It's based on up to date glibc-2.4-r4 and 2.5 ebuilds but it also includes this (more details below):
* --enable-kernel bumped to 2.6.16 (2.4) or 2.6.17 (2.5, 2.5.90.2006XXXX)
* Added back the nomalloccheck USE.
* Build with -DNDEBUG=1 if USE -debug.
* CVS update from HEAD (2.5.90.2006XXXX only).
* Glibc madvise support.
* Binutils --hash-style= support. Requires binutils 2.17.50.0.3 or later.
* Binutils -Bdirect support for faster dynamic loading of libraries. Requires patched binutils. Probably canï½´t be used with --hash-style.
Any suggestions are welcome!
*overlay (thx nesl247 for hosting)
http://svn.evolution-mission.org/toolch ... ibs/glibc/
*distfiles (thx googlecode)
http://palgene-overlay.googlecode.com/svn/distfiles/
*changelog
http://svn.evolution-mission.org/toolch ... og.overlay
http://palgene-overlay.googlecode.com/svn/ is personal overlay.
there include useful changes for experimental, like --as-needed,gcc-4.2 suport.
want to report
test on 64bit system (FEATURES="test")
testing
FEATURES="test" will pass now.
elf/check-localplt test needs some modify for glibc-ssp-compat if only don't have USE="glibc-compat20".
so, scripts/data/localplt-{i386,x86_64}-linux-gnu.data was changed in this ebuild.
But i can't test on amd64.
*additional export func.
0010_all_glibc-ssp-compat.patch:__stack_chk_fail
6901_all_2.4-amd64-strings-20061210:__bzero (perhaps hidden on amd64?)
FEATURES="test" USE="alltest" will pass with required toolchain.
testing-note
Code: Select all
note testing glibc by testsuite
you can test glibc by glibc testsuite set enviroment FEATURES="test" when emerging glibc.
skip some tests by default. you can check all tests with USE="alltest".
!WARNING! SYSTEM BECOME HIGH LOAD ON SOME TESTS.
* default skip tests
**nptl/tst-mutex5,tst-mutex5a: "premature timeout: 1.994088 difference"
**nptl/tst-cond10,tst-cond16: "Timed out: killed the child process"
Failing sometimes randomly for me.
They are timing problem, i think it depend on kernel timer or preemptible or scheduler.
application must engage timing can use rt(realtime) code, thus it should be harmless.
see http://www.osdl.org/developer_bugzilla/show_bug.cgi?id=34
note
-tst-mutex5a is same tst-mutex5 execlude #define TYPE PTHREAD_MUTEX_ADAPTIVE_NP.
-tst-mutexpi* is using PI MUTEX.
-tst-cond17 is same tst-cond16 execlude #define UNLOCK_AFTER_BROADCAST 1.
**nptl/tst-cancel1 : "Didn't expect signal from child: got `Aborted'"
this seem to fail with not supported toolchain (gcc,binutils and other debugging with gdb).
DW_CFA_val_{offset{,_sf},expresion} is supported since gcc-4.2, binutils-2.17.
see http://sourceware.org/ml/libc-alpha/2006-09/msg00039.html
on gcc-4.2 or above, this test does not ignore now.
old threads
Tweaked glibc 2.4 + 2.5 + snapshot ebuilds
Glibc with amd64 performance patches
) last night and decided to upgrade my gcc to the most recent snapshot. On a whim I decided to see if a new glibc snapshot was available so I dug up this old thread. Lo and behold there is a new ebuild (20070119) available in pal_gene's overlay. I swiped the ebuild then modified it to download and install glibc-linuxthreads and ports 20070115 and branch update 20070119. I had to modify the glibc-2.5-patches-1.4.1 tarball and delete all the linuxthreads related patches as they didn't apply (I'm guessing they were already included in the glibc-linuxthreads snapshot?). Other than that all the patches took and it emerged fine. Everything is working smoothly.
