Re: [RFC] Cross-gprof cleanup in gmon_io

[Ethan Solomita <ethan at cs dot columbia dot edu>]

Daniel Jacobowitz wrote:
> 
> I think the two of you are talking around each other.... Jason, is
> sparc64 a 32-bit or 64-bit userland?  i.e. what broken behavior did it
> fix?
> 
	We were talking around each other, but we actually figured it out. The
summary is that sparc64 can be either 32-bit or 64-bit userland
(although typically it's 32-bit). The new gprof opens the binary aout to
determine the pointer size, whereas the old gprof was a 32-bit app and
so assumed that sizeof(void *) was correct. The kernel's aout is clearly
using 64-bit pointers, and so now gprof expects to see them in the
output file.

	It's a change from the old expectations, but it's not unreasonable, and
does not affect user profiling. If fixes things for Jason because NetBSD
was putting out what it thought were reasonable 64-bit pointers. It
broke things for Linux because Linux's kernprof had already adapted to
casting the pointers down to 32-bits.
	-- Ethan