[Dwarf-Discuss] Size cost of debug info [from Dwarf-Discuss Digest, Vol 52, Issue 7]

Joel Brobecker brobecker@adacore.com
Mon Oct 24 21:52:11 GMT 2011

> I don't see a problem which cries out to be solved.  My experience
> must be different from Cary's.

Our own experience is very similar to Cary's, and we have been
contacted by a few customers because of that. Our own rule of thumb
is that debugging info is about 10 times the size of the actual
executable. A lot of it is because of mindless duplication.

Memory and disk space may be cheap, but the time it takes to read
large amounts of data, only to write them back, can be quite a bit
of a time waster. Imagine you're working on a large project, you
modify one tiny file, recompile it (instantaneous), then spend
a long time waiting for the link to finish.  Even in the best
conditions, when you have 200 engineers doing that all day, you're
losing a lot of money.

That being said, I am not sure that there is anything that needs
to be addressed at the DWARF reference level. DWARF already helps
trying to compress a bit the amount of data required to encode
the info we want to store. We can, on the other hand, have
significant gains by eliminating the duplication, and/or leaving
the debug info in the .o files (I really dislike needing the .o
files for debugging, but you have to admit that it significantly
reduces the size of the exe). Google has some other ideas which
sound very promising as well.


More information about the Dwarf-discuss mailing list