Re: Recent seems to have broken toolchain
On 31 May 2016 at 12:21, Dimitry Andric wrote: > > Maybe elftoolchain's ar does something different? Assuming the .a files > are built with the system ar, of course. We don't use the ELF Tool Chain ar yet so it won't be that. (ELF Tool Chain's ar, brandelf, elfdump are updated descendants of the ones in the base system, but there are still some changes that exist only in the FreeBSD version so I have not yet migrated them.) ___ freebsd-current@freebsd.org mailing list https://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to "freebsd-current-unsubscr...@freebsd.org"
Re: Recent seems to have broken toolchain
On Tue, May 31, 2016 at 10:25:54AM -0400, Ed Maste wrote: > On 30 May 2016 at 15:51, Steve Kargl > wrote: > > > > It happens with both /usr/bin/ld and /usr/local/bin/ld. I remove the > > binutils port and still had the issue. I tried reverting recent changes > > to elftoolchain/libelftc, the resulting tree would not build. > > The elftoolchain changes are a good candidate to check, although they > should have no impact at all on the linker. > > What error did you encounter when trying to build with them reverted? See my reply to dim@. I don't remember the error message, but it concerned a missing function with a cpp_ prefix. I just tried to reproduce the error and everything built as expected. I was hoping that you might have an 'Aha!' moment and know where to look. If I give gfortran48 the -Wl,--allow-multiple-definition the code builds without issue. So, it seems a compile-time option for ld (or loader script) has been flipped. -- Steve ___ freebsd-current@freebsd.org mailing list https://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to "freebsd-current-unsubscr...@freebsd.org"
Re: Recent seems to have broken toolchain
On Tue, May 31, 2016 at 06:21:06PM +0200, Dimitry Andric wrote: > > > > The elftoolchain changes are a good candidate to check, although they > > should have no impact at all on the linker. > > > > What error did you encounter when trying to build with them reverted? > Ed is referring to a different an error after I tried revert to a known good revision. 190 9:20cd /usr/src 191 9:20svn revert -R . 192 9:21cd contrib/elftoolchain/libelftc 193 9:21svn merge -r head:295577 . 194 9:21cd /usr/src/lib/libelftc 197 9:22svn merge -r head:295577 . 198 9:22make This actually builds now. Previously, I go an error about a missing function (which I can't remember). > > Yes, pointm.o is in both libtgt.a and liboa.a. In the past, during > > linking, the symbols are resolved from the first of -ltgt or -loa > > depending on the order on the command line. > > Maybe elftoolchain's ar does something different? Assuming the .a files > are built with the system ar, of course. > I tried reverting ar to a previously known good revision. That did not help. I was hoping either you or Ed would have an 'Aha!' moment and recognize where to look in the source. Unfortunately, the problem appears on the master node of my small cluster, which in the middle of a fairly long computation. I won't be able to bisect the tree until at least Friday. -- Steve ___ freebsd-current@freebsd.org mailing list https://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to "freebsd-current-unsubscr...@freebsd.org"
Re: Recent seems to have broken toolchain
On 31 May 2016, at 16:25, Ed Maste wrote: > > On 30 May 2016 at 15:51, Steve Kargl > wrote: >> >> It happens with both /usr/bin/ld and /usr/local/bin/ld. I remove the >> binutils port and still had the issue. I tried reverting recent changes >> to elftoolchain/libelftc, the resulting tree would not build. > > The elftoolchain changes are a good candidate to check, although they > should have no impact at all on the linker. > > What error did you encounter when trying to build with them reverted? From Steve's initial mail: > gfortran48 -O2 -pipe -march=native -mtune=native -static -funroll-loops \ > --param max-unroll-times=4 -ftree-vectorize -Wall\ > -rpath /usr/local/lib/gcc48 -I/home/kargl/modules -o acolor acolor.f90 \ > globalm.o saxm.o -L/home/kargl/lib -L. -L/usr/local/lib -L. -ltgt -loa \ > -L/home/kargl/lib -L. -L/usr/local/lib -lm90 -llapack -lblas > ./liboa.a(pointm.o): In function `__pointm_MOD_l2norm2': > pointm.f90:(.text+0x490): multiple definition of `__pointm_MOD_l2norm2' > /home/kargl/lib/libtgt.a(pointm.o):pointm.f90:(.text+0x0): first defined here > > Yes, pointm.o is in both libtgt.a and liboa.a. In the past, during > linking, the symbols are resolved from the first of -ltgt or -loa > depending on the order on the command line. Maybe elftoolchain's ar does something different? Assuming the .a files are built with the system ar, of course. -Dimitry signature.asc Description: Message signed with OpenPGP using GPGMail
Re: Recent seems to have broken toolchain
On 30 May 2016 at 15:51, Steve Kargl wrote: > > It happens with both /usr/bin/ld and /usr/local/bin/ld. I remove the > binutils port and still had the issue. I tried reverting recent changes > to elftoolchain/libelftc, the resulting tree would not build. The elftoolchain changes are a good candidate to check, although they should have no impact at all on the linker. What error did you encounter when trying to build with them reverted? ___ freebsd-current@freebsd.org mailing list https://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to "freebsd-current-unsubscr...@freebsd.org"
Re: Recent seems to have broken toolchain
On Mon, May 30, 2016 at 09:29:40PM +0200, Dimitry Andric wrote: > On 29 May 2016, at 04:27, Steve Kargl > wrote: > > > > I have a Fortran application that has built forever on FreeBSD-current; > > that is, until recently. It now dies with the following error: > > > > gfortran48 -O2 -pipe -march=native -mtune=native -static -funroll-loops \ > > --param max-unroll-times=4 -ftree-vectorize -Wall\ > > -rpath /usr/local/lib/gcc48 -I/home/kargl/modules -o acolor acolor.f90 \ > > globalm.o saxm.o -L/home/kargl/lib -L. -L/usr/local/lib -L. -ltgt -loa \ > > -L/home/kargl/lib -L. -L/usr/local/lib -lm90 -llapack -lblas > > ./liboa.a(pointm.o): In function `__pointm_MOD_l2norm2': > > pointm.f90:(.text+0x490): multiple definition of `__pointm_MOD_l2norm2' > > /home/kargl/lib/libtgt.a(pointm.o):pointm.f90:(.text+0x0): first defined > > here > > > > Yes, pointm.o is in both libtgt.a and liboa.a. In the past, during > > linking, the symbols are resolved from the first of -ltgt or -loa > > depending on the order on the command line. > > If you run the above command line with -v, what linker does it use? I > suspect this may be something caused by a newer binutils port, or even > by a newer gfortran. > > If this is using /usr/local/bin/ld, you might want to try downgrading > your binutils. > It happens with both /usr/bin/ld and /usr/local/bin/ld. I remove the binutils port and still had the issue. I tried reverting recent changes to elftoolchain/libelftc, the resulting tree would not build. For the record, % /usr/bin/ld -v GNU ld 2.17.50 % /usr/local/bin/ld -v GNU ld (GNU Binutils) 2.25.1 I forgot to note that the problem does not occur for amd64 FreeBSD at r299122. Unfortunately, I won't have time until the end of the week to try to bisect the src tree (which is going to be a monumental pain with keeping everything in sync). As a workaround, I have added -Wl,--allow-multiple-definition to the compiler options for gfortran48. -- Steve ___ freebsd-current@freebsd.org mailing list https://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to "freebsd-current-unsubscr...@freebsd.org"
Re: Recent seems to have broken toolchain
On 29 May 2016, at 04:27, Steve Kargl wrote: > > I have a Fortran application that has built forever on FreeBSD-current; > that is, until recently. It now dies with the following error: > > gfortran48 -O2 -pipe -march=native -mtune=native -static -funroll-loops \ > --param max-unroll-times=4 -ftree-vectorize -Wall\ > -rpath /usr/local/lib/gcc48 -I/home/kargl/modules -o acolor acolor.f90 \ > globalm.o saxm.o -L/home/kargl/lib -L. -L/usr/local/lib -L. -ltgt -loa \ > -L/home/kargl/lib -L. -L/usr/local/lib -lm90 -llapack -lblas > ./liboa.a(pointm.o): In function `__pointm_MOD_l2norm2': > pointm.f90:(.text+0x490): multiple definition of `__pointm_MOD_l2norm2' > /home/kargl/lib/libtgt.a(pointm.o):pointm.f90:(.text+0x0): first defined here > > Yes, pointm.o is in both libtgt.a and liboa.a. In the past, during > linking, the symbols are resolved from the first of -ltgt or -loa > depending on the order on the command line. If you run the above command line with -v, what linker does it use? I suspect this may be something caused by a newer binutils port, or even by a newer gfortran. If this is using /usr/local/bin/ld, you might want to try downgrading your binutils. -Dimitry signature.asc Description: Message signed with OpenPGP using GPGMail
Recent seems to have broken toolchain
I have a Fortran application that has built forever on FreeBSD-current; that is, until recently. It now dies with the following error: gfortran48 -O2 -pipe -march=native -mtune=native -static -funroll-loops \ --param max-unroll-times=4 -ftree-vectorize -Wall\ -rpath /usr/local/lib/gcc48 -I/home/kargl/modules -o acolor acolor.f90 \ globalm.o saxm.o -L/home/kargl/lib -L. -L/usr/local/lib -L. -ltgt -loa \ -L/home/kargl/lib -L. -L/usr/local/lib -lm90 -llapack -lblas ./liboa.a(pointm.o): In function `__pointm_MOD_l2norm2': pointm.f90:(.text+0x490): multiple definition of `__pointm_MOD_l2norm2' /home/kargl/lib/libtgt.a(pointm.o):pointm.f90:(.text+0x0): first defined here Yes, pointm.o is in both libtgt.a and liboa.a. In the past, during linking, the symbols are resolved from the first of -ltgt or -loa depending on the order on the command line. The system is amd64 FreeBSD 11.0-CURRENT r300782M. I tried scanning the svn-src-head mailing list archive for a possible candidate commit that is causing the problem. Unfortunately, there is large volume of commits silencing errors from static analysis tools.o Note, the above error does not occur on an i386 FreeBSD 11.0-CURRENT r300379 system. -- Steve ___ freebsd-current@freebsd.org mailing list https://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to "freebsd-current-unsubscr...@freebsd.org"