Re: [RFA 2/4] dwarf2_physname

[Tom Tromey <tromey@redhat.com>]

>>>>> "André" == André Pönitz <andre.poenitz@nokia.com> writes:

André> Now the results of a highly unscientific approach at measuring "time"

FWIW, I do this sort of timing too, I think it is pretty reasonable
provided that there isn't much background noise on the system.

André> # gdb 6.8 from Ubuntu 9.04 (optimized build)

GDB 7.0 as a baseline would be more interesting.  It may be a little
slower than 6.8 due to the C++ name canonicalization.  If so, using 6.8
as the baseline would make Keith's patch seem slower than it is.

André> Block 'ptype':
André> expr-cumulativ loses  ~280% (!) against tromey-python

I'm surprised by the archer-tromey-python result, I have no reason to
believe it should be faster than anything else.

André> [...], whereas 
André> optional-psymtab gains >~10%

FWIW that number is really the same as the delayed-symfile number,
because optional-psymtab only takes the fastest path if you built with a
modified GCC, to get the DWARF index.

Here are some numbers showing the speedup on a smallish (about 80 KLOC)
C++ program if you build it the right way:

CVS HEAD		1.83user 0.04system 0:02.06elapsed
delayed-symfile		1.30user 0.05system 0:01.50elapsed
optional-psymtab	0.43user 0.04system 0:00.51elapsed

I haven't tried it on my big test cases yet.

Tom