[gclist] Two conservative collectors

Matthew P. Gordon mpg@geodesic.com
Fri, 19 Nov 1999 13:58:44 -0600

Michael Spertus wrote:
> is only "good enough for a debugger". Matt, how did you determine that the
> reports you were seeing came from holes in the stack and not temporaries as
> Hans suggests?

Amusingly, it was exactly what Charles suggested: I eventually found
that the pointers were being stored in addresses which didn't have
symbolic information, and I went and tracked them down in the debugger. 
However, that's only for lack of some basic information on stack
boundaries.  With some closer tracking, it probably could have been
reported by the library as being in the stack.
	In the end, I found that knowing when a value got put on the stack to
within a few instructions turned out to be very valuable information,
for which I absolutely needed a debugger.  But, it wouldn't have been
possible without the GC stack backtracing to tell me what address to
watch.  I'd say the techiniques are very complimentary, and not mutually