Cary Coutant wrote: > Rod Evans wrote: > >> STV_ELIMINATE = STV_HIDDEN/STV_INTERNAL + don't produce a symbol table >> entry. BTW, I never have know what STV_INTERNAL implied - we've just >> considered it another form of STV_HIDDEN. >> >> This might seem trivial, in that if a symbol is hidden then its symbol >> is only written to the .symtab and that can be stripped. But, as >> diagnosability (?) has become a central theme for us, we want o keep >> as much symbol table information as possible - striping isn't so >> common >> anymore. But, we do have lots of "auxiliary" symbols that are just >> needed to aid the relocation process (ie. tie one piece of code to >> another), and after which they have no use. Removing this >> "scaffolding" >> clutter is what we're trying to do. We've also added more local >> symbol information to the runtime environment, again to aid runtime >> observability - removing any clutter is useful. > > I'm not saying I'm opposed to STV_ELIMINATE (yet), but if you don't > want the symbols in the symbol table in the first place, why not just > convert any relocations to use the section symbol when generating > the .o file? No need for the linker to eliminate them if they're not > in the symbol table to begin with.
We have a need for *.o's to reference information from other .o's. -- Rod.
