|Debugging Strategy Survey email@example.com (1992-03-31)|
|Re: Debugging Strategy Survey firstname.lastname@example.org (1992-04-02)|
|Re: Debugging Strategy Survey email@example.com (Jim Hill) (1992-04-21)|
|From:||firstname.lastname@example.org (Jeffery S Brown)|
|Organization:||Los Alamos National Laboratory|
|Date:||Thu, 2 Apr 1992 19:29:01 GMT|
We employ software watchpoints as follows:
Our debugger developed at Los Alamos (ldb) has a "watch" command that
instruments the attached user process (at the line/label level) with
checks for a condition specified in the "watch" command. An instrumented
program stops when the condition is satisfied. If the scope of the
"watch" command was the entire program, then the program stops when the
memory is corrupted. The checks are done "on the fly" in the attached
user process so as not to degrade performance of the instrumented code.
Other debuggers have similar features but check the "watch" condition in
the debugger causing considerable performance problems.
Hardware support helps here too, such as an set of registers that can be
set such that an signal is generated if memory is written to that is
bracketed by these registers (available on the C90).
lbd runs on CRI hardware under unicos
Jeff Brown, LANL
[I added watchpoints to GDB, using the hardware support on the 386, they
are indeed very useful. -John]
Return to the
Search the comp.compilers archives again.