
o Fixed by jejb's recent patch to 2.6 and backport to 2.4 (Thanks!). Actually we know if fixes one of these, have yet to see if it fixes both.
make[1]: *** [libio/tests] Error 2
make[2]: *** [/glibc-cvs/build-hppa/dlfcn/tststatic.out] Error 139
Priority: High.
o Static object dlopening RTLD_LAZY and tryint to call the function descriptor stub put in place by RTLD_LAZY. This is a suspect PLABEL problem with the dynamic loader, needs more analysis. Perhaps randolph/dave's patches to binutils for converting relocations might be helpfull here since the dlopen might expect that r19 should be setup for the jump to the stub?
make[1]: *** [dlfcn/tests] Error 2
make[2]: *** [/glibc-cvs/build-hppa/posix/tst-regex.out] Error 139
Priority: High.
o Regresssion. I think it's somehow related to restoring r19 after a signal handler. Need to think more about the restoration of r19 across different syscalls.
make[2]: [/glibc-cvs/build-hppa/posix/annexc.out] Error 1 (ignored)
Ignored.
make[1]: *** [posix/tests] Error 2
make[2]: *** [/glibc-cvs/build-hppa/linuxthreads/tst-cancel2.out] Error 1 make[2]: *** [/glibc-cvs/build-hppa/linuxthreads/tst-popen.out] Error 1 make[2]: *** [/glibc-cvs/build-hppa/linuxthreads/tst-popen2.out] Error 1
Priority: High.
o Regresssions. Reasons unkown. Children die of SIGSEGV, related to restoring r19 and signal handlers?
make[1]: *** [linuxthreads/tests] Error 2
make[2]: *** [/glibc-cvs/build-hppa/rt/tst-timer.out] Error 139
Priority: High.
o Regression. Dies after returning from rt_sigsuspend and we restore r19, perhaps we shouldn't?
make[1]: *** [rt/tests] Error 2
make[2]: *** [/glibc-cvs/build-hppa/elf/neededtest.out] Error 14 make[2]: *** [/glibc-cvs/build-hppa/elf/neededtest2.out] Error 14 make[2]: *** [/glibc-cvs/build-hppa/elf/neededtest3.out] Error 17 make[2]: *** [/glibc-cvs/build-hppa/elf/neededtest4.out] Error 2 make[2]: *** [/glibc-cvs/build-hppa/elf/circleload1.out] Error 9
Priority: Low
o Been here since the start of time, related to the fact that we aren't properly initializing _r_debug in the dynamic loader.
make[2]: *** [/glibc-cvs/build-hppa/elf/tst-tls13.out] Error 1
Priority: Low.
o We don't even have tls enabled let alone implemented? :)
make[1]: *** [elf/tests] Error 2
make: *** [check] Error 2
Thanks to everyone who helped me out!
Cheers,
Carlos.