Re: FreeBSD 10 and zfsd

On Tue, Aug 26, 2014 at 1:51 PM, Alan Somers <asomers_at_freebsd.org> wrote:
> On Tue, Aug 26, 2014 at 1:48 PM, Garrett Cooper <yaneurabeya_at_gmail.com> wrote:
>>
>>> On Aug 26, 2014, at 12:45, Alan Somers <asomers_at_freebsd.org> wrote:
>>>
>>>> On Tue, Aug 26, 2014 at 1:39 PM, Mark Felder <feld_at_freebsd.org> wrote:
>>>> August 26 2014 10:21 AM, "Alan Somers" <asomers_at_freebsd.org> wrote:
>>>>>
>>>>> Merged into the projects/zfsd/head branch by change 270604. Merging
>>>>> to head is blocked by three issues:
>>>>> a) The atf-ksh93 hack. The correct solution is to modify all the test
>>>>> programs (not the test cases) so they can run under /bin/sh. That
>>>>> will take some effort.
>>>>
>>>> Can you provide a link to the atf-ksh93 code?
>>>
>>> There's not much to it.  It simply sets an environment variable and
>>> calls atf-sh.  What's more interesting is libtest.kshlib. In order to
>>> eliminate atf-ksh93, we would need to modify libtest.kshlib to run
>>> under /bin/sh.  Alternatively, it may be easier to split
>>> libtest.kshlib into two files: a small /bin/sh-compatible that can be
>>> sourced by the ATF test programs, and a ksh93 script that only needs
>>> to be sourced by the ATF test cases.
>>
>> I'm working on integrating the dtrace testcases into kyua, then atf under Kyua as time permits, but I'm not rewriting the core test code (yet), because I want to reduce divergence with upstream.
>>
>> Why not just require ksh93 from ports?
>
> It's not that simple.  It's certainly possible to put a "require.progs
> ksh93" in the test case header.  But you also need to use ksh93
> features before invoking the test case script.  For example, see
> zpool_upgrade_004_pos_body in
> tests/sys/cddl/zfs/tests/cli_root/zpool_upgrade/zpool_upgrade_test.sh
> .  Notice how it sources some .kshlib scripts, then checks the $KEEP
> variable.  Other tests do similar things.  Perhaps they could be fixed
> by wrapping those lines in a here document that gets fed to ksh93.
> But that's very ugly.  I'd rather fix all the tests to not require
> ksh93 before invoking the separate test case script.
>
> If you change the interpreter script from atf-ksh93 to atf-sh, you'll
> see errors like this:
>
> # kyua debug zfsd_test:zfsd_import_001_pos
> set: Illegal option -A
> set: Illegal option -A
> zfsd_test:zfsd_import_001_pos  ->  broken: Premature exit; test case
> exited with code 2

    This should be plugged into kyua-testers though instead of hacking
around ATF -- even if it's mostly a thin layer on top of the ATF
tester; we should work towards a common goal with the DTrace/ZFS
testing though because they sort of follow the same design patterns.
    My mentor jmmv_at_ might have valuable input to provide here in what
should be done (I've CCed him).
Thanks!
-Garrett