The symbol auxiliary

[Andrew Cagney <ac131313 at redhat dot com>]

Thinking out loud.

The architecture vector, using gdbarch_data(), lets modules add 
additional, per-architecture information.

I'm wondering if the symbol table should have something similar.  A 
per-module symbol auxiliary table, indexed by the `struct symbol 
address, that manages that modules symbol information.  The table would 
be populated strictly on demand, and flushed when every the 
corresponding symbol table was flushed (or when a memory crunch occured).

For instance, the dwarf2cfi could use this as a clean way of retaining 
the CFI information being loaded.  It can probably even extend to the 
readers where info, such as what dwarf2loc needs, could be delayed until 
needed.

The assumption being that almost zero of the information currently being 
read into the basic symbol table is used.  Instead of reading in all 
that info, and wasting all that memory, the readers could pull in just 
just the basics ([mangled] name, address, ???), and then read the rest 
on on-demand.  While having symaux's would significantly increase the 
the overhead of a fully loaded symbol, there would be so few that 
nothing would notice.

As for implementation, this could be done pretty much independant of the 
main symbol table.  Just that `struct symbol' address is needed.

Andrew