Hi,
On +2021-09-02 18:41:12 -0400, Simon South wrote:
> Patching strace to add a "--trace-path" parameter to the two tests'
> definitions as in the patch below seems to fix this issue on both
> AArch64 and x86-64, and is less drastic than disabling the tests
> altogether.
>
> The changes limit strace's output during testing to only calls affecting
> files in the test directory itself, effectively filtering out the
> 'readlink("/proc/self/exe")' call from glibc that is throwing the tests
> currently. You can see a number of other places in gen_tests.in where
> this is done, presumably for similar reasons.
>
> Does this seem reasonable?
>
I worry about disabling tests in too general a way, since it creates
a hiding place which conceivably someone really clever may be able exploit.
So I wonder whether what you are doing is making it possible to
configure tests to have (narrow) context-sensitive expectations
(e.g., in this case making the test handle the error as an expected one,
as opposed to being made ignorant of it), or building in a static and
probably too general configuration.
A proper configurability, ISTM, would be preferable to any other form
of more general filtering.
Sorry if this is just noise from a lurker with insufficient knowledge
of the issue and discussions. If so please ignore ;/
of13> definitions as in the patch below seems to fix this issue on both
> --
> Simon South
> si...@simonsouth.net
>
>
> diff --git a/tests/gen_tests.in b/tests/gen_tests.in
> index 8b4e2e9..cc3ca63 100644
> --- a/tests/gen_tests.in
> +++ b/tests/gen_tests.in
> @@ -623,8 +623,8 @@ quotactl-xfs-v -v -e trace=quotactl
> read-write -a15 -eread=0,5 -ewrite=1,4 -e trace=read,write -P
> read-write-tmpfile -P /dev/zero -P /dev/null
> readahead -a1
> readdir -a16
> -readlink -xx
> -readlinkat -xx
> +readlink -xx --trace-path=test.readlink.link
> +readlinkat -xx --trace-path=test.readlinkat.link
> reboot -s 256
> recv-MSG_TRUNC -a26 -e trace=recv
> recvfrom -a35
>
>
>
--
Regards,
Bengt Richter
