[DynInst_API:] commit: symtab and parseAPI optimizations now on master

[Bill Williams <bill@xxxxxxxxxxx>]

I've just merged various SymtabAPI and ParseAPI memory optimizations 
back to our master branch. This work covers the following things:
1) Line information is now parsed per-CU, provided that you ask for the 
file/line corresponding to an address. Other line information queries 
still require a full line info parse.
2) Symbols' demangled names are generated on-the-fly, rather than being 
stored persistently. This should result in lower memory overhead, some 
additional startup time (as building the demangled name indexes has 
become more expensive), and roughly comparable query speeds (we're 
replacing a log n set of string compares with a hash and a demangle).
3) ParseAPI's block and function extent data structures should consume 
much, much less space in the common case (mostly non-overlapping blocks 
and functions). It is possible that there will be a small CPU tradeoff 
for some users.
4) StackwalkerAPI no longer accidentally triggers a full binary parse by 
asking ParseAPI for the blocks containing a given address.
IMPORTANT INTERFACE CHANGES: the Symbol class has been revised in line 
with my recent RFC to provide foo_names_begin/foo_names_end accessors. 
The old container-style interfaces *must* be discarded in order for 
optimization 2) to actually work. This has propagated through various 
users of symbol names; many interfaces that were returning string 
references now must return strings by value, and the getCallee method in 
BPatch_loopTreeNode now returns a std::string rather than a const char*.
This is not an official release, but users who are interested in these 
optimizations are encouraged to try them out.
--bw

Bill Williams
Paradyn Project
bill@xxxxxxxxxxx