This is the mail archive of the
mailing list for the elfutils project.
Re: [patch] Fix run-unstrip-n.sh regression on CentOS-5 ppc
- From: Mark Wielaard <mark at klomp dot org>
- To: elfutils-devel at lists dot fedorahosted dot org
- Date: Wed, 24 Jul 2013 23:32:52 +0200
- Subject: Re: [patch] Fix run-unstrip-n.sh regression on CentOS-5 ppc
On Wed, Jul 24, 2013 at 07:36:44PM +0200, Jan Kratochvil wrote:
> So I have changed/improved the algorithm - L_LD can be IMO compared exactly
> but otherwise the ranges should be compared for every module, not just the
> first one.
Thanks for the example. I see now how you can have multiple ranges overlap
from the link map compared to the phdrs. Please do include the explanation
in the commit message (as you done with the last commit) so we can easily
find it back in the future.
> Again I am not much happy from this code, it should be using NT_FILE instead,
> but when we keep compatibility with old OSes elfutils should not regress
Yeah, all this is somewhat fragile. Thanks for going above and beyond to
make the heuristics match up anyway, especially in the face of not having
build-ids around. NT_FILE would be helpful indeed, but it is fairly recent
and if you have NT_FILE you probably also have build-ids for everything,
so that would make things much more reliable.
> 2013-07-24 Jan Kratochvil <email@example.com>
> * dwfl_segment_report_module.c (dwfl_segment_report_module): Check for
> conflicts all the modules, not just the first one. Compare L_LD if it
> is equal, not if it is in a module address range.
This looks fine to me.