15 Feb
2018
15 Feb
'18
8:42 a.m.
I've successfully debugged analogous modern code longline with smart
people by judicious use of valgrind. Never underestimate the power of
sending them current source, and running a tool against it with good
diagnostic. Its probable that electric fence would have got there too,
but this method worked for the person in question (event driven code,
C++, linked to the libevent library, lots of malloc/free)
It was quite entertaining for me to do this because back in the 80s
when I worked at Leeds university I had to run student debug clinic,
and the litany was "bring me a fresh listing" which got rid of 3/4 of
the problems up front. So here I am, debugging and the moral
equivalent is "send me a link to the checked in code repository in
github, and valgrind output, and I'll find the problem". I am pretty
sure some of the problems vanished once I got code up to date, before
sending it in.
On Thu, Feb 15, 2018 at 9:50 AM, Dave Horsfall <dave(a)horsfall.org> wrote:
On Thu, 15 Feb 2018, George Michaelson wrote:
I (and I am self-confessed as the worlds least
competent programmer)
frequently found the delay between where corruption in malloc is applied and
where its detected sufficiently confusing that I would have been tempted to
follow this well trodden "remove the free()" path.
MALLOC_CONF and Electric Fence are your friends; I never released any of my
dynamic code without first running it past those (and some of my linked
lists were hideous).
--
Dave Horsfall DTM (VK2KFU) "Those who don't understand security will
suffer."