[DTrace-devel] [PATCH] sdt and syscall probe descriptions broken when multiple probes

[Eugene Loh]

On 10/18/2020 10:46 PM, Kris Van Hees wrote:

> On Mon, Oct 19, 2020 at 01:07:51AM -0400, eugene.loh at oracle.com wrote:
>
>> diff --git a/test/unittest/syscall/tst.entry.sh b/test/unittest/syscall/tst.entry.sh
> Can you look to see whether you can reuse one of the trigger executables we
> already have?  In that case you can probably also turn this into a .d test.
Just a quick note on this.  I think what you suggest is possible, but 
I'm not real enthusiastic.

The default trigger mechanism relies on -c, which we do not yet support.

One could use trigger-timing:after, but that would mean the D script 
would not know what pid is of interest.  Maybe that's okay, but it means 
the test would not be as discriminating.

If one uses trigger-timing:before, there remains the question which 
trigger to use.  A general problem with these triggers is that they tend 
not to have general behaviors.  They're really tailored for particular 
purposes, and it seems to me to be a little clumsy to try to hijack a 
trigger for a new purpose.

Nevertheless, a couple of triggers might be usable.  E.g., 
syscall-tst-args repeatedly calls mmap().  And pid-tst-float repeatedly 
calls usleep (well, syscall nanosleep).

I have a revised patch using such an approach, but will look at some 
other stuff before sending it out.  Meanwhile, if you have any 
additional comments about how important you think it is to change this 
from a .sh to a .d test, let me know.