The cost of stack traces

Wed May 10 19:14:00 GMT 2006

Anthony Green wrote:
> On Wed, 2006年05月10日 at 11:32 -0700, David Daney wrote:
>>>This "familiar" stuff would necessarily contain an IP->source_line 
>>mapping. 
>>> No, not quite. It would contain a "stack region" to "stack trace"
> mapping. Once we hit something that looks familiar, and memcmp the rest
> of the stack, we know what the rest of the stack trace will look like.

My point was that the "rest of the stack" can be thought of as a 
sequence of mappings from IP->source_line. If you were to keep a cache 
of these things, the overhead would not be much different from that of 
an IP->source_line map for all stack frames encountered. You would lose 
the sequence information, but that does not matter as it is available 
from the new trace.
David Daney,