Re: Failing ssp tests

[Julio Merino <jmmv%NetBSD.org@localhost>]

On Sun, Mar 20, 2011 at 1:32 PM, David Laight <david%l8s.co.uk@localhost> wrote:
> On Sun, Mar 20, 2011 at 01:26:00PM +0000, Julio Merino wrote:
>> On Sun, Mar 20, 2011 at 12:41 PM, Joerg Sonnenberger
>> <joerg%britannica.bec.de@localhost> wrote:
>> > On Sun, Mar 20, 2011 at 09:39:59AM +0000, Julio Merino wrote:
>> >> As an example, doing "echo adlkfjldskfjd | ./h_fgets 11" in
>> >> src/tests/lib/libc/ssp results in a crash with the binary built with
>> >> USETOOLS=no, but not with the one built with nbmake-amd64 (i.e. the
>> >> ones I have in destdir or in /usr/tests/lib/libc/ssp).
>> >
>> > Works for me (TM). Fresh release build installed into a VM, all tests
>> > crash as expected.
>>
>> I can imagine it is working well for you and pretty much everyone
>> else; otherwise we might have already seen complaints from the
>> continuous testers ;-)
>>
>> What I'm trying to figure out is what can be causing ssp to not work
>> when I'm using nbmake-amd64.  If you take a look at the build logs I
>> pasted in the original mail, the gcc commands run in the tools vs.
>> non-tools cases are exactly the same -- with the exception of the
>> paths used.  That said, the host system should be the same as that in
>> the destdir because I installed the new release sets onto / ...
>
> Why not look at the core dump and see why it is failing!

No, no.  The test helper (h_fgets) _has_ to fail when invoked as "echo
1234567890123 | h_fgets 11" (because it obviously causes a buffer
overflow).  But is is _not_ crashing, which implies that ssp is not
working (right?).

-- 
Julio Merino / @jmmv