On Tue, Oct 25, 2005 at 12:44:42PM -0700, James E Wilson wrote: > On Tue, 2005-10-25 at 01:08, Christophe LYON wrote: > > On occasions, I wonder whether it wouldn't make sense to generate > > different infos in debug_frame and eh_frame: > > That is probably a reasonable solution if it can be implemented > cleanly. There are already some differences. debug_frame info can > contain unaligned data, but for eh_frame we try to avoid this, because > some targets which aren't native DWARF2 users may lack the necessary > relocs. There are also already other differences. > > The long term solution here is to get gdb to start using debug_frame > info exclusively, instead of disassembling prologues as it does now for > most targets. Then, people will start caring more about the accuracy of > the debug_frame info than its size, and gcc will be fixed to emit full > debug_frame info, including for the epilogue.
It already does use .debug_frame exclusively, on most targets, if .debug_frame is present. i.e. if the dwarf2 information includes incorrect information about the epilogue, gdb will fall down in the epilogue. The prologue analyzers are used strictly when no debug info is available (or for a couple of thorny targets, MIPS in particular, that's a TODO). -- Daniel Jacobowitz CodeSourcery, LLC
