This is the mail archive of the gdb@sourceware.cygnus.com mailing list for the GDB project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]

Re: gdb performance is impossible!!


Anglister, Shlomo wrote:
> 
> Hi,
> 
> What we have is more and more people/companies/groups moving to C++.
> Moreover, after so many years of SW development we all reached the
> conclusion that shared libraries is the solution for large applications.
> 
> It seems that debugging large programs using many/large shared libraries is
> a pain and a common problem to all architectures.
> 
> I think the real problem is that OS providers HP, SGI, SUN, IBM provide very
> poor and basic documentation on how to debug large applications.
> 

	This could very well be the case. However, IMHO, gdb itself
does not scale very well for large applications. For instance,
the minimal symbol lookups by name are accomplished by linear search.
In many configurations of gdb, psymtab lookup for a C++ method 
decays into a minimal symbol table lookup and so takes the hit due
to the linear search. Even otherwise, the fact that the lookup 
symbol could match the mangled name or the demangled name forces
gdb to use linear search (lookup_partial_symbol()). 

	Some of HP's customers are debugging large C++ programs
with dozens of shared libraries and a quarter million symbols and
the performance was not very good. Almost all the tuning we did 
was platform specific though (such as deferring demangling)

Srikanth. 


> We found out that compiler switches + secret debug environment variables
> could make the change !!!!!!
> 

	What platform is this and what are the options ?

Srikanth

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]