Working ccache configuration for buildworld on amd64?
On Fri, 15 Aug 2008 11:22:01 -0400 "Maxim Khitrov" http://lists.freebsd.org/mailman/listinfo/freebsd-questions>> wrote: >/ I think what's happening is that there is a collision in hash values />/ generated by ccache. That's the only thing I can think of, because />/ crt1.c is compiled twice; once from /usr/src/lib/csu/amd64/crt1.c, and />/ a second time from /usr/src/lib/csu/i386-elf/crt1.c. If LIB32 is />/ disabled in src.conf, only the first compilation takes place. If the />/ generated hash values are the same, by some chance, then the actual />/ problem is that the file is not compiled a second time when, in fact, />/ it should be. This is only a guess, however. / That collision isn't going to happen for several reasons, but it's missing the point I made earlier, that the build is failing on a cache miss. If the kind of situation you're describing is happening, then it's happening earlier, and the observed error is just a side-effect. Nothing is wrong with ccache; it works just fine. What's happening is that the build system is passing compiler options in the CC variable when invoking make for lib32, and the recommended make.conf config for ccache is blowing those away. Here's how the lib32 build is invoked with ccache enabled: cd /usr/src; MAKEOBJDIRPREFIX=/usr/obj/lib32 _SHLIBDIRPREFIX=/usr/obj/usr/src/lib32 MACHINE=i386 MACHINE_ARCH=i386 INSTALL="sh /usr/src/tools/install.sh" PATH=/usr/obj/usr/src/tmp/legacy/usr/sbin:/usr/obj/usr/src/tmp/legacy/usr/bin:/usr/obj/usr/src/tmp/legacy/usr/games:/usr/obj/usr/src/tmp/usr/sbin:/usr/obj/usr/src/tmp/usr/bin:/usr/obj/usr/src/tmp/usr/games:/sbin:/bin:/usr/sbin:/usr/bin CC="/usr/local/libexec/ccache/world-cc -m32 -march=k8 -mfancy-math-387 -DCOMPAT_32BIT -iprefix /usr/obj/usr/src/lib32/usr/ -L/usr/obj/usr/src/lib32/usr/lib32 -B/usr/obj/usr/src/lib32/usr/lib32" CXX="/usr/local/libexec/ccache/world-c++ -m32 -march=k8 -mfancy-math-387 -DCOMPAT_32BIT -iprefix /usr/obj/usr/src/lib32/usr/ -L/usr/obj/usr/src/lib32/usr/lib32 -B/usr/obj/usr/src/lib32/usr/lib32" OBJC="cc -m32 -march=k8 -mfancy-math-387 -DCOMPAT_32BIT -iprefix /usr/obj/usr/src/lib32/usr/ -L/usr/obj/usr/src/lib32/usr/lib32 -B/usr/obj/usr/src/lib32/usr/lib32" LD="ld -m elf_i386_fbsd -Y P,/usr/obj/usr/src/lib32/usr/lib32" AS="as --32" LIBDIR=/usr/lib32 SHLIBDIR=/usr/lib32 make -DNO_CPU_CFLAGS -DCOMPAT_32BIT -DWITHOUT_BIND -DWITHOUT_MAN -DWITHOUT_NLS -DWITHOUT_INFO -DWITHOUT_HTML DESTDIR=/usr/obj/usr/src/lib32 -f Makefile.inc1 libraries cd /usr/src; make -f Makefile.inc1 _prereq_libs; make -f Makefile.inc1 _startup_libs; make -f Makefile.inc1 _prebuild_libs; make -f Makefile.inc1 _generic_libs; Here's how the compiler gets called from the 2nd level make with the recommended settings: /usr/local/libexec/ccache/world-cc -O2 -fno-strict-aliasing -pipe -I/usr/src/lib/libc/include -I/usr/src/lib/libc/../../include -I/usr/src/lib/libc/i386 -D__DBINTERFACE_PRIVATE -I/usr/src/lib/libc/../../contrib/gdtoa -DINET6 -I/usr/obj/lib32/usr/src/lib/libc -I/usr/src/lib/libc/resolv -DPOSIX_MISTAKE -I/usr/src/lib/libc/locale -DBROKEN_DES -DPORTMAP -DDES_BUILTIN -I/usr/src/lib/libc/rpc -DYP -DNS_CACHING -DSYMBOL_VERSIONING -Wsystem-headers -Wall -Wno-format-y2k -Wno-uninitialized -Wno-pointer-sign -c Here's what's supposed to be happening (without ccache): cc -m32 -march=k8 -mfancy-math-387 -DCOMPAT_32BIT -iprefix /usr/obj/usr/src/lib32/usr/ -L/usr/obj/usr/src/lib32/usr/lib32 -B/usr/obj/usr/src/lib32/usr/lib32 -O2 -fno-strict-aliasing -pipe -I/usr/src/lib/libc/include -I/usr/src/lib/libc/../../include -I/usr/src/lib/libc/i386 -D__DBINTERFACE_PRIVATE -I/usr/src/lib/libc/../../contrib/gdtoa -DINET6 -I/usr/obj/lib32/usr/src/lib/libc -I/usr/src/lib/libc/resolv -DPOSIX_MISTAKE -I/usr/src/lib/libc/locale -DBROKEN_DES -DPORTMAP -DDES_BUILTIN -I/usr/src/lib/libc/rpc -DYP -DNS_CACHING -DSYMBOL_VERSIONING -Wsystem-headers -Wall -Wno-format-y2k -Wno-uninitialized -Wno-pointer-sign -c /usr/src/lib/libc/i386/gen/_ctx_start.S Notice the difference? Try this instead: # Special ccache for buildworld .if exists(/usr/local/libexec/ccache/world-cc) && !defined(NOCCACHE) && \ (!empty(.CURDIR:M/usr/src*) || !empty(.CURDIR:M/usr/obj*)) CC := ${CC:C,^cc,/usr/local/libexec/ccache/world-cc,1} CXX := ${CXX:C,^c\+\+,/usr/local/libexec/ccache/world-c++,1} .endif Instead of hard-configuring CC/CXX, it selectively modifies it only on the first invocation of make. Subsequent invocations are allowed to modify CC as needed using the original (modified) values as a seed. -j ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to "[EMAIL PROTECTED]"
Re: Working ccache configuration for buildworld on amd64?
On Fri, 15 Aug 2008 11:22:01 -0400 "Maxim Khitrov" <[EMAIL PROTECTED]> wrote: > I think what's happening is that there is a collision in hash values > generated by ccache. That's the only thing I can think of, because > crt1.c is compiled twice; once from /usr/src/lib/csu/amd64/crt1.c, and > a second time from /usr/src/lib/csu/i386-elf/crt1.c. If LIB32 is > disabled in src.conf, only the first compilation takes place. If the > generated hash values are the same, by some chance, then the actual > problem is that the file is not compiled a second time when, in fact, > it should be. This is only a guess, however. That collision isn't going to happen for several reasons, but it's missing the point I made earlier, that the build is failing on a cache miss. If the kind of situation you're describing is happening, then it's happening earlier, and the observed error is just a side-effect. ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to "[EMAIL PROTECTED]"
Re: Working ccache configuration for buildworld on amd64?
On Fri, Aug 15, 2008 at 10:48 AM, RW <[EMAIL PROTECTED]> wrote: > On Thu, 14 Aug 2008 16:04:52 -0400 > "Maxim Khitrov" <[EMAIL PROTECTED]> wrote: > >> This is an old problem, but so far I haven't been able to find a >> solution. When ccache is used to build world on amd64, the process >> fails when /usr/src/lib/csu/i386-elf/crt1.c is compiled. If >> WITHOUT_LIB32 is added to src.conf, this problem does not happen. >> Likewise, building without ccache works fine. > > I take it that you've already tried removing any unnecessary settings > such as CFLAGS. > > What interesting about this is that it's failing on a compile; i.e. on a > cache miss, when ccache is doing next to nothing. That suggests that > there's either a problem in the way that the real compiler is invoked > by ccache, or that the real failure occurred during the building > of the toolchain and it's dependencies. > > I'd try setting CCACHE_RECACHE temporarily in the environment, to flush > out the old cached files, and see if it makes a difference. I only set the CPUTYPE in make.conf, CFLAGS are untouched. Clearing out the repository (ccache -C) doesn't help. I think what's happening is that there is a collision in hash values generated by ccache. That's the only thing I can think of, because crt1.c is compiled twice; once from /usr/src/lib/csu/amd64/crt1.c, and a second time from /usr/src/lib/csu/i386-elf/crt1.c. If LIB32 is disabled in src.conf, only the first compilation takes place. If the generated hash values are the same, by some chance, then the actual problem is that the file is not compiled a second time when, in fact, it should be. This is only a guess, however. - Max ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to "[EMAIL PROTECTED]"
Re: Working ccache configuration for buildworld on amd64?
On Thu, 14 Aug 2008 16:04:52 -0400 "Maxim Khitrov" <[EMAIL PROTECTED]> wrote: > This is an old problem, but so far I haven't been able to find a > solution. When ccache is used to build world on amd64, the process > fails when /usr/src/lib/csu/i386-elf/crt1.c is compiled. If > WITHOUT_LIB32 is added to src.conf, this problem does not happen. > Likewise, building without ccache works fine. I take it that you've already tried removing any unnecessary settings such as CFLAGS. What interesting about this is that it's failing on a compile; i.e. on a cache miss, when ccache is doing next to nothing. That suggests that there's either a problem in the way that the real compiler is invoked by ccache, or that the real failure occurred during the building of the toolchain and it's dependencies. I'd try setting CCACHE_RECACHE temporarily in the environment, to flush out the old cached files, and see if it makes a difference. ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to "[EMAIL PROTECTED]"
Working ccache configuration for buildworld on amd64?
This is an old problem, but so far I haven't been able to find a solution. When ccache is used to build world on amd64, the process fails when /usr/src/lib/csu/i386-elf/crt1.c is compiled. If WITHOUT_LIB32 is added to src.conf, this problem does not happen. Likewise, building without ccache works fine. Has anyone out there been able to find a ccache configuration that would work in this situation? I tried disabling ccache for some of the directories under /usr/src and /usr/obj, but it only caused problems in other stages of the build process. Here are the default ccache settings from make.conf: .if exists(/usr/local/libexec/ccache) && !defined(NOCCACHE) && \ (!empty(.CURDIR:M/usr/src*) || !empty(.CURDIR:M/usr/obj*)) CC= /usr/local/libexec/ccache/world-cc CXX= /usr/local/libexec/ccache/world-c++ .endif And here is what I tried using to isolate the problem, albeit with no success: .if exists(/usr/local/libexec/ccache) && !defined(NOCCACHE) && \ (!empty(.CURDIR:M/usr/src*) || !empty(.CURDIR:M/usr/obj*)) && \ empty(.CURDIR:M/usr/src/lib/csu*) && \ empty(.CURDIR:M/usr/obj/usr/src/lib/csu*) && \ empty(.CURDIR:M/usr/obj/lib32/usr/src/lib/csu*) CC= /usr/local/libexec/ccache/world-cc CXX= /usr/local/libexec/ccache/world-c++ .endif Please let me know if you have a working solution. - Max ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to "[EMAIL PROTECTED]"