Say, have you got any idea about the reason for this build failure? http://koji.fedoraproject.org/koji/taskinfo?taskID=584409 It's dying on an AC_CHECK_LIB(xslt, xsltLibxmlVersion) configure check, which is odd because the other three platforms got past that just fine before ppc choked. I last built postgresql in rawhide on 25-March; all was well then, and the changes in the postgresql SRPM since then are utterly trivial/unrelated, so it seems this has to be due to some recent change in libxslt. If necessary I'll crank up a ppc test system and poke into the details, but I thought you might have a clue where to look.
Hum, no. Are you sure it's not just a problem on the build root. Are you explicitely BuildRequire libxslt (-devel) ? http://koji.fedoraproject.org/koji/buildinfo?buildID=46679 libxslt-1.1.23-2.fc10 seems to have built fine, ppc included So I'm inclined to blame either a missing require from the spec (by why would that work on other arches ?) or a PPC build root problem (which happen from time to time ...) Daniel
Hum, I'm afraid it's related to a problem in libxslt-1.1.23-2 key patch it kills the python bindings... Pushing a libxslt-1.1.23-3, hopefully it will solve this problem Daniel *** This bug has been marked as a duplicate of 444455 ***
Unfortunately the fix for 444455 didn't fix this one, so it's not a duplicate. I've reproduced this on a personal ppc machine, and what I'm seeing in the config.log is configure:8002: checking for xsltLibxmlVersion in -lxslt configure:8037: gcc -o conftest -O2 -g -pipe -Wall -Wp,-D_FORTIFY_SOURCE=2 -fexceptions -fstack-protector --param=ssp-buffer-size=4 -m32 -Wall -Wmissing- prototypes -Wpointer-arith -Winline -Wdeclaration-after-statement -Wendif-labels -fno-strict-aliasing -fwrapv -D_GNU_SOURCE -I/usr/include/libxml2 conftest.c - lxslt -lxml2 -lpam -lssl -lcrypto -lgssapi_krb5 -lz -lreadline -lcrypt -ldl -lm >&5 /usr/bin/ld: /tmp/ccrnnP3S.o(.text+0xc): unresolvable R_PPC_REL24 relocation against symbol `xsltLibxmlVersion' /tmp/ccrnnP3S.o: In function `main': /home/tgl/rpmwork/BUILD/postgresql-8.3.1/conftest.c:41: relocation truncated to fit: R_PPC_REL24 against symbol `xsltLibxmlVersion' defined in .sdata section in /usr/lib/gcc/ppc64-redhat-linux/4.3.0/../../../../lib/libxslt.so /usr/bin/ld: final link failed: Nonrepresentable section on output collect2: ld returned 1 exit status configure:8043: $? = 1 configure: failed program was: | #ifdef __cplusplus | extern "C" | #endif | char xsltLibxmlVersion (); | int | main () | { | return xsltLibxmlVersion (); | ; | return 0; | } configure:8061: result: no configure:8071: error: library 'xslt' is required for XSLT support
Oh ... [tgl@g3 libxslt-1.1.23]$ grep -r xsltLibxmlVersion . ./libxslt/xslt.c:const int xsltLibxmlVersion = LIBXML_VERSION; xsltLibxmlVersion is a variable, not a function. So the question is not so much why it failed as why it ever worked. Obviously this is a bug in the Postgres configure script, not in libxslt --- I suppose some recent change in gcc is what exposed it.
yes the test is wrong. In general for portable detection of libxslt use the xslt-config script, If you want to do a linking based test either declare the item properly or link against void xsltInit(void) exported from libxslt/xslt.h Daniel