On 4/9/19 10:57 AM, Kees Cook wrote:
On Tue, Apr 9, 2019 at 9:56 AM shuah <shuah@xxxxxxxxxx> wrote:
On 4/9/19 10:06 AM, Kees Cook wrote:
On Mon, Apr 8, 2019 at 2:32 PM shuah <shuah@xxxxxxxxxx> wrote:
Couple of things to consider:
- will this work on embedded test system with limited applications shell
support?
I was assuming the availability of "stdbuf" which is part of
coreutils, and "perl" which is already required for at least one other
test (sysctl). What do you have in mind for the "maximum supported
environment" (and where's best to document this)? Also, I see python
is used by several tests. Should python be considered instead of perl?
- It does add dependecy on perl for running tests - not a problem since
kernel build system uses it, however it might not be the case when
tests get run on system without perl.
I could likely rewrite the "prefix.pl" in C, and the test could use a
built binary instead? Or in python (as noted above). Where are tests
being run right now with such a limited environment?
Some embedded systems as I recall with just busybox env. I am thinking
shell rather C/perl/python.
It can't be done with shell alone if we want to retain the unbuffered
output. Don't we already ship binaries in this situation, though?
(i.e. such environments clearly aren't building test binaries...) Why
not also ship kselftest binaries?
They don't build them and run from install. Let's give this a try with
perl and see how this will work. Once you have the emit_tests covered,
I will upload it to a kselftest topic branch and we can ask Kernci and
Linaro folks to test this in their environments.
Thanks for working on this.
thanks,
-- Shuah
