Collision-resistant hashing

[Jonathan Schilling]

Another thought on this is that there are currently times when users 
*have* to use the names linkers see for symbols, such as when passing 
in -B options to the linker (such as -Bsymbolic to protect a name in
a dynamic library from being overridden).  If the linker symbol
table name is based on a hash, it may be more difficult for users
to figure out the name to pass in.  The general problem is that
there often aren't "mangle" utilities that linkers can use, and 
thus you can't pass in a real C++ name.

(From discussions with the IA-64 ABI people here, I understand 
that there will now be "protected" and "hidden" attributes for symbols
in object files that can be turned on via a source language pragma.  
This would allow real names to be used and avoid linker -B options.  
If so for all such options then the hashing scheme wouldn't be an issue.)

Jonathan Schilling		SCO, Inc.		jls at sco.com