This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the EGCS project.
Re: Contribution to egcs
>>>>> "Jeffrey" == Jeffrey A Law <law@cygnus.com> writes:
Jeffrey> It sounds like a lot of work down the wrong direction. I
Jeffrey> strongly recommend you pursue removing redundant stuff in
Jeffrey> the linker, possibly using additional information
Jeffrey> provided by the compiler (which is how stabs does it).
Since dwarf2 is a binary format, I wonder if using linkonce sections
might not buy us a lot. Suppose there were a linkonce section for
each type, say. Then, the other sections would contain relocations
for the final location of this section. Presto, magic, the linker
already knows how to deal with this. Similarly, we could have a
dedicated dwarf2 section for each template instantiation, which would
blow away duplicate debug info for template instantiations just like
we already blow away code. (Do we already do this?)
Of course, some profiling is in order to see what dwarf information is
being replicated. (Is it types, as I assumed above? Or function
declarations for functions in header files? ...)
Just a thought off the top of my not-really-a-dwarf2-expert head.
--
Mark Mitchell mark@codesourcery.com
CodeSourcery, LLC http://www.codesourcery.com