This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: Marking the end of a function


Richard Henderson writes:
 > On Fri, Sep 27, 2002 at 09:37:03AM +0100, Andrew Haley wrote:
 > > Do you know of some way that exporting these symbols from object files
 > > that contain CNI methods might hurt?
 > 
 > They'll be dynamically bound whether or not they are legal Java 
 > symbols.  If you try to subtract the end symbol from the start
 > symbol, you'll get link errors if they're dynamicly bound.  If
 > you don't, then you'll have a dynamic relocation anyway (RELATIVE).

I think I begin to understand your objection: are you saying that if a
source file contains something like

     .word foo
     .word foo_end

and both are put in a DSO, these will not be fixed up to point
directly to the code but to PLT entries?  That would indeed break
things!

 > For very recent Solaris, and recent binutils, one can use the
 > visibility("hidden") attribute to prevent the symbol from escaping.
 > Otherwise you're out of luck.

That's useful to know.

 > Is there any reason _not_ to use _Unwind_Find_FDE for systems that
 > already use dwarf2 unwind information?  That includes most of the
 > targets that do dynamic linking.  If you fell back on other symbols
 > for the sjlj eh...

I don't want to do it again for IA-64, mostly!  In any case, is
_Unwind_Find_FDE and the structures it uses public?

Andrew.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]