RFA: shrink minimal symbols

[Tom Tromey]

>>>>> "Daniel" == Daniel Jacobowitz <drow@false.org> writes:

Daniel> The OO.o/firefox numbers are with shared libraries loaded, I assume.

Yeah.

Daniel> Do you have any performance comparison?  I'm not too worried about
Daniel> lookup performance - number of lookups is relatively low though with
Daniel> full debug info I'd expect it to be linear to the number of minimal
Daniel> symbols.  But I am concerned about startup time.

Good point.  It does slow things down a bit :(

I did: 'time gdb -batch-silent' either with a file name or -p $PID:

OO.o before and after
    2.47user 0.29system 0:03.13elapsed 88%CPU
    2.51user 0.30system 0:04.98elapsed 56%CPU

firefox
    0.75user 0.16system 0:00.92elapsed 99%CPU
    0.77user 0.14system 0:00.92elapsed 99%CPU

libgcj
    1.28user 0.10system 0:01.43elapsed 96%CPU
    1.51user 0.09system 0:01.76elapsed 91%CPU

So, back to the drawing board, I think.  Perhaps using two different
types of hash table is in order -- that would keep the space savings
but ameliorate the performance problems, I think.  I'll look a bit
later.

I have another space-savings patch I'm toying with ... I'll time it
before sending it.

Tom