Re: [DTrace-devel] [PATCH v2 2/4] test: correct file permissions

On 1/6/26 21:22, Kris Van Hees wrote:
> On Tue, Jan 06, 2026 at 08:20:10PM -0500, Eugene Loh wrote:
>> On 1/6/26 00:54, Kris Van Hees wrote:
>>
>>> On Mon, Jan 05, 2026 at 11:50:53PM -0500, Eugene Loh wrote:
>>>> The patch otherwise reads right to me.  The tests, however, are not happy.
>>>> The following tests fail intermittently:
>>>>
>>>>       test/demo/script/interp.d
>>>>       test/demo/script/tracewrite.d
>>>>       test/unittest/lockstat/tst.lockstat-summary.d
>>>>       test/unittest/scripting/tst.arg0.d
>>>>       test/unittest/scripting/tst.basic.d
>>>>       test/unittest/scripting/tst.trace.d
>>>>
>>>> These tests fail consistently:
>>>>       test/unittest/scripting/tst.assign.d
>>>>       test/unittest/scripting/tst.pgid.d
>>>>       test/unittest/scripting/tst.pid.d
>>>> The problem for these tests is
>>>>       -#!/usr/sbin/dtrace -qs
>>>>       +#!dtrace -qs
>>>> Since these files used to be invoked as scripts, the -q was ignored.  The .r
>>>> files relied on that.  So the interpreter files should drop the -q.  That
>>>> does not fix them entirely;  it simply makes their failures intermittent.
>>> Yes, I caught (I think) all of the opposite cases where I added a pragma to
>>> ensure that the -q behaviour was retained, but I don't think I considered the
>>> case where the script was not invoked using the #! and that therefore the -q
>>> was not actually effective.
>>>
>>>> I'll look more at the intermittent failures, but there may be more than one
>>>> thing going on there.  (lockstat-summary looks different)
>>> Thanks.
>> The intermittent failures were weird, but that lockstat-summary.d should
>> also fail is just plain insane.  The problem actually appears to be with a
>> different patch.  So, I'm withdrawing the "intermittent failure" complaint
>> about this patch, which should be ready for a final version.  Elsewhere,
>> I'll report on what's really going on.
> Yes, I believe it is due to:
> 	[3/4] bpf: allocate the buffers BPF map to fit highest CPU id
>
> I must have introduced a bug in that patch.

Right.  And I reviewed it.  We replace num_online_cpus with max_cpuid.  
That should be max_cpuid+1.  Is it too late to rework that patch?  If 
so, I can supply a patch to amend it.  (Anyhow, events on the highest 
cpuid are missed.)