On Thu, Dec 18, 2014 at 11:12 AM, Rainer Orth <r...@cebitec.uni-bielefeld.de> wrote: > Currently, the new gcc.dg/lto/chkp-privatize test FAILs on Solaris/x86 > with gas and ld for 64-bit: > > UNRESOLVED: gcc.dg/lto/chkp-privatize > c_lto_chkp-privatize_0.o-c_lto_chkp-privatize_1.o execute -fPIC -flto > -flto-partition=max -fcheck-pointer-bounds -mmpx > FAIL: gcc.dg/lto/chkp-privatize > c_lto_chkp-privatize_0.o-c_lto_chkp-privatize_1.o link, -fPIC -flto > -flto-partition=max -fcheck-pointer-bounds -mmpx > > output is: > ld: fatal: relocation error: file c_lto_chkp-privatize_0.o: section > [4].rela.text: invalid relocation type: 0x28 > ld: fatal: symbol referencing errors > collect2: error: ld returned 1 exit status > > This relocation type 0x28 (40) is R_X86_64_PLT32_BND, which Solaris 10 > and 11 ld know nothing about. > > Given that there's no test whatsoever in gcc/configure.ac that the whole > toolchain supports the necessary mnemonics and relocs, the logical place > seems to be the testsuite. > > The following patch does just that, turning the mpx effective-target > test into a link test, thus checking both assembler and linker used. > > Tested with the appropriate runtest invocations on i386-pc-solaris2.11 > with as/ld (both 32 and 64-bit tests unsupported due to missing > assembler support), gas/ld (32-bit tests pass, 64-bit tests unsupported > due to missing linker support), and x86_64-unknown-linux-gnu (both 32 > and 64-bit tests pass). > > I wonder what's the point of having mpx-dg.exp at all: I'd rather move > the single test to target-supports.exp, avoiding the whole flurry of > mpx-dg.exp inclusions all over gcc/testsuite. Ilya? However that may > be, it can be done as a followup. > > Ok for mainline? > > Rainer > > > 2014-12-18 Rainer Orth <r...@cebitec.uni-bielefeld.de> > > * lib/mpx-dg.exp (check_effective_target_mpx): Change into link test. > Add main.
OK. Thanks, Uros.
