Re: [Grep-devel] [platform-testers] new snapshot available: grep-3.1.48-7eea

[Jim Meyering]

On Sat, Dec 15, 2018 at 10:55 PM Bruno Haible <address@hidden> wrote:
>
> On HardenedBSD 11/x86_64 (with libsigsegv installed), the stack-overflow test
> fails.
>
> Since HardenedBSD 11 is a derivate of FreeBSD 11, I tried this as well:
> On FreeBSD 11/x86_64 (with libsigsegv installed), the stack-overflow test
> succeeds.
>
> What's the difference?
>
> On FreeBSD 11 (with libsigsegv):
> 1 million opening parentheses -> "grep: in:1: Unmatched ( or \("
> 2 million opening parentheses -> "grep: stack overflow"
>
> On HardenedBSD 11 (with libsigsegv):
> grep never printed "stack overflow"
> 1 million opening parentheses -> "grep: in:1: Unmatched ( or \("
> 2 million opening parentheses -> "grep: in:1: Unmatched ( or \("
> 4 million opening parentheses -> "grep: stack overflow"
> 10 million opening parentheses -> "grep: stack overflow"
>
> So, HardenedBSD just needs a larger input to make the stack overflow.
>
> 'ulimit -a' shows the difference: The stack size limit (in kiB) is
> - on FreeBSD 11:      524288  (so, 1/2 GiB)
> - on HardenedBSD 11: 1048576  (so, 1 GiB).
>
> The attached patch fixes the problem for me. But you may consider
> to derive the maximum try from the `ulimit -s` value.

Nice. Thank you for investigating and fixing.
I'll be happy to use your patch, but am adding one more term: 1000, to
test up to 10 million parentheses. That should give us a little more
time before systems with even larger stack limits cause a false
failure here.

I don't want to attempt to guess a limit from ulimit-reported numbers,
because it would have to be a heuristic and hence would provide one
more opportunity for test malfunction.