970917 eh problems - bootstrap failure

Robert Lipe robertl@dgii.com
Thu Sep 18 07:27:00 GMT 1997


> Since SCO5 has ELF, I'm not too worried about not supporting table-based
> unwinding for the native COFF toolchain.  We should support it for the
> native ELF assembler, though.

Sounds like a reasonable compromise to me.   SCO is pushing hard to move
to ELF anyway.  For example, they've announced the compilers they're 
releasing by the end of the year won't even support COFF generation.

To make this work well, we'll need to make DWARF2_UNWIND_INFO a runtime
variable, right?   Then we can do the moral equivalent of:
	 DWARF2_UNWIND_INFO (TARGET_ELF) ? 1 : 0 ;
Or is there a better way, Jason?

Oh.   Also to reach this compromise, the patch I submitted yesterday
to sco5.h will have to be applied I believe in order to shorten up the
section names anyway.

meissner@cygnus.com contributed:

> The original COFF did indeed restrict section names to 8 characters 

Thank you for that info.

Jason Merrill added:
> I'm just wondering if other targets will have the same problem down the road.

I don't know how long the "gcc_exception_table" has been present, but 
I don't recall hearing anyone else tripping over this.   Whether this 
means that I'm on the last OS on the planet that cares about COFF or 
whether there are others that are just unrepresented on gcc2/egcs is
something I can't guess.





More information about the Gcc mailing list