The nice thing about this change is that is has to be correct[*] if it still compiles. (And it does, I checked :-)

3 minor considerations:

indra/llxml/llcontrol.cpp (Diff revision 1)

1) Any reason not to do the same for name?
2) Remove space between "const" and "&", so that it's easer to visually distinguish from binary operator &
3) LLSD::map_const_iterator seems to support operator ->, so we could write itr->second instead of (*itr).second .

1. Brad Kittenbrink April 27, 2011, 11:56 a.m. (April 27, 2011, 11:56 a.m.)

   1) Premature optimization is the root of all evil.  In all seriousness though, I found that the recursive LLSD copying was causing a significant number of memory allocations.  I didn't notice the name copying in the profiling, so I didn't look at it.  Making this consistent is probably a good idea.
   2) I think this is a bad idea, that would be far less readable imho.
   3) I'd be fine with this change at some point, but I don't think it really matters.

[*] Assuming the data referenced by control_map isn't changed by other threads meanwhile. But I guess that'd also be necessary for copying it correctly, isn't it?