Ulrich Drepper - glibc 2.2.3 (original) (raw)

This is the mail archive of the libc-alpha@sources.redhat.commailing list for the glibc project.

• To: GNU libc testers
• Subject: glibc 2.2.3
• From: Ulrich Drepper
• Date: 26 Apr 2001 22:57:44 -0700
• Cc: VGER gcc list
• Reply-To: drepper at cygnus dot com (Ulrich Drepper)

I've uploaded to

ftp://sourceware.cygnus.com/pub/glibc/releases

the files

glibc-2.2.3.tar.bz2 (also .gz) glibc-2.2.2-2.2.3.diff.bz2 (also .gz) glibc-linuxthreads-2.2.3.tar.bz2 (also .gz)

which contain the third test release for glibc 2.2.3. For those who used the test releases there is also the file

glibc-2.2.3pre3-2.2.3.diff.bz2 (also .gz)

This is not a pure bugfix release. There were a few bigger problems discovered since 2.2.2 so an upgrade is highly recommended. But 2.2.3 also contains a quite large number of new features. Coming from glibc 2.2.2 the user-visible changes include:

Version 2.2.3

• Intel's IA-64 math library is largely integrated. It provides fast and accurate implementatations for most basic and standard math functions in float, double, and long double format.

• Stephen Moshier implemented j0, j1, jn, y0, y1, yn, lgamma, erf, erfc, and asin for the 96-bit long double format and logl for the 128-bit long double format.

• The beginning of a last-bit accurate math library by IBM Haifa were added. The basic double functions exist today. Contributed by Abraham Ziv ziv@il.ibm.com, Moshe Olshansky olshansk@il.ibm.com, Ealan Henis ealan@il.ibm.com, and Anna Reitman reitman@il.ibm.com.

• An asynchronous name lookup library was added. The interface is designed after POSIX AIO. The proposal was circulated beforehand to get comments. No negative ones came in. Implemented by Ulrich Drepper.

• Port to S390/64bit contributed by Martin Schwidefsky schwidefsky@de.ibm.com.

• David Mosberger davidm@hpl.hp.com implemented the setcontext family of functions for Linux/IA-64.

• The RPC code is now thread safe. Threads can now use the same service of different services at the same time. Patch by Eric Norum eric.norum@usask.ca with some help by Ulrich Drepper.

• Martin Schwidefsky schwidefsky@de.ibm.com implemented the setcontext family of functions for Linux/S390.

• Ulrich Drepper drepper@redhat.com implemented the setcontext family of functions for Linux/x86.

• Port to Linux/CRIS contributed by Axis Communications.

The asynchronous name lookup code is not yet documented in the manual but my original whitepaper is available at

http://www.cygnus.com/~drepper/asynchnl.pdf

One other change is noting worthwhile. To implement atexit() in the present of dlopen() according to the upcoming Unix standard the definition of atexit() is changed in a way which can cause problems with most of today's gccs. The problem is some code in gcc which was necessary in libc5 but not with glibc. A patch for gcc is available at

http://gcc.gnu.org/ml/gcc-patches/2001-03/msg01187.html

For a description of the problem look at

http://sources.redhat.com/ml/libc-alpha/2001-03/msg00048.html

This release has be thoroughly tested but only with compilers which are known to be working. There are many compiler versions which are not working correctly and the configure script is not able to catch them all. If something goes wrong during your compilation first try a different compiler and less aggressive optimizations.

Some platforms (such as Arm and MIPS) are known to have some problems. In the case of Arm it is suspected that the FPU emulator is broken. The Cris and HP/PA ports are not yet complete. Talk to the people who are doing the ports and not to the glibc lists.

One final bit, the RPC code is now thread-safe but this didn't come without a price. Due to the "clever" (NOT!) design of the interface it is not possible to transparently replace the use of the global variable rpc_createerr with a hidden function call which does the right thing. The reason is that there is a data type with the same name. Very clever, Sun! There are two possibilities implemented:

• use get_rpc_createerr() to the variable reference

• define _RPC_MT_VARS before including the RPC headers. This will define a macro rpc_createerr which can cause the problem described above

As usual, report success and failures to libc-alpha@sourceware.cygnus.com.

And my standard disclaimers:

• unless you are rolling your own distribution it is probably not wise to compile and install glibc yourself. It's too risky, you can make your system unusable.

• never write to me directly about glibc issues. I won't read it. Always use the mailing list.

I want to thank again all the people who contributed. As usual of major help were

Andreas Jaeger Andreas Schwab Bruno Haible Jakub Jelinek Roland McGrath

Many others helped, please look at the ChangeLog to get an impression.

Thanks!

-- ---------------. ,-. 1325 Chesapeake Terrace Ulrich Drepper \ ,-------------------' \ Sunnyvale, CA 94089 USA Red Hat --' drepper at redhat.com ------------------------

• Follow-Ups:
  • Re: glibc 2.2.3
    * From: Andreas Jaeger