[DTrace-devel] [PATCH 4/5] test: fix test/demo/user/libc.d

[Kris Van Hees]

On Wed, Sep 07, 2022 at 03:58:13PM +0100, Nick Alcock wrote:
> On 7 Sep 2022, Kris Van Hees via DTrace-devel stated:
> 
> > Most distributions support /bin/date whereas some only support /bin/date.
> 
> Um, that's a bit confusing :P
> 
> But yes, for old non-unified-/usr systems, /bin/date is probably more
> likely to exist than /usr/bin/date (though every system I can see has
> a symlink in the one place point to the other).

My Debian VM does not have /usr/bin/date.

> > Mark the test as '@@skip' for now because probe* bvars are broken for pid
> > probes.
> 
> They're not meant to be, and surely @@xfail is more appropriate, or
> we'll just forget to unskip this. (This neither crashes the system nor
> takes a long time, so @@xfail...)

The trouble is that the test is NOT failing at all - but it should.  But since
the reason for failure is not what is being tested here, I'll just leave the
test without @@skip or @@xfail - and I will add a new test that should actually
test for this broken behaviour (and fail because of it).

> > -/* @@runtest-opts: -c /usr/bin/date */
> > +/* @@runtest-opts: -c /bin/date */
> > +/* @@trigger: none */
> > +/* @@skip: probefunc and other probe* bvars are broken for pid probes */
> >  
> > -pid$target:libc.so::entry
> > +pid$target:libc.so:m*:entry
> 
> This'll need later changes anyway, since "libc.so" is the wrong name and
> we can't use globs there yet...

>From what I can see, libc.so is the name we should be using here per DTrace
convention.  But we'll see what the future brings.  Main reason for this change
is that we run into trouble trying to enable pid probes for all functions in
libc.so at the same time.

That is another issue to solve in a future patch.