------- Comment #3 from dave at hiauly1 dot hia dot nrc dot ca 2005-12-09 18:31 ------- Subject: Re: [4.1 Regression] hppa64 EH failures
> ------- Comment #2 from joseph at codesourcery dot com 2005-12-09 16:58 > ------- > Subject: Re: [4.1 Regression] hppa64 EH failures > > On Fri, 9 Dec 2005, danglin at gcc dot gnu dot org wrote: > > > Is this with the current CVS version of GAS? A critical bug affecting > > data relocations was recently fixed. > > This is with the current binutils release (2.16.1); when I started 4.1 > branch testing I made that use release binutils rather than mainline > binutils which is being used with mainline GCC testing, though the > mainline tester had only recently started using mainline binutils for > HP-UX before 4.1 branched and no problems had been observed before then > with 2.16.1. When there's a new binutils release I expect to switch the > 4.1 branch testing to that. Ok. Can't do much about the bug in 2.16.1. 2.16.1 incorrectly handles the fixup associated with the difference of two symbols. The two 32-bit portions of a 64-bit difference are swapped in 2.16.1. This affects dwarf debugging output and EH exception support. I recently made dwarf2 EH support the default on both the 32 and 64-bit hpux ports to use EH exception support. I would have liked more time between the gas fix and enabling this feature, but Richard was pushing. I believe that you can avoid these errors by using the sjlj exception support. In a sense, these errors aren't regressions as the dwarf2 eh support is new. With cvs binutils, I don't see any regressions relative to the previous sjlj support on this target. Dave -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=25317