Robert Elz <[email protected]> wrote:

>   Why would anyone ever care?

Okay, there's one "no" vote.  But I think it is based on a
misunderstanding of my proposal.

I agree with Robert that we don't care if a program allocates
a lot of memory that it doesn't free before exiting.  And I'm
not proposing to change that, which would, as Robert points
out, make the code more complicated for no performance gain.

What I am proposing is to re-arrange the code slightly to show
ASan that we don't care.  That would 1) make it easier to use
ASan and 2) make it easier to find leaks we do care about.

Whether the resulting code is "mangled" is a matter of opinion
and important to deciding whether to make the change; I invite
people to study the sample patch in my proposal.

And yes, there are leaks we care about.  If a program were to
leak memory in proportion to the amount of data it processes,
then the resulting memory growth would limit the size of the
folder the program could operate on.

That Robert (and I) have hit no such limit speaks to the
quality of the nmh code base: it is most likely that there is
no such progressive leak.  A fact that supports a "no" vote
against this work; as I said, it won't find bugs right away.

Like valgrind, ASan lets you ignore certain conditions.
I have been using ASan productively with the option
"detect_leaks=false", and I can continue to do so.

 < Stephen

Reply via email to