On the other hand I managed to build tools/gcc with the latest FreeBSD
Clang (including even static-linking with a bit of manual help running
nbgmake again in the build directory because for some reason it wasn't
running the right commands when run from build.sh + nbmake)
Tools continued to build until:
link mandoc/mandoc
ld: error: duplicate symbol: strcasecmp
>>> defined at strcasecmp.c:61 (/more/work/woods/m-NetBSD-current/tools/compat/../../common/lib/libc/string/strcasecmp.c:61)
>>> strcasecmp.lo:(strcasecmp) in archive /build/woods/fezzik/current-amd64-amd64-tools/lib/libnbcompat.a
>>> defined at strcasecmp.c:60 (/usr/src/lib/libc/string/strcasecmp.c:60)
>>> strcasecmp.o:(.text+0x110) in archive /usr/lib/libc.a
ld: error: duplicate symbol: strncasecmp
>>> defined at strncasecmp.c:61 (/more/work/woods/m-NetBSD-current/tools/compat/../../common/lib/libc/string/strncasecmp.c:61)
>>> strncasecmp.lo:(strncasecmp) in archive /build/woods/fezzik/current-amd64-amd64-tools/lib/libnbcompat.a
>>> defined at strcasecmp.c:85 (/usr/src/lib/libc/string/strcasecmp.c:85)
>>> strcasecmp.o:(.text+0x2A0) in archive /usr/lib/libc.a
cc: error: linker command failed with exit code 1 (use -v to see invocation)
--- mandoc ---
*** [mandoc] Error code 1
nbmake[7]: stopped in /more/work/woods/m-NetBSD-current/tools/mandoc
1 error
This is because I'm static-linking, of course, and I'm guessing
libnbcompat should not include strcasecmp() and strncasecmp() on
FreeBSD. There's a real bug underlying here though, hidden from anyone
not doing static linking.
The problem is likely that on FreeBSD there is just one .o file in
libc.a containing both strcasecmp() and strncasecmp(), as well as
strcasecmp_l() and strncasecmp_l(), so if anything else that's needed
from libc.a also needs str[n]casecmp_l() it will cause the linker to
have to bring in libc.a's strcasecmp.o thus causing the conflict.
Now as to why str[n]casecmp() end up in libnbcompat.a, well I guess
that's the puzzle. Here are the mentions from the config.log:
$ fgrep -i strcasecmp config.log
configure:5309: checking for strcasecmp
conftest.c:186:6: warning: incompatible redeclaration of library function 'strcasecmp' [-Wincompatible-library-redeclaration]
char strcasecmp ();
conftest.c:186:6: note: 'strcasecmp' is a builtin with type 'int (const char *, const char *)'
| #define HAVE_STRCASECMP 1
| #define HAVE_STRCASECMP 1
| #define HAVE_STRCASECMP 1
| #define HAVE_STRCASECMP 1
configure:5399: checking whether strcasecmp is declared
| #define HAVE_STRCASECMP 1
| #define HAVE_DECL_STRCASECMP 1
| #define HAVE_STRCASECMP 1
| #define HAVE_DECL_STRCASECMP 1
| #define HAVE_STRCASECMP 1
| #define HAVE_DECL_STRCASECMP 1
| #define HAVE_STRCASECMP 1
| #define HAVE_DECL_STRCASECMP 1
| #define HAVE_STRCASECMP 1
| #define HAVE_DECL_STRCASECMP 1
| #define HAVE_STRCASECMP 1
| #define HAVE_DECL_STRCASECMP 1
ac_cv_func_strcasecmp=yes
ac_cv_have_decl_strcasecmp=yes
#define HAVE_STRCASECMP 1
#define HAVE_DECL_STRCASECMP 1
Looking at tools/compat/Makefile I'm not seeing anything obvious that
will use the configure choices to avoid including unnecessary objects in
the final library, and I must say I'm very surprised by this omission!
--
Greg A. Woods <gwoods%acm.org@localhost>
Kelowna, BC +1 250 762-7675 RoboHack <woods%robohack.ca@localhost>
Planix, Inc. <woods%planix.com@localhost> Avoncote Farms <woods%avoncote.ca@localhost>
Attachment:
pgpZ3fzS3NcXm.pgp
Description: OpenPGP Digital Signature